Roger Brown posted on Tuesday, September 18, 2007 - 1:22 pm
I am trying build a typical GMM on continuous outcome measures (upper section of the picture), then trying to model the categories (C) as a function of the intercept and slope of another growth model (see lower section of the picture).
A am a little unsure how to set up the syntax for this model. Both growth models cannot come under %OVERALL% since that would establish latent classes on both growth structures, correct?
My initial thought was:
title: growth curve modeling sleep data: file is sub1.dat; Variable: names are id m0 m2 m4 m6 m8 p1 p2 p3 p4 p5; usevariables are m0 m2 m4 m6 m8 p1 p2 p3 p4 p5; missing all .; classes = c(2); analysis: type = mixture; algorithm = integration; model: %overall% i s | m0@0m2@2m4@4m6@6m8@8; pi ps | p1@1p2@2p3@3p4@4p5@5; c#1 on pi p2;
Roger Brown posted on Wednesday, September 19, 2007 - 4:20 pm
Linda - Bengt,
I can get the above model working by staging it, first building the GMM and writing out the class probabilities and classes, and then reloading the data with this class variable into a new latent growth model with intercept and slope predicting the class variable. Not very elegant, but seems to work. Any idea how I could do this as a single inclusive run? Thanks.
Your initial thought is good. To not have the P process depend on c you should hold its parameters equal across classes. Given the defaults, this means holding its growth factor means equal across classes. You do that by giving equality statements () in Overall. That should work. Looks like an interesting model.