Ordinal Discriminant Validity
Ordinal Discriminant Validity Assessment · Also known as: discriminant validity for ordinal data, polychoric discriminant validity, ordinal HTMT, ordinal AVE-based discriminant validity
Ordinal discriminant validity assesses whether a latent construct measured by ordinal (Likert-type) items is empirically distinct from other constructs in the same instrument. It applies polychoric correlations and ordinal-appropriate factor loadings to standard discriminant validity criteria such as the Fornell-Larcker rule and the Heterotrait-Monotrait ratio (HTMT), ensuring that validity conclusions are not distorted by the non-continuous nature of ordered-response data.
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 ordinal discriminant validity when your scale contains Likert or other ordered-response items and you need to demonstrate that two or more theoretically distinct constructs are also empirically distinct. It is the appropriate choice whenever a standard CFA or SEM analysis uses WLSMV or similar ordinal estimators. Do not use standard (Pearson-based) discriminant validity procedures on ordinal items with fewer than about five response categories, or when the item distributions are highly skewed — in those cases the polychoric approach is essential. Ordinal discriminant validity is not needed when items are genuinely continuous (e.g., reaction times, physiological measures), in which case standard ML-based CFA with Pearson correlations is appropriate.
Strengths & limitations
- Corrects for the attenuation of correlations caused by treating ordinal data as continuous, producing less biased validity estimates.
- Compatible with modern SEM estimators (WLSMV, ULS) that are specifically designed for ordinal data.
- Supports both loading-based (AVE, Fornell-Larcker) and correlation-based (HTMT) discriminant validity criteria within the same ordinal framework.
- Provides more accurate construct-level distinctions when response distributions are skewed or categories are few.
- Aligns with best-practice reporting requirements in journals that require ordinal-appropriate measurement models.
- Polychoric correlations require the assumption that each ordinal item is a categorisation of an underlying normal continuous variable; violations of this normality assumption can distort results.
- Large samples (typically n > 200) are needed for stable polychoric correlation estimates, especially with many items or few response categories.
- The HTMT threshold (0.85 vs. 0.90) is debated in the literature, and no single cutoff applies universally across all constructs and domains.
- Software support for ordinal discriminant validity (polychoric + WLSMV + AVE/HTMT) is less automated than standard continuous-data pipelines and may require manual computation of AVE from ordinal loadings.
- Discriminant validity evidence alone does not establish full construct validity; convergent validity, content validity, and nomological validity must also be evaluated.
Frequently asked
Why can I not simply use Pearson-based AVE for Likert items?
Pearson correlations applied to ordinal items underestimate the true linear relationship between the underlying continuous variables, which in turn attenuates factor loadings and AVE. This can make constructs appear less well-defined than they are, and can also mask discriminant validity failures. Polychoric correlations recover the latent continuous correlations and give more accurate results.
Which estimator should I use for the ordinal CFA step?
The most widely recommended estimator is WLSMV (mean- and variance-adjusted weighted least squares), available in lavaan (R), Mplus, and other SEM software. It uses the polychoric correlation matrix and provides accurate standard errors and fit statistics for ordered-categorical data. ULS (unweighted least squares) is a simpler alternative that is less sensitive to distributional assumptions.
What HTMT threshold should I apply?
The most common guideline is HTMT < 0.85 as a conservative criterion, or HTMT < 0.90 as a more lenient threshold. Henseler and colleagues (2015) originally suggested 0.85 for conceptually distinct constructs. The choice depends on how theoretically separable the constructs are — more similar constructs may require the stricter cutoff.
Can I assess ordinal discriminant validity in PLS-SEM?
PLS-SEM does not natively use polychoric correlations, but the HTMT criterion can still be computed from polychoric item correlations externally. For confirmatory purposes in the ordinal context, covariance-based SEM (CB-SEM) with WLSMV is generally preferred over PLS-SEM.
Does passing the HTMT test guarantee the constructs are distinct?
No. HTMT is a strong and sensitive criterion, but discriminant validity evidence is one component of construct validity. You should also check convergent validity (AVE >= 0.50, loadings >= 0.70), content validity, and nomological validity to make a comprehensive case that each construct is well-defined and distinct.
Sources
- Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56(2), 81–105. DOI: 10.1037/h0046016 ↗
- Henseler, J., Ringle, C. M., & Sarstedt, M. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1), 115–135. DOI: 10.1007/s11747-014-0403-8 ↗
How to cite this page
ScholarGate. (2026, June 3). Ordinal Discriminant Validity Assessment. ScholarGate. https://scholargate.app/en/psychometrics/ordinal-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
- Ordinal CFAPsychometrics↔ compare
- Ordinal EFAPsychometrics↔ compare
- Ordinal Reliability AnalysisPsychometrics↔ compare