SPSS FACTOR procedure for exploratory factor analysis

Description

Running exploratory factor analysis in SPSS with deliberate choices of extraction, retention and rotation rather than the misleading menu defaults.

SPSS FACTOR defaults to principal components extraction, the Kaiser eigenvalue-above-1 rule and Varimax rotation, a combination methodologists have long criticised (Costello and Osborne, 2005). Done properly, FACTOR is run from syntax with principal axis factoring or maximum likelihood, a fixed number of factors chosen by parallel analysis (run via a macro such as O'Connor's or the Python plugin), and an oblique rotation (Oblimin or Promax) when constructs are expected to correlate. Output includes KMO, Bartlett's test, communalities, pattern and structure matrices and the factor correlation matrix. Rigour comes from overriding defaults with justification and interpreting the pattern matrix after oblique rotation.

Attributes

Loss type
Siloed
Stage
Evaluation
Effort
Days

Steps

  1. Screen the data and confirm a sample of at least about 200 with adequate item variance.
  2. Run FACTOR with /PRINT KMO and check KMO of 0.60 or above and a significant Bartlett's test.
  3. Determine the number of factors with parallel analysis (O'Connor's SPSS syntax or the Python plugin) rather than the eigenvalue-above-1 default.
  4. Specify /EXTRACTION PAF or ML and /CRITERIA FACTORS(k), with /ROTATION OBLIMIN or PROMAX unless factors are theoretically orthogonal.
  5. Interpret the pattern matrix, flagging loadings below about 0.40 and cross-loadings above about 0.30, and check communalities.
  6. Save the syntax and iterate item removal one item at a time, then confirm in a new sample.

Sources

  • SPSS FACTOR
  • RELIABILITY procedures; syntax automation; parallel analysis via Python plugin
  • Field, A. (2024) Discovering Statistics Using IBM SPSS Statistics, 6th ed.
  • Costello, A. B. & Osborne, J. W. (2005) Best practices in exploratory factor analysis. Practical Assessment, Research & Evaluation
  • O'Connor, B. P. (2000) SPSS and SAS programs for determining the number of components using parallel analysis. Behavior Research Methods
  • The Inquiry Series, C7 — SPSS for Survey Analysis — EFA, Reliability & Scale Validation

Also known as

  • SPSS factor analysis
  • FACTOR command

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics