I have a sample of data with 1,609 cases. I'm trying to conduct CFA, build two factors, and examine different variables that can predict the factors. Unfortunately, every variable I'm using has missing data. Now, the factors' indicators (a1, a2 & b1, b2) are ordinal-level measures each with 5 categories. I wanted to use FIML to take care of the missingness. This is what I did under Scenario 1. It worked fine, and I retained my full sample of 1,609 cases. But, the model didn't converge when I used the WLSMV estimator, which is what I used under Scenario 2.
My question - How do I retain my full sample and account for the ordinal nature of my factor indicators if I use the WLSMV estimator?
VARIABLE: NAMES ARE: x1 x2 x3 a1 a2 b1 b2; USE VARIABLES x1 x2 x3 a1 a2 b1 b2; MISSING ARE ALL (-999);
ANALYSIS: ESTIMATOR IS MLR;
MODEL: Factor1 BY a1 a2; Factor2 BY b1 b2; Factor1 Factor2 ON x1 x2 x3;
x1 x2 x3;
VARIABLE: NAMES ARE: x1 x2 x3 a1 a2 b1 b2; USE VARIABLES x1 x2 x3 a1 a2 b1 b2; CATEGORICAL ARE a1 a2 b1 b2; MISSING ARE ALL (-999);
ANALYSIS: PARAMETERIZATION IS DELTA; ESTIMATOR IS WLSMV;
I want to save factor scores from a factor analysis. When I designate the items as categorical, the saved file with the factor scores contains fewer cases than when I designate the items as being continuous. I was assuming that was because the categorical option was handling missing data differently and not giving me factor score estimates for cases with some missing values but from the above discussion this would not appear to be the case. Does this make sense to you? Also, I am having a hard time getting the categorical analysis to provide SRMR in the output. This was the case even after including a "LISTWISE=ON;" statement in my syntax which from other posts on the discussion board indicated would result in a categorical model still giving us SRMR. Any help you can provide will be most appreciated.
I am trying to save factor scores and notice that when I designate the items as categorical, Mplus saves factor scores for fewer cases than when I designate the items as continuous. My first hunch was that categorical analysis excludes cases with missing values but the earlier posts in this thread suggest otherwise. Any thoughts about why the categorical option is saving factor scores for fewer cases? A second question about categorical/WLSMV analysis - from a post on the discusssion board I thought I could get SRMR in a categorical analysis as long as I included a MODEL = NOMEANSTRUCTURE; statement in the ANALYSIS command but this didn't produce SRMR in the output. Is there a way to get SRMR from a categorical model? Thanks!