Student posted on Tuesday, April 28, 2015 - 9:39 am
Hi Drs. Muthen,
I wondered what the best approach would be to use GMM or LCGA with a latent factor outcome? Is there a way to adapt Example 8.6? For example, would this adaptation of the code work, if I was interested in whether class trajectory membership predicts latent outcome "o"?:
VARIABLE: NAMES ARE y1–y4 x o1 o2 o3; CLASSES = c(2); ANALYSIS: TYPE = MIXTURE; MODEL: %OVERALL% i s | y1@0y2@1 y3@* y4@*; o BY o1 o2 o3; i s ON x; c ON x; o ON c x;
UG ex 8.6 shows that you don't say "o ON c" because the o means change by default over the c classes (that's what the on ON c regression means).
Student posted on Tuesday, April 28, 2015 - 12:04 pm
I'm still unclear as to how to test for association between class membership and latent outcome. I understand that for observed variable outcomes, you just add them to the USEVARIABLES to get the mean of each class, then can test for differences using MODELTEST. However, when the outcome is a latent variable, what would you do?
If you identify the latent variable in the model command, does Mplus output just automatically give you the means for each class for the latent outcome? Please see code below (I just removed the "o ON c" command):
MODEL: %OVERALL% i s | y1@0y2@1 y3@* y4@*; o BY o1 o2 o3; i s ON x; c ON x; o ON x;
Yes, the o factor gets means estimated as the default for all classes except a reference class where it is zero.
Student posted on Tuesday, April 28, 2015 - 10:55 pm
Thank you- I had an additional question:
In Mplus, is there a way one could potentially test for indirect effects through a latent mediator variable in this scenario? For example, trajectory membership -> mediator (latent variable) -> outcome (latent variable)?