Statistics Glossary → Factor extraction method (ML, PAF, minres)
Factor extraction method (ML, PAF, minres)concept
The method that sets the criterion by which factor loadings are estimated from the observed correlation matrix in exploratory factor analysis.
Maximum likelihood (ML) assumes the data are multivariate normal and finds the loadings that make the observed correlations most probable; this yields a chi-square test of fit, fit indices such as RMSEA, and standard errors for the loadings. Principal axis factoring (PAF) places communality estimates on the diagonal of the correlation matrix and updates them iteratively; it needs no distributional assumption. Minres (minimum residual) or unweighted least squares (ULS) minimizes the sum of squared off-diagonal differences between the observed and model-implied correlations. Generalized least squares (GLS) weights these differences so that variables with large unique variance count for less.
When the data depart markedly from normality or the sample is small, PAF and minres usually give more stable solutions and are widely used with Likert-type items; ML is preferred when a test of fit and statistical inference are required. The principal components (PC) method analyses all of the variance without separating common from unique variance, so strictly speaking it is a data-reduction method rather than factor analysis, and it tends to show higher loadings. With a well-defined factor structure, different methods often give similar results; very different results may signal a weak structure.
An everyday example
A psychologist investigates how many dimensions a 20-item anxiety scale has. The items use a five-point Likert format and the responses are skewed, so she chooses to extract factors with PAF or minres. If a reviewer asks her to show the model's fit with chi-square and RMSEA, however, she will also need to report the ML solution.
In MerQur
It is chosen with the 'Extraction method (ML, PAF, Minres/ULS, GLS, PC)' parameter in Exploratory Factor Analysis; the default is maximum likelihood. In Cronbach's Alpha and Reliability Analysis, McDonald's omega_t is computed from a one-factor minres solution.
Analyses where this concept appears
Cronbach's Alpha and Reliability AnalysisExploratory Factor Analysis (EFA)
Related concepts
Factor rotation (varimax, promax, oblimin)CommunalityModel fit indices (χ², CFI, TLI, RMSEA, SRMR)