I have a dataset of 77 items, ranked by 17 people, with many ties (actually: Q-sorted under a forced quasi-normal distribution in 15 bins). I am interested in common patterns of sorting items across people, so I want some exploratory factor analysis (PCA or PAF, probably) with items as cases, and people as variables (I know, it's odd - it's called Q Methodology.

The end result should be factor scores on these items, as "ideal-typical" ways to rank order these items. (Again, that's not me, that's Q).

I want to assume / preserve the ordinal nature of this data, so that leaves Spearman's or Kendall's as correlation coefficients for the correlation matrix. (I'm vaguely aware of Polychoric correlations, but I have no idea how that works, so I'd rather not touch it).

Two questions:

  1. Can I use PAF and/or PCA on a Spearman's and/or Kendall's correlation coefficient?
  2. How would I have to interpret the results differently (from a Pearson's based PAF/PCA) – if at all?

I know there are some answers around this issue, but there remains disagreement in answers and comments, as well as the literature. Also, the Q-circumstances are a bit special (or not, I don't know).

  • This A seems to suggest Spearman works (no word on Kendall), though with different interpretation.
  • This Q on SSCP seems to have to do with it.
  • This A says "classic (linear) factor analysis" is only for Pearson's
  • Basilevsky 2009 says "A Factor analysis can be based on a correlation matrix using Kendall's tau. Nothing new of major importance arises with respect to the interpretation of the results (...).

Ps.: since this (also) pertains to Q-Methodology, I'd appreciate if someone could add qmethod as a tag; see QMethod for a description and more.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.