I am comparing several factor rotation methods in a Q-methodological analysis while holding the extraction method constant as PCA. I use both orthogonal rotations (Varimax, Quartimax, Equamax) and oblique rotations (Promax and Direct Oblimin). In conventional Q methodology (qmethod package), defining Q-sorts are weighted using $$ w = \frac{l}{1-l^2}, $$ and these weights are used to calculate statement z-scores for the factor array. For orthogonal rotations this is straightforward because the rotated loadings are correlation-like coefficients. Under Promax or Direct Oblimin, however, the pattern matrix contains regression-like coefficients, which can exceed $|1|$ . In one of my Promax solutions, for example, a pattern coefficient is $1.008$ . Applying the conventional Q-weight formula gives a weight of approximately $-64.9$ (counterintuitive). Since I want to compare across different rotations, the derived z-scores for oblique rotations seem to be not comparable. For the oblique solution I have the factor-correlation matrix $\Phi$ and structure matrix $$ S = P\Phi. $$ I am trying to determine the statistically appropriate way to construct statement-level scores/z-scores for comparing viewpoints across orthogonal and oblique rotations . I have considered: Applying $$ \frac{p}{1-p^2} $$ to the pattern coefficients, as my Q-method software currently does. Applying the traditional weighting formula to the structure coefficients, $$ \frac{s}{1-s^2}, $$ because these are Q-sort-fa…

Full article content could not be extracted automatically. Read the original below.