Glossary · Assessment Science
Construct Validity
Construct validity is the body of evidence that a test measures the specific psychological trait or ability it is designed to measure — such as logical reasoning, numerical ability, or coding proficiency — rather than a confounding factor like test-format familiarity or reading speed. It is not a single number but a cumulative case built from multiple evidence sources.
Why it matters in hiring and assessment.
A test can be well-designed, psychometrically reliable, and commercially popular while still failing to measure what it claims to measure. For example, a verbal reasoning test administered in a second language may be measuring language proficiency more than reasoning ability. A coding challenge delivered under extreme time pressure may be measuring anxiety tolerance as much as technical skill. Construct validity evidence surfaces these problems before they corrupt the selection process.
In modern psychometrics — following the unified validity framework in the Standards for Educational and Psychological Testing — construct validity is the overarching category. It encompasses four main sources of evidence:
- Content evidence: Do the items representatively sample the domain? A Python test that covers only list comprehensions and ignores error handling lacks content coverage.
- Convergent evidence: Do scores correlate with other tests of the same construct? A logical reasoning test should correlate with established reasoning measures.
- Discriminant evidence: Do scores NOT correlate with conceptually unrelated constructs? If a coding test correlates strongly with a personality extraversion scale, something is wrong.
- Criterion-related evidence: Do scores predict outcomes that the construct should predict? High logical-reasoning scores should predict training success in analytical roles.
For buyers of off-the-shelf assessments, asking vendors for their construct validity documentation — not just face validity ("it looks relevant") — is one of the most important due-diligence questions in a procurement decision.
Example.
A company wants to assess "problem-solving ability" for a data analyst role. It uses a test with abstract pattern-completion items. Construct validity evidence would include: a content review confirming the items represent the domain (convergent), a correlation with an established spatial reasoning test (convergent), no significant correlation with a verbal memory test (discriminant), and a demonstrated relationship with analyst performance ratings after six months (criterion). Each piece of evidence strengthens or weakens the case that the test measures what it claims.
Related terms.
- Criterion Validity
One key source of construct validity evidence — whether scores predict real-world performance outcomes.
- Reliability Coefficient
A test cannot demonstrate strong construct validity if it is unreliable — inconsistent scores cannot represent a stable underlying trait.
- Adverse Impact
Construct invalidity (measuring the wrong thing) is a common root cause of unexpected adverse impact — the test filters on an irrelevant trait that differs by group.
- Item Response Theory
IRT models provide item-level statistics that support construct validity investigation, including differential item functioning (DIF) analysis.