Exploratory Factor Analysis (EFA): Concept, Key Terminologies, Assumptions, Running, Interpreting

Exploratory Factor Analysis (EFA) is presented as a practical way to build and validate new measurement scales by compressing many questionnaire items into a smaller set of underlying, unobservable “latent” factors. Instead of treating dozens of observed variables as unrelated, EFA groups items that move together—based on their intercorrelations—so a construct like “service quality” can be represented by components such as teaching, administration, or leadership. That matters because researchers often need objective instruments for concepts that can’t be directly observed, and EFA helps reveal the factor structure when there’s limited prior knowledge about how items should cluster.

The transcript distinguishes EFA from Confirmatory Factor Analysis (CFA): EFA searches for latent patterns in the data, while CFA tests a pre-specified structure tied to hypotheses. For scale development, the workflow begins with data reduction—often using principal component analysis as the extraction method—then checks whether the dataset is suitable for factor modeling. Two core diagnostics are emphasized: the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity. KMO values between 0.5 and 1 indicate the sample is appropriate for factor analysis; values below 0.5 suggest the correlations are too weak or the sample is inadequate. Bartlett’s test should be significant, meaning the correlation matrix is not an identity matrix—there must be enough relationships among variables to justify factor extraction.

Once suitability is confirmed, EFA relies on several interpretive statistics. Commonality indicates how much variance a variable shares with the factor solution and should be sufficiently high (the transcript uses a threshold above 0.6). Uniqueness—computed as 1 minus commonality—should be low, since high uniqueness implies an item isn’t well explained by the factors. The number of factors to retain is guided by eigenvalues (Kaiser rule: eigenvalues greater than 1) and supported by scree plots, where the “elbow” suggests a reasonable cutoff. Factor loadings—correlations between items and factors—are then inspected to assign items to dimensions.

Rotation is treated as a key step for interpretability. Varimax (orthogonal) is recommended as a common choice in scale-development papers because it aims to minimize cross-loadings and keep factors independent (orthogonal). The transcript contrasts this with oblique rotations such as Direct Oblimin, which allow factors to correlate, and briefly notes other options like Quartimax and Promax. In practice, items that fail to load clearly (for example, not loading above a minimum threshold such as 0.5) or that cross-load meaningfully on multiple factors are candidates for removal to achieve a cleaner factor structure.

A worked example is described using a CSR scale with multiple dimensions (development responsibilities, ethical responsibilities, relational responsibilities, and information sharing responsibilities). After running EFA with principal component analysis and Varimax rotation, the solution yields four factors, with the retained components explaining a substantial portion of variance (about 57.8% in the example). The transcript ends with guidance on reporting: document the extraction and rotation method, minimum loading criteria, commonality/uniqueness checks, KMO and Bartlett’s test results, the final factor solution, and which items were deleted and why (typically due to poor loadings or problematic cross-loadings).