In one study, I used GGMM to identify several trajectories of a dependent variable from grade 9 through 11) and found some very interesting results with all of my time-invariant covariates set at baseline (9th grade), and a distal outcome in the 11th grade. Now, a peer is asking about changes in the effects of the covariates over time. In other words, what would happen if covariate "A" were measured at 10th instead of 9th grade? If I wanted to assess the effects of covariate "A" at 10th grade without changing the membership of my 3 trajectories, what should I do?
I am doing a LGMM exercise using a single continuous measure (N>2000) over 4 time points; 0, 1, 2, 4 (represneting 0, 6, 12, and 24 months). Is it OK to use both linear and square root expressions of time in the same analysis? My model command is: Model: %overall% i by PCLb-PCLf@1; s by PCLb@0PCLd@1PCLe@2PCLf@4; sqrt by PCLb@0PCLd@1PCLe@1.414PCLf@2; [PCLb-PCLf@0 i s sqrt]
%c#1% [i*71 s*2 sqrt*-2.8]
The model runs properly and gives me a better (lower) BIC than either of the simple linear OR square root models. But I'm not sure if I can have two diferent expressions of time in the model.
For your advice please.
bmuthen posted on Friday, August 13, 2004 - 4:53 pm
I think this is ok. It is just another non-linear function of time.