Corinna posted on Wednesday, April 01, 2015 - 5:13 am
Hello, I am currently running a GMM with a categorical outcome and several predictors (categorical and continuous). Based on the various literature, I am not quite sure if it is better to first run an unconditional model and decide on the number of classes and only then add the predictors in later separate analyses or if I can immediately enter the predictors and decide on the best model already with the predictors included? Moreover, I wanted to ask if it is possible to include predictors that were only estimated at the second assessment point but not at the first one (when I have four assessment points for my analysis)? I would be very thankful for a short answer. Many greetings
I would make it simple and first decide on the number of classes without covariates. Then add covariates and see if the number of classes decision changes. If it does, see if there is a need for direct effects from some covariates to some latent class indicators.
I would not use later-occurring covariates to influence the latent class variable if the latter refers to class membership at the beginning.
Corinna posted on Thursday, April 02, 2015 - 12:40 pm
Hey, thank you very much for your answer. I followed your advice: When doing the analysis with the unconditional model a 2-class solution fits the best. But if I am conducting a conditional model I get a 3-class solution. Moreover, the data then show direct effects of some of the covariates on the classes. Does this then legitimate to add the covariates at the beginning when deciding on the number of classes? Thank you very much!