Robust Exploratory Factor Analysis
Also known as: robust EFA, robust factor analysis, outlier-resistant factor analysis, EFA with robust estimation
Robust exploratory factor analysis discovers the latent factor structure of a set of items using estimation methods that are resistant to outliers and violations of multivariate normality. It applies the same measurement model as standard EFA but replaces classical covariance estimation with robust counterparts — such as minimum covariance determinant or iteratively reweighted least squares — so that a small fraction of atypical cases cannot distort the recovered factor loadings.
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 robust EFA when the data may contain multivariate outliers, when distributions of item responses are noticeably non-normal (heavy tails, skew, or contamination), or when clinical, educational, or applied samples are likely to include extreme scores. Robust EFA is especially valuable in early scale development, where item pools are untested and unusual response patterns are common. Do not use it as a routine replacement for standard EFA on clean, well-screened, approximately normal data — the additional complexity is unnecessary, and interpreting down-weighted cases requires care. Also avoid it when the sample is very small (n < 100 is often insufficient for MCD-based methods) or when ordinal items require a polychoric correlation approach that is not yet paired with a robust estimator in your software.
Strengths & limitations
- Produces factor loadings that represent the majority of respondents rather than being distorted by a few outlying cases.
- Provides a built-in outlier-detection mechanism — the case weights themselves flag influential observations for inspection.
- Retains the full interpretive framework of standard EFA (loadings, communalities, rotation) so results are familiar to psychometric audiences.
- More trustworthy in applied and clinical settings where extreme response patterns are substantively meaningful rather than simple data errors.
- Requires larger samples than standard EFA; MCD-based estimators are unstable with fewer than approximately 100 cases.
- Software support is less universal than for standard EFA; specialized packages (e.g., rrcov or robustbase in R) are often needed.
- Interpretation of down-weighted cases can be ambiguous — it is not always clear whether flagged observations are errors or genuinely informative extreme scores.
- Robust standard errors and fit statistics are not yet standardised across software, complicating cross-study comparison.
Frequently asked
How does robust EFA differ from standard EFA?
The factor model and rotation are identical. The difference lies in the estimation of the input covariance matrix: standard EFA uses the classical sample covariance, whereas robust EFA uses an outlier-resistant estimator such as MCD or IWLS. This makes the recovered loadings insensitive to a small fraction of extreme cases.
When should I prefer robust EFA over standard EFA?
Prefer robust EFA when you have substantive reason to expect outliers or non-normality — clinical samples, convenience samples with extreme scorers, or item pools that have not been pre-screened. In well-screened, approximately normal data, standard EFA is adequate.
What sample size do I need for robust EFA?
MCD-based robust EFA generally requires at least 100 cases, and preferably more, because the algorithm searches over subsets of observations. The standard EFA guidelines (at least 5 cases per item, at least 3 items per factor) remain necessary but are not sufficient on their own for robust methods.
Does robust EFA change the rotation step?
No. Rotation — whether orthogonal (varimax) or oblique (oblimin, geomin) — is applied to the robustly extracted factor matrix in exactly the same way as in standard EFA. The robustness enters through the covariance estimation, not through the rotation.
What should I report when using robust EFA?
Report the specific robust estimator used (e.g., MCD with a given breakdown point, or IWLS with the chosen weight function), the proportion of cases assigned low weights, the extraction method and rotation, the number of factors retained and the criterion used, and whether the factor solution differed meaningfully from a standard EFA on the same data.
Sources
- Yuan, K.-H., & Bentler, P. M. (2000). Robust mean and covariance structure analysis through iteratively reweighted least squares. Psychometrika, 65(1), 43–58. DOI: 10.1007/bf02294185 ↗
- Pison, G., Rousseeuw, P. J., Filzmoser, P., & Croux, C. (2003). Robust factor analysis. Journal of Multivariate Analysis, 84(1), 145–172. DOI: 10.1016/S0047-259X(02)00007-6 ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Exploratory Factor Analysis. ScholarGate. https://scholargate.app/en/psychometrics/robust-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
- EFAStatistics↔ compare
- Item Response TheoryPsychometrics↔ compare
- Robust Confirmatory Factor AnalysisStatistics↔ compare