1

I have a tricky data set for factor analysis with 71 ordered likert-type items. The items have five levels. I'm using psych::fa.

I use ``avaan::lavCorto get a smoothed covariance matrix. to avoid errors withpolychoric`.

The base analysis I've tried is:

test1 <- fa(postImputedCor,
            nfactors = 11,
            n.obs = nrow(ddItemsOrdered),
            rotate = "oblimin",
            fm = "pa",
            smooth = TRUE,
            oblique.scores = TRUE,
            cor = "poly")

This uses the default "regression" method of weighting the scores, but I have tried them all ("Anderson", "tenBerge", "Thurstone"). I have also tried "uls" and "wls" in addition to "pa" as factoring methods.

The warning I get is:

 Warning message:
In fa.stats(r = r, f = f, phi = phi, n.obs = n.obs, np.obs = np.obs,  :
  The estimated weights for the factor scores are probably incorrect.  Try a different factor score estimation method.

Are there other options I should try in order to get reliable factor scores?

hare
  • 21
  • 1
    I am not an expert of psych and I don't know its options, but classic f. score estimation methods are not meant for the analysis of inferred correlations such as tetrachoric or polychoric. Special methods are needed for that, and I don't know if psych offers them. – ttnphns Feb 21 '23 at 13:14
  • And start with checking this https://stats.stackexchange.com/q/32152/3277 – ttnphns Feb 21 '23 at 13:15
  • 1
    Thank you for your comment! psych::fa does have an option for a polycoric correlation. It has functions for calculating these correlations, but they are generating errors when I try to use the data as input (rather than a a correlation matrix; both should be allowed according to the docs). – hare Feb 22 '23 at 21:18
  • The link you mentioned is about getting individual factor scores as output. While this would be useful, I'm concerned right now that analyses I'm doing have other incorrect outputs related to it (communalities, loadings, etc.). – hare Feb 22 '23 at 21:21
  • 1
    The core of my comment was that when the analysis is based on polychoric correlations, the traditional methods of computation of factor scores are inapplicable. – ttnphns Feb 23 '23 at 10:05
  • Thank you for this explanation. – hare Feb 24 '23 at 14:13
  • Confirming the warning occuring with ,cor="poly" https://stats.stackexchange.com/questions/618250/factors-or-components-in-stata-and-r-psych-lavaan-confusing-method-labels-an – Johan Jun 09 '23 at 09:37

0 Answers0