Exploratory Factor Analysis (EFA)

EFA looks for structure rather than testing it. Given a set of items, it identifies how many underlying factors plausibly account for their shared variance and which items group together. It is the right tool when an instrument is new, adapted from another context, or being applied to a new population.

Three decisions shape the result: extraction method (principal axis factoring or maximum likelihood for latent constructs, PCA for pure data reduction), rotation (oblique such as Promax when factors are expected to correlate, orthogonal Varimax when they are not), and how many factors to retain — eigenvalues over 1 is the crude default, while parallel analysis and the scree plot are more defensible.

Check sampling adequacy before interpreting anything: KMO above 0.60 and a significant Bartlett's test of sphericity. Then read the pattern matrix, retaining items that load cleanly above about 0.50 on a single factor and reconsidering items that cross-load or fall below the threshold. EFA normally precedes CFA on a separate sample.