I have 4 observed variables for husbands which represent scores for husbands at the start of marriage and at 3 annual follow ups. I also have 4 parallel variables for wives. I have missing data as some couples left the study, but I want to use all available data. I can easily create latent variable intercepts and slopes for husbands and for wives. But, I want to use these 4 latent variables to predict four outcome categories at year 4 (still participating [reference group], divorced, withdrawn from the study by intent, and dropped from the study for lack of response). Ordinarily, I would conduct a multinomial logistic regression. But I want to do so with the latent variables intercepts/slopes estimated with missing data. I have looked through the manual, but have not seen any examples of doing a multinomial logistic regression with latent variable predictors. Can this be done in one analysis? Do I need to do the growth curve analyses first and save the resulting latent variables for input into the multinomial logistic analysis?
bmuthen posted on Tuesday, May 13, 2003 - 10:16 pm
This cannot be done in version 2, but will be available in version 3 due out this fall. Multinomial logistic regression in Mplus is handled by latent class analysis and in version 2 latent classes cannot be influenced by latent continuous variables. So the answer is yes to the last question.