Discriminant Validity

Discriminant validity is the evidence that a construct is empirically distinct from the other constructs in your model — that what you are calling job satisfaction is not, statistically, the same thing as organisational commitment. Without it, the path coefficient between two constructs cannot be interpreted.

Three approaches are in use. The Fornell-Larcker criterion requires the square root of each construct's AVE to exceed its correlations with all other constructs. Cross-loadings should show each indicator loading most strongly on its own construct. The HTMT ratio, the most sensitive of the three, should fall below 0.85 or 0.90.

When discriminant validity fails, the honest options are to merge the two constructs, drop the overlapping indicators, or revisit the conceptual definitions that allowed them to blur. Reporting all three criteria — or at minimum HTMT with a bootstrap interval — is increasingly expected in high-quality journals.