Brian Don posted on Monday, November 11, 2013 - 2:31 pm
I am working on a latent class growth analysis with 4 waves of data. For class 1, the intercept, slope, and quadratic terms were significant, while for the 2nd class only the intercept was significant. When running conditional analyses, I have regressed the predictors only onto the slope and intercept of the 2nd class (as seen in the following syntax), because of the fixed quadratic term.
You can do it this way, but it is fine to keep the quadratic slope in the second class - when adding predictors of it, it might have significant slopes, which would mean it varies (due to more power for instance).
Brian Don posted on Tuesday, November 19, 2013 - 10:01 am
Thanks very much for your help. One last question: I've been having problems with the classes switching when I attempt to fix the quadratic component for class 2. This occurred in a previous analysis about a year ago, and I was instructed by Linda to use the STARTS = 0 command and use starting values drawn from the unconditional model. So, I used the following syntax and just wanted to be sure this was correct:
ANALYSIS: TYPE = COMPLEX MIXTURE; Starts = 0;
MODEL: %OVERALL% i s q | AnxM1@0AnxM2@3AnxM3@6AnxM4@11; i-q@0; c on sexf Dep SE Psup RAS_full; %c#1% [i*0.396 s*-0.048 q*0.003]; i s q on sexf Dep SE Psup RAS_full; %c#2% [i*1.038 s*-0.037 q@0]; i s on sexf Dep SE Psup RAS_full;