Discriminant Validity
Also known as: discriminant validity evidence, divergent validity, DV, AVE-based discriminant validity
Discriminant validity is evidence that a latent construct is empirically distinct from other constructs it should differ from. Originating in Campbell and Fiske's multitrait-multimethod framework (1959), it is a core component of construct validity and a mandatory check in scale development and structural equation modeling.
Key highlights
- Provides direct empirical evidence that conceptually distinct constructs are also statistically distinct, strengthening theoretical precision.
- Multiple established criteria (Fornell-Larcker, HTMT, chi-square difference) allow triangulation across methods.
- HTMT is sensitive to moderate levels of overlap that the Fornell-Larcker criterion can miss, improving detection of poor discriminant validity.
- Routinely reported in SEM and scale studies, making findings immediately interpretable by reviewers and readers.
- Detecting failed discriminant validity early prompts revision of the measurement model before costly data collection.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Apply discriminant validity tests whenever you develop or validate a multi-construct scale and report a structural equation model with two or more latent variables. It is required when submitting scale validation studies to peer-reviewed journals. Do not rely on discriminant validity evidence alone — it must accompany convergent validity evidence. Avoid discriminant validity tests when constructs are intentionally conceptualized as facets of a broader single construct (e.g., sub-dimensions of a unidimensional scale), as some correlation is theoretically expected and the distinction does not apply in the same way.
Strengths & limitations
- Provides direct empirical evidence that conceptually distinct constructs are also statistically distinct, strengthening theoretical precision.
- Multiple established criteria (Fornell-Larcker, HTMT, chi-square difference) allow triangulation across methods.
- HTMT is sensitive to moderate levels of overlap that the Fornell-Larcker criterion can miss, improving detection of poor discriminant validity.
- Routinely reported in SEM and scale studies, making findings immediately interpretable by reviewers and readers.
- Detecting failed discriminant validity early prompts revision of the measurement model before costly data collection.
- The Fornell-Larcker criterion has been shown to be liberal and can fail to detect inadequate discriminant validity even when HTMT reveals problems.
- All criteria depend on model fit: poor fitting CFA models yield unreliable validity estimates.
- HTMT thresholds (0.85 vs. 0.90) remain debated, and no single cutoff covers all research contexts.
- Discriminant validity is sample- and context-dependent; evidence obtained in one population may not generalize.
- For formative (composite) measurement models, reflective-based criteria such as AVE and HTMT do not directly apply.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between discriminant validity and convergent validity?
Convergent validity shows that indicators of the same construct are highly related to each other (typically assessed via factor loadings and AVE). Discriminant validity shows that different constructs are sufficiently distinct. Both are required for a complete case of construct validity; they address opposite ends of the validity spectrum.
Which criterion should I report — Fornell-Larcker, HTMT, or the chi-square difference test?
Current best practice is to report HTMT as the primary criterion, supplemented by the Fornell-Larcker criterion. The chi-square difference test is useful for targeted pairs of highly similar constructs. Relying solely on Fornell-Larcker is outdated given evidence of its insensitivity.
What HTMT threshold should I use?
The original Henseler et al. (2015) recommendation is 0.85 as a conservative threshold. A threshold of 0.90 is sometimes used for conceptually adjacent constructs. Additionally, a bootstrapped 95% confidence interval for HTMT that excludes 1.0 provides inferential support for discriminant validity.
Can discriminant validity fail even if factor loadings are high?
Yes. High factor loadings establish convergent validity within a construct but say nothing about how much overlap exists between constructs. Two constructs can each have high internal loadings yet still correlate so strongly with each other that discriminant validity fails.
Does discriminant validity apply to formative measurement models?
Not directly. AVE-based criteria and HTMT are designed for reflective models, where indicators are caused by the latent construct. For formative composites, discriminant validity is assessed differently, typically by examining the full collinearity between composites rather than through AVE comparisons.
Sources
- 1.Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56(2), 81–105.
- 2.Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Discriminant Validity. ScholarGate. https://scholargate.app/psychometrics/discriminant-validity