In using Mplus 6.1 to fit a single-factor model with categorical indicators using WLSMV (Theta parameterization), I receive this error message in the output:
THE MODEL CONTAINS A NON-ZERO CORRELATION BETWEEN DEPENDENT VARIABLES. SUCH CORRELATIONS ARE IGNORED IN THE COMPUTATION OF THE FACTOR SCORES.
Can you please tell me what this means? I do not have any residual covariances in the model (nor do any show up unexpectedly in the output), so the only correlation between the dependent variables should be that predicted by the latent factor.
It looks like a new check added in Version 6.1 is not working as intended. The message should be ignored. The factor scores that are saved are correct.
Tait Medina posted on Wednesday, March 12, 2014 - 1:13 pm
My model (factor analysis with ordinal variables) contains correlated residuals (that is, one WITH statement). Using the SAVE=FSCORES options, is it possible to compute factor scores that take into account these residual covariances?
It sounds like you are using WLSMV where this is not allowed. You can use maximum likelihood to do this.
Tait Medina posted on Thursday, March 13, 2014 - 12:11 pm
I am using WLSMV and THETA parameterization. I am working with a factor model for ordinal items in multiple populations using the minimal conditions for identification discussed in Millsap and Yun-Tein, 2004. I am trying to extract factor scores, however I have correlated residuals and am obtaining this warning:
"THE MODEL CONTAINS A NON-ZERO CORRELATION BETWEEN DEPENDENT VARIABLES.SUCH CORRELATIONS ARE IGNORED IN THE COMPUTATION OF THE FACTOR SCORES."
The CATEGORICAL option is available with ML. You will need to specify your residual covariance using the BY option. WITH is not allowed with ML because each residual covariance requires one dimension of integration. When you specify: