Multi-group Discriminant Validity Assessment
Also known as: cross-group discriminant validity, multi-sample discriminant validity, MGDV, discriminant validity across groups
Multi-group discriminant validity assessment tests whether constructs measured by a scale are empirically distinct not just in one sample but consistently across two or more groups (e.g., cultures, genders, age cohorts). It extends standard discriminant validity criteria — such as the AVE rule and the HTMT ratio — into a multi-group confirmatory factor analysis framework to verify that conceptual distinctness is replicable across subpopulations.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use multi-group discriminant validity assessment when you are adapting or validating a multi-factor scale across culturally, demographically, or organisationally distinct samples and need to demonstrate that conceptually separate constructs remain empirically distinct in each group. It is especially important before conducting multi-group structural equation modeling, because discriminant validity failures can bias path estimates. Do not use this approach as a substitute for measurement invariance testing — establish at least metric invariance first. Avoid when the sample size in any group is too small (generally n < 150 per group for multi-factor models), as inter-factor correlations become unstable and HTMT ratios unreliable.
Strengths & limitations
- Provides rigorous, group-specific evidence of construct distinctness rather than assuming it transfers automatically from one-sample results.
- Combines multiple complementary criteria (Fornell-Larcker AVE, HTMT, chi-square difference test) for a comprehensive verdict.
- Naturally integrates into the multi-group CFA workflow already required for measurement invariance testing.
- Flags cross-group differences in construct boundaries that can reveal genuine cultural or demographic variation in how concepts are understood.
- Supports publishable cross-cultural or comparative validation studies with methodological rigor.
- Requires large, balanced group samples; small or highly unequal group sizes inflate Type I and Type II errors for the chi-square difference test.
- The Fornell-Larcker AVE criterion has been criticised as insufficiently sensitive to near-discriminant-validity failures; HTMT is now generally preferred.
- Multi-group CFA requires well-specified, locally identified models — any misspecification in the base model propagates into the discriminant validity assessment.
Frequently asked
Do I need to establish measurement invariance before testing multi-group discriminant validity?
Yes. At minimum, configural invariance (same factor pattern across groups) is required so that inter-factor correlations are estimated within a comparable factor structure. Metric invariance (equal loadings) is also recommended because it ensures that the scale of the latent factors is comparable across groups, making inter-factor correlations directly interpretable.
Should I use the Fornell-Larcker criterion or the HTMT ratio?
Both, when possible. The Fornell-Larcker AVE criterion is widely reported but has lower sensitivity — it can pass even when constructs are not truly distinct. The HTMT ratio is more sensitive and is now broadly preferred. A threshold below 0.85 is conservative; some authors allow up to 0.90 for conceptually related constructs. Report both criteria per group for transparency.
What if discriminant validity holds in one group but fails in another?
This is a substantively meaningful finding: the two constructs may be conceptually blurred in that group for cultural, linguistic, or contextual reasons. Investigate whether item wording or construct definitions need revision for that group, or report the limitation explicitly and interpret cross-group comparisons cautiously.
How large a sample do I need per group?
A rough minimum is about 150–200 cases per group for models with three or more factors and five or more items per factor. Smaller groups yield unstable inter-factor correlations, making AVE and HTMT computations unreliable and chi-square difference tests underpowered.
Is multi-group discriminant validity different from simply running discriminant validity in each group separately?
Mostly it uses the same criteria per group, but the multi-group CFA framework allows formal statistical tests of whether inter-factor correlations differ across groups, which separate group analyses cannot provide. The chi-square difference test and fit-index changes (ΔCFI, ΔRMSEA) are only available in the simultaneous multi-group framework.
Sources
- Fornell, C. & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. DOI: 10.1177/002224378101800104 ↗
- Vandenberg, R. J. & Lance, C. E. (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70. DOI: 10.1177/109442810031002 ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-group Discriminant Validity Assessment. ScholarGate. https://scholargate.app/en/psychometrics/multi-group-discriminant-validity
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Construct ValidityPsychometrics↔ compare
- Convergent ValidityPsychometrics↔ compare
- Discriminant ValidityPsychometrics↔ compare
- Multi-group confirmatory factor analysisPsychometrics↔ compare
- Multi-group measurement invariancePsychometrics↔ compare