Growth Modelling in Experimental Designs
Message/Author
 Kieran Ayling posted on Wednesday, August 14, 2019 - 3:09 am
Hi, I am trying to do some growth modelling on some data from an experimental study, where various psychological measures were repeatedly measured over the following two years. Its a screening study so there were two arm (screened or not screened), but really there are three groups I wish to model and compare (no screening, screened: positive result, screened: negative result).

I read the Muthen & Curran - 1997 Psych Methods Paper dealing with this in a two group scenario but even with this I am struggling to understand what Mplus syntax is required at each of the 5 steps in the approach described. Is there a worked example somewhere I might be able to look at, and modify for the specifics of this study?
 Bengt O. Muthen posted on Wednesday, August 14, 2019 - 6:37 am
Take a look at the video and handout of our Short Course Topic 3 (or 4) that you find on our website:

http://www.statmodel.com/course_materials.shtml
 Kieran Ayling posted on Friday, August 23, 2019 - 3:27 am
Thanks Bengt, so I have been back to those and trying to follow along with the 1997 psych methods paper, but I start becoming confused at step 3.

Am looking at the syntax on page 70 of handout 4, which I believes relates to this (perhaps without t ON I statement - step 4?) trying to understand what is happening. Understand that variance and mean of slope etc are constrained to be the same, with t being the treatment effect. But there are bits I am not sure what they are doing was hoping you could explain those lines I've put * in front of:

i s q | y1@0 y2@1 y3@2 y4@3 y5@5 y6@7 y7@9 y8@11;
i t | y1@0 y2@1 y3@2 y4@3 y5@5 y6@7 y7@9 y8@11;
*[y1-y8] (1);
*[i@0];
i (2);
s (3);
i WITH s (4);
[s] (5);
[q] (6);
t@0 q@0;
*q WITH i@0 s@0 t@0; y1-y7 PWITH y2-y8;
t ON I;
 Bengt O. Muthen posted on Friday, August 23, 2019 - 6:24 am
[y1-y8] (1); holds the intercepts equal across time. This is an alternative parameterization for growth modeling which usually fix these at zero. In this parameterization, you instead fix the intercept factor mean at zero which you see on the next line. I chose this alternative version for ease of interpretation in this case.

q WITH i@0 s@0 t@0; says that the quadratic factor is uncorrelated with the other factors - which is natural because we have already fixed its variance at zero so it wouldn't make sense to have its covariances free (this is actually automatically done by Mplus now)