Polytomous Confirmatory Factor Analysis
Also known as: CFA for ordered categories, ordinal CFA, categorical CFA, WLSMV-CFA
Polytomous confirmatory factor analysis (CFA) tests a pre-specified factor structure when items have three or more ordered response categories (e.g., Likert scales). By working with polychoric correlations and robust estimators such as WLSMV, it avoids the distortions that arise when ordered categorical data are treated as continuous.
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 polytomous CFA when you have a theorised factor structure and your items carry three or more ordered response categories. It is the correct choice over standard (continuous) CFA whenever Likert-type or graded-response items are the primary data, particularly when category distributions are skewed or floor/ceiling effects are present. Do not use it when items are truly continuous or when the factor structure is unknown and needs to be discovered — in the latter case, polytomous EFA should precede CFA. Also avoid it if sample size is very small (below roughly 200 cases), as WLSMV requires adequate cell frequencies to estimate thresholds stably.
Strengths & limitations
- Correctly handles the ordinal, non-interval nature of polytomous items by working with polychoric correlations rather than Pearson correlations.
- WLSMV is robust to violations of multivariate normality that are common in attitude and personality scales.
- Provides formal hypothesis tests of the theorised factor structure along with global and local fit indices.
- Allows direct comparison of nested models (e.g., constrained vs. unconstrained loadings) via chi-square difference tests.
- Supports downstream measurement invariance testing across groups with the same estimator framework.
- Yields threshold estimates that are directly interpretable as cut-points on the underlying latent response continuum.
- Requires a reasonably large sample — thresholds and polychoric correlations are estimated from sparse category-pair frequencies and become unstable with small samples.
- WLSMV fit statistics behave differently from ML-based indices; researchers must be familiar with adjusted chi-square differences and WRMR.
- The polychoric correlation model assumes bivariate normality of underlying latent responses, an assumption that cannot be directly verified.
- If the theorised structure is wrong, the model will misfit but will not automatically suggest a better structure — misspecification diagnosis requires careful inspection of modification indices.
Frequently asked
Why not just run standard CFA with ML when my items are on a 5-point Likert scale?
Treating ordinal categories as continuous and using ML with Pearson correlations underestimates factor loadings and distorts chi-square and RMSEA values, particularly when distributions are skewed. WLSMV with polychoric correlations is the established solution for ordered categorical items.
How large a sample do I need?
As a rough rule of thumb, at least 200 cases are needed for stable polychoric correlation estimates and reliable WLSMV fit statistics. More categories per item and more symmetric distributions allow slightly smaller samples; sparse cells in extreme categories demand larger ones.
Can I compare fit across groups (measurement invariance) with polytomous CFA?
Yes. Measurement invariance testing for polytomous items follows the same sequence of configural, metric, and scalar models as in standard CFA, but thresholds replace intercepts and the appropriate chi-square difference test must account for WLSMV scaling.
What is the difference between polytomous CFA and the graded response model in IRT?
Both model ordered categorical item responses via thresholds and a latent trait, and they are mathematically closely related. Polytomous CFA is estimated in a covariance structure framework and emphasises global model fit and factor structure; the graded response model is estimated in an IRT framework and emphasises item-level parameter interpretation and person scoring. For large, carefully calibrated item banks, IRT is preferred; for confirmatory scale validation, polytomous CFA is standard.
Which software supports polytomous CFA?
Mplus is the most widely used package and implements WLSMV natively. The R packages lavaan (with estimator='WLSMV') and sem also support ordinal CFA, and OpenMx provides flexible model specification for advanced users.
Sources
- Flora, D. B. & Curran, P. J. (2004). An empirical evaluation of alternative methods of estimation for confirmatory factor analysis with ordinal data. Psychological Methods, 9(4), 466–491. DOI: 10.1037/1082-989X.9.4.466 ↗
- Muthen, B. (1984). A general structural equation model with dichotomous, ordered categorical, and continuous latent variable indicators. Psychometrika, 49(1), 115–132. DOI: 10.1007/BF02294210 ↗
How to cite this page
ScholarGate. (2026, June 3). Polytomous Confirmatory Factor Analysis. ScholarGate. https://scholargate.app/en/psychometrics/polytomous-confirmatory-factor-analysis
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.
- Confirmatory factor analysisPsychometrics↔ compare
- Item Response TheoryPsychometrics↔ compare
- Measurement InvariancePsychometrics↔ compare
- Ordinal CFAPsychometrics↔ compare
- Polytomous EFAPsychometrics↔ compare