

GMM with classes in addition to LGM 

Message/Author 


Hi, Not sure what to do if I wanted to create my classes, a LGM and examine the relationship between classes and the intercept and slope. Can you tell me what I am doing wrong. Thanks usev are ag; !where ad are outcome variables and eg !are variables for the classes classes = c (2); Analysis: type=mixture; starts = 50 2; model: %overall% i s a@0 b@1 c@2 d@3; c on i s; output:..... plot:..... series = e(1) f(2) g(3); 


Typically, c influences i and s, not the other way around. The c influence on i and s is such that i and s have different means across the classes. You will have this model as the default if you say only i s  etc in the overall part of the model. If in contrast you really intend to have c influence i and s, you should have observed indicators of c. And you should also specify the model so that the means of i and s do not vary across the classes  if not, you have a reciprocal interaction loop problem where c influences i and s and i and s influence c. 

Diane Chen posted on Monday, November 22, 2010  1:57 pm



I am interested in looking at how class membership for one GGMM model (2class, treated as observed/hard binned) predicts intercept/slope in LGM for another variable. Here is my input: MODEL: vi vs vq Var14t@4 Var15t@3 Var16t@2 Var17t@1 Var18t@0 Var19t@1 Var20t@2 Var21t@3 Var22t@4 Var23t@5 Var24t@6 Var25t@7; vi vs vq ON RPIclass; Results indicate that vi ON RPIclass is significant Est/SE = 2.376, p=.018. I'm not sure how exactly to interpret this finding. Any help would be appreciated. Thanks! 


If say RPIclass=1 means high trajectory vs 0 being low trajectory, this says that a high trajectory membership leads to a lower intercept growth factor than the low trajectory membership. 

Back to top 

