Ordinal Exploratory Factor Analysis
Also known as: ordinal factor analysis, polychoric EFA, categorical EFA, EFA for ordinal data
Ordinal exploratory factor analysis discovers latent factors underlying a set of ordinal items — typically Likert scales — by computing polychoric correlations among the items and then applying a weighted least squares estimator. It avoids the distortions that arise when continuous EFA methods are naively applied to ordered categorical responses.
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 EFA whenever items are measured on ordered categorical scales — Likert-type scales with roughly 2 to 7 categories — and your goal is to discover the latent factor structure without a pre-specified model. It is the appropriate default for scale development and validation studies in psychology, education, and health research. Do not use standard Pearson-based EFA on Likert items with five or fewer categories, as this systematically underestimates loadings and can produce spurious factors. Ordinal EFA is not needed when items already yield approximately symmetric, continuous distributions with at least seven ordered response options. It also requires larger samples than continuous EFA; fewer than 200 respondents can make polychoric estimation unstable, especially for rare response categories.
Strengths & limitations
- Corrects the attenuation of factor loadings that occurs when Pearson correlations are computed on coarse ordinal scales.
- Produces statistically consistent estimates of factor loadings because it models the true continuous underlying response variables.
- WLSMV estimation provides global fit indices (RMSEA, CFI, SRMR) that allow formal evaluation of the extracted factor solution.
- Handles asymmetric and skewed item distributions better than normal-theory maximum likelihood applied to ordinal data.
- Directly comparable to confirmatory factor analyses using the same estimation framework, facilitating the EFA-to-CFA validation pipeline.
- Polychoric correlation estimation is computationally intensive and can be numerically unstable with sparse frequency cells, requiring larger samples than standard EFA.
- Not all software supports ordinal EFA with polychoric matrices and weighted least squares; implementation knowledge is required.
- The assumption of bivariate normality for the latent response variables may not hold, and violations can bias polychoric estimates.
- With many items, the polychoric correlation matrix can be non-positive-definite due to estimation errors, requiring regularisation before factoring.
Frequently asked
When can I use standard EFA instead of ordinal EFA?
Standard Pearson-based EFA is acceptable when items have at least seven roughly symmetric response categories and the distribution of responses is approximately normal. With five or fewer Likert categories, or with skewed distributions, ordinal EFA with polychoric correlations is strongly preferred because Pearson correlations between ordinal variables are attenuated, leading to underestimated loadings and potentially wrong factor counts.
What sample size do I need?
Polychoric correlation estimation is more sample-hungry than Pearson. A minimum of 200 cases is a common rule of thumb, with larger samples (400+) recommended when items are skewed, categories are sparse, or the expected loadings are modest. With fewer than 100 cases, polychoric estimates are often unstable and should be treated with caution.
Which estimator should I choose — DWLS or WLSMV?
WLSMV (mean-and-variance-adjusted weighted least squares, implemented in Mplus as the default for ordinal CFA/EFA) is generally preferred because its chi-square statistic and fit indices are corrected for the non-normality of the weight matrix. DWLS (used in lavaan) produces equivalent parameter estimates but its standard fit indices are less accurate without adjustment. For most practical purposes in ordinal EFA, either estimator gives very similar factor loadings.
Can I mix ordinal and continuous items in the same analysis?
Yes, provided you use a mixed correlation matrix — polychoric for ordinal-ordinal pairs, polyserial for continuous-ordinal pairs, and Pearson for continuous-continuous pairs. Most modern software (e.g., lavaan, Mplus) can compute and analyse such a matrix, but sample size requirements increase and results should be interpreted carefully.
Does ordinal EFA replace item response theory for Likert data?
They answer related but distinct questions. Ordinal EFA reveals the number of latent dimensions and which items load on each — a structural question. Item response theory (IRT) models the probability of each response category as a function of the latent trait level — a measurement precision question. For unidimensional scales, graded response IRT models are often the next step after ordinal EFA confirms a single dominant factor.
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 ↗
- Muthén, 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). Ordinal Exploratory Factor Analysis. ScholarGate. https://scholargate.app/en/psychometrics/ordinal-exploratory-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
- Cronbach's AlphaStatistics↔ compare
- EFAStatistics↔ compare
- Item Response TheoryPsychometrics↔ compare
- Structural Equation ModelingResearch Statistics↔ compare