Qiwen Zheng posted on Thursday, June 16, 2016 - 8:58 pm
I'm fitting a two-part random intercept mixture model, in which a binary (u) and a continuous (m) dependent variables are used. I'm having trouble to specify different residual var/cov structure of random intercepts of these two variables across classes.
The binary part uses a logistic model, while the MIXTURE is modeled ONLY ON the continuous part.
When I specify the residual variance/covariance of the random intercepts in each class in the BETWEEN part of the model, I got the FATAL ERROR: CLASS-SPECIFIC BETWEEN VARIABLE PROBLEM. The code runs successfully if I do not specify u, m and u with m in each class, in which case I then obtain equal var/cov estimates in both classes.
The data is longitudinal with each person with 8 repeated measures.
Below is the model for the continuous part: m_ij=a_0i+a_1*X1_ij+e_ij a_0i=a_0+b_01*X2_i+b_0i where i is person and j is time. u is a binary dependent variable and m is a continuous dependent variable. X1 is a time-varying covariate and X2 is a time-invariant covariate. Model for binary dependent variable has the same two covariates with logistic link function and random intercept.
(Continued in the next post)
Qiwen Zheng posted on Thursday, June 16, 2016 - 8:59 pm
My Mplus code: VARIABLE: Names are ID u m X1 X2; Missing are all (-999) ; usevariables are u m X1 X2; cluster = ID; within = X1; between=X2; CLASSES = c(2); categorical=u; ANALYSIS: type = twolevel random mixture; MODEL: %WITHIN% %Overall% u on X1; m on X1; m*; %c#1% u on X1(1); m on X1; m*; %c#2% u on X1(1); m on X1; m*;
%BETWEEN% %Overall% u on X2; m on X2; u with m; u;m; [u$1];[m]; %c#1% u on X2(2); m on X2; u(3); m; u with m; [u$1*](4); %c#2% u on X2(2); m on X2; u(3); m; u with m; [u$1*](4);