Types of treatment in trajectory classes PreviousNext
Mplus Discussion > Growth Modeling of Longitudinal Data >
Message/Author
 Greta posted on Thursday, March 20, 2008 - 3:35 am
Hi, I would like to test a Growth Mixture Model with longitudinal data. I assume that different interventions produce different within-class trajectories.
My question is: how can I reflect this assumption in the model statement? I thought of regressing the slope on intervention status for each class separately. But I do not really know how to handle the treatment variable (three different treatments). Should I specify it as a GROUPING variable, use the KNOWCLASS option or simply introduce it dummy coded?
Do I have to make any specifications in the GRAPHS option, in order to get "splitted" trajectories for the different interventions (similar to Fig.2 in Muthén et al. 2002, Biostatistics)?

Thanks a lot!
 Linda K. Muthen posted on Saturday, March 22, 2008 - 9:39 am
You can create two dummy variables to represent the three treatments or use a multiple group approach. You don't need any special specifications for the PLOT command.
 Susan E. Collins posted on Thursday, September 25, 2008 - 6:27 pm
Hi there,

I am planning a GMM model like the one in Muthen et al. (2002)--i have this figure from p.150 of the topic 6 short course notes-- however, instead of dummy-coding my 3-group treatment variable, i wanted to make it a knownclass variable. that way, i would regress the slope factor on latent class, treatment group and the effect of tx group would be free to vary across latent class.

my problem is that I wanted to add a continuous, measured covariate "X" and look at the interaction between the 3-group, knownclass treatment and the covariate_X, to determine whether the effect of covariate_X on the slope varies across the known classes (treatment group) AND/OR across latent class. I am not sure if this is possible? how? or should i just dummy code tx and use the 2 dummy-coded variables and 2 treatmentxcovariate_x interactions?

thanks for any advice you might have!

--susan
 Bengt O. Muthen posted on Thursday, September 25, 2008 - 6:43 pm
For what you describe in the first paragraph it seems like you don't need a knownclass variable, but can do it via 2 dummies and the latent class variable.

But to answer paragraph 2, you can do this via the construction:

%cg#1.c#1%
s on x;
%cg#1.c#2%
s on x;
etc

that combines the knownclass cg variable with the real class c variable. Then you can allow any equalities (or not) across these combinations that you like.
 Susan E. Collins posted on Saturday, September 27, 2008 - 10:57 am
Thanks so much for the info! that was really helpful. Unfortunately, even with that addition, my model is not converging. I think it might be because of the start values, but i am not sure. b/c the code is a little long, I sent it and our previous exchanges here to the support email address. Thanks so much if you get a chance to give me any tips on where my code is going wrong...
--Susan
 Bengt O. Muthen posted on Saturday, September 27, 2008 - 11:04 am
I would not give start values, but use the program defaults.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: