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

Exploratory factor analysis (EFA) is a data-reduction method used to test whether a set of questionnaire items actually clusters into a smaller number of underlying, unobservable constructs. It matters for scale development because it turns theory-driven item groupings into an evidence-based structure: if items intended to measure “ethical responsibilities,” “research and development responsibilities,” and “philanthropic responsibilities” don’t load together in the data, the proposed scale needs revision.

EFA works by examining systematic interdependence among observed variables—survey items respondents answer—and grouping them based on how strongly they correlate. The goal is to summarize information from many variables into fewer factors that can be named and interpreted. In the example used for scale development, the researcher created 19 items for “University social responsibility,” grouped theoretically into three dimensions: 7 ethical responsibility items, 6 research and development items, and 6 philanthropic responsibility items. EFA then checks whether the empirical correlations support that three-factor structure by looking for “factor loadings,” which represent how strongly each item relates to each extracted factor.

EFA is contrasted with confirmatory factor analysis (CFA). CFA is used when there’s a prior theory to test—relationships are confirmed rather than discovered. EFA is used when prior knowledge is limited or when no established scale exists, because it searches for latent patterns and reduces variables into a smaller set of composite factors.

Before running EFA, several diagnostics determine whether the data are suitable. Kaiser-Meyer-Olkin (KMO) measures sampling adequacy; values above 0.50 indicate factor analysis is appropriate. Bartlett’s test of sphericity checks whether the correlation matrix is meaningfully different from an identity matrix (i.e., whether variables are correlated enough to justify factor extraction). Commonality reflects how much variance an item shares with the factor solution; low commonality suggests poor fit. The number of factors is guided by eigenvalues (typically retaining factors with eigenvalue > 1), supported by a scree plot and/or percentage of variance explained (a common rule of thumb is retaining factors that explain roughly 60–70% of variance).

Running EFA in SPSS involves selecting Dimension Reduction → Factor, using principal component extraction, and applying rotation to clarify the factor structure. The session emphasizes Varimax rotation (orthogonal), which aims to produce a pattern where items load strongly on one factor and weakly on others. Model fit is assessed by comparing reproduced vs. observed correlations; a key rule of thumb is that non-redundant residuals should be under 50%.

In the worked example, the initial EFA output produced a three-factor solution consistent with expectations, with KMO = 0.931, a significant Bartlett test, and a cumulative variance close to 60% for three factors. However, some items failed to load adequately (factor loadings below the 0.50 threshold), including rd1, rd2, and pr1. After removing those problematic items one by one and re-running the analysis, the factor structure became clean: ethical responsibilities, research and development responsibilities, and philanthropic responsibilities aligned with the intended three-factor model. Reporting then follows a structured template: document the extraction method, rotation, loading and commonality criteria, KMO and Bartlett results, items removed and why, and the final factor loadings and variance explained.