Description
Discovering the latent dimensions underlying a set of survey items, using parallel analysis to decide objectively how many factors to retain.
Exploratory factor analysis models item correlations as arising from a smaller number of latent factors. The critical decision is how many factors to retain. The widely used Kaiser rule (eigenvalues above 1) systematically over-extracts. Horn's parallel analysis compares observed eigenvalues with those from random data of the same size, retaining only factors whose eigenvalues exceed the 95th percentile of the random ones. A rigorous EFA checks factorability (KMO, Bartlett's test), uses common-factor extraction (principal axis or minimum residual, not principal components), applies an oblique rotation when factors are expected to correlate, and removes items with weak or cross-loadings iteratively. In UX it is mostly confined to academic scale development.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Siloed
- Stage
- Definition
- Data
- Quantitative
- Effort
- Days
Steps
- Ensure a sample of at least about 200, or 5-10 respondents per item, and clean the data first.
- Check factorability: KMO of 0.60 or above (0.80 or above preferred) and a significant Bartlett's test.
- Run parallel analysis with 1,000 random datasets and retain factors whose eigenvalues exceed the 95th percentile of random eigenvalues.
- Extract with principal axis or minimum residual factoring (factor_analyzer) and apply an oblique rotation such as oblimin or promax.
- Remove items with primary loadings below about 0.40 or cross-loadings above about 0.30, one at a time, re-running after each removal.
- Name factors from their highest-loading items and confirm the structure in a separate sample with CFA.
Sources
- factor_analyzer
- semopy
- pyreadstat (SPSS import)
- scikit-learn (MICE)
- pingouin (reliability)
- pandas
- Horn, J. L. (1965) A rationale and test for the number of factors in factor analysis. Psychometrika
- Fabrigar, L. R. & Wegener, D. T. (2012) Exploratory Factor Analysis
- The Inquiry Series, C3 — Survey Pipeline in Python — From Raw Responses to Structural Equation Model
- Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)
Also known as
- EFA
- Horn's parallel analysis
Categories
- Analytics & Behaviour
- Survey & Psychometrics