Scree plot vs. parallel analysis

Description

Comparing the subjective scree elbow with the objective parallel analysis criterion to decide how many factors a UX questionnaire really measures.

Retaining too many or too few factors is the most consequential error in EFA. Cattell's scree plot charts eigenvalues in descending order and retains factors above the 'elbow', which is subjective and can be ambiguous. The Kaiser rule (eigenvalues above 1) over-extracts. Parallel analysis compares observed eigenvalues against those from many random datasets of the same size, retaining factors whose eigenvalues exceed the random 95th percentile; simulation studies (Zwick and Velicer, 1986) found it among the most accurate rules. Velicer's MAP test is a further check. Rigour comes from using parallel analysis as the primary criterion, the scree and MAP as corroboration, and examining interpretability of adjacent solutions (k minus 1 and k plus 1).

Attributes

Loss type
Siloed
Stage
Evaluation
Effort
Hours

Steps

  1. Compute eigenvalues from the correlation matrix (reduced matrix for common factor analysis where possible).
  2. Draw the scree plot and note the elbow position independently before running parallel analysis.
  3. Run parallel analysis with at least 1,000 random datasets, retaining factors whose eigenvalues exceed the 95th percentile.
  4. Optionally run Velicer's MAP test as a further check.
  5. If criteria disagree, fit the k minus 1, k and k plus 1 solutions and choose based on interpretability and item loadings.
  6. Report all criteria and the rationale for the final choice.

Sources

  • SPSS FACTOR
  • RELIABILITY procedures; syntax automation; parallel analysis via Python plugin
  • Cattell, R. B. (1966) The scree test for the number of factors. Multivariate Behavioral Research
  • Horn, J. L. (1965) A rationale and test for the number of factors in factor analysis. Psychometrika
  • Zwick, W. R. & Velicer, W. F. (1986) Comparison of five rules for determining the number of components to retain. Psychological Bulletin
  • The Inquiry Series, C7 — SPSS for Survey Analysis — EFA, Reliability & Scale Validation
  • Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)

Also known as

  • Factor retention decision
  • Cattell's scree test vs. Horn's parallel analysis
  • Velicer's MAP test

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics