method note
Exploratory or confirmatory? Which factor analysis, and when
Should I use exploratory or confirmatory factor analysis?
Use exploratory factor analysis when you do not know how many dimensions the construct has, and confirmatory factor analysis when theory or prior work already proposes a structure. They are not rival methods but consecutive stages: EFA reveals a structure, CFA tests it on an independent sample. Running EFA then CFA on the same data is not confirmation.
Why both are needed
EFA asks the data how many dimensions are present and reads the answer off the data. CFA asks whether a proposed dimensional structure is consistent with the data — it tests a hypothesis. Asking the first question and reporting the second question's answer is the most common error in scale development.
If you test the structure EFA produced on the same sample with CFA, fit will look good — you built the model on that data. That is a tautology, not evidence of validity. Split the sample or collect a second one.
Checks before running EFA
The KMO measure of sampling adequacy and Bartlett's test of sphericity answer the prior question of whether factor analysis is applicable at all. KMO above .60 is the usual floor; a significant Bartlett test shows the correlation matrix differs from an identity matrix. Neither is evidence of validity; they only clear the way.
Choosing a rotation is a theoretical decision, not a technical one: orthogonal (such as varimax) if you expect the factors to be uncorrelated, oblique (promax, oblimin) if you expect them to correlate. In the social sciences dimensions usually do correlate, yet varimax is habitually chosen by default.
Reporting order
Sample and collection, KMO and Bartlett, extraction method, rotation and its justification, the factor-count decision and its criterion, the loading table, removed items with reasons, explained variance, reliability coefficients. For CFA, additionally: estimator, measurement model results, fit indices, and convergent/discriminant validity evidence.
The work behind this note
- Factor Analysis and Validity in Social Sciences: Application of Exploratory and Confirmatory Factor Analyses2017 · 2,435 citations
- Structural Equation Modeling (SEM) for Social and Behavioral Sciences Studies: Steps Sequence and Explanation2024 · 44 citations
- An Investigation of the Characteristics of Learning Organizations in Turkish Companies: Scale Validation2014 · 38 citations
Other method notes
- Which fit index to report, and whenWhich fit indices should I report in structural equation modelling?
- Validity and reliability: what to report, and howHow are validity and reliability reported, and how do they differ?
- Common method variance: preventing, testing, correctingWhat is common method variance and how is it controlled?
Written for researchers and graduate students. Conventions described here are conventions, not rules — check them against your own design before you report.