Psychometric Validation Research Note

Psychometric Validation Research Note

Assessment evidence is tied to an intended interpretation, population, and use. Validity is not a general vendor badge.

The EEOC's employment-test guidance says employers remain responsible for selection procedures. The Uniform Guidelines clarification discusses criterion, content, and construct validity.

SIOP's recommendations for AI-based selection assessments emphasize intended interpretation, relevance, reliability, validity, population, and transport to a new setting.

Reliability questions include inter-rater agreement, test-retest behavior, prompt equivalence, scoring repeatability, minimum input, model-version stability, human-review effects, and uncertainty.

Validity questions include construct definition, relationships with adjacent measures, outcome relevance, incremental value over existing evidence, prediction interval, error costs, and the evidence supporting the actual use.

Testimonials, face validity, buyer agreement, correlation with another assessment, or a compelling sample report do not establish selection validity.