I have a question about how model fit is computed when using TYPE = EFA versus TYPE = GENERAL with censored variables and the WLSMV estimator. My situation is outlined below.
I ran an EFA with censored variables using
TYPE = EFA 1 3; ESTIMATOR = WLSMV;
Based on the EFA, I decided on a single-factor model. I then used
TYPE = GENERAL; ESTIMATOR = WLSMV;
and specified the single-factor model in the MODEL command.
I was expecting the goodness of fit to be the same for these two analyses, but it was drastically different.
It appears that the discrepancy has to do with the specification of some variables as censored, because when I removed the CENSORED statement from the VARIABLE command, the goodness of fit was identical for both analyses.
Should it be the case that these two analyses yield different goodness of fit with censored variables? If so, could you please refer me to a citation that explains the discrepancy?