The umbrella question of measurement—convergent, discriminant and factorial evidence, and what a validation story looks like.
Construct validity is the master question: is your score actually about the theoretical thing you claim? Evidence stacks up from many directions—content review, correlations with similar instruments (convergent), low correlations with unlike ones (discriminant), and factor structure.
No single formula—evidence triangulation: MTMM matrix, factor loadings, hypothesis tests.
A new engagement scale correlates .74 with an established one (convergent) and only .18 with tenure (discriminant)—the construct story holds.
Every name you attach to a score—"engagement", "readiness", "potential"—is a construct claim. Build the evidence before marketing the noun.
A factor analysis alone doesn't prove validity; the theory has to name the factors and the patterns must match predictions.
No—it's a body of evidence. Any vendor showing a single "validity coefficient" is selling, not measuring.
Reliability = consistent; validity = on-target. A broken clock is perfectly reliable and invalid.
Turn the concept into a live assessment：Construct Validity
Browse templates →