Message/Author 


I am working on a GMM where I would like to include timevarying moderators. Has anyone attempted this? If so, what does your syntax look like? I keep running into all kinds of errors and can't seem to find a lot our there on this issue. 


Do you mean a model like 6.13 where the a variables are then regressed on another set of timevarying variables? 

bmuthen posted on Thursday, December 22, 2005  2:51 pm



I think Linda meant ex 6.12. The extension to having the timevarying covariates be moderators should be possible. 


I did mean 6.12. 


Dear Linda & Bengt: I have a 2class solution for a ZIP mixture GMM model with random intercept for drugs (mrj) with timevarying covariates (pst). The overall model is i s mrj15@2 mrj16@1 mrj17@0 mrj18@1 mrj19@2 mrj20@3; s@0; mrj15 ON pst15; mrj16 ON pst16; mrj17 ON pst17; mrj18 ON pst18; mrj19 ON pst19; mrj20 ON pst20; The estimates for the mean intercept and slope fro each class are: class1 i*.505 s*0.058; class2 i*.407 s*0.041. The estimates of the regression part of mrj15 ON pst15,…., mrj20 ON pst20 are .94, .48, .66, .89, .95, and .98, respectively. I have two questions: 1) How can I plot (by hand) these two classes and taking into account the estimates from the timevarying covariates, or the PLOT3 option can do it for me? 2) Probably this is related to the first one. How can I put in specific values for the timevarying covariate and estimate the outcome from this model. Can Mplus do it? Thanks so much. Delfino 


This is a further explanation from the former question. The data set that I have is longitudinal with a cohort sequential design with 7 time points, reflecting subject’s ages from 15 to 20 years old. I have self reported substance use measures from subjects as outcome so that I believe the zero inflated passion (ZIP) is best option for this analysis. We are looking for mixtures, first by fixing intercepts and slopes to zero (a second step will be taking this solution and continue with intercepts and slopes random) , with time varying covariates (TVC). I was able to run this model but when I request TECH7 option I’ve get all classes with zero weighted means in all classes what is the program doing? Would you please advice. 


I don't think Mplus provides this plot because of the combination of numerical integration and x's influencing the u's. Unfortunately, it is tricky to do this yourself because with noncontinuous outcomes and random effects, you have to use numerical integration to get the estimated outcome probabilities. We have this on our list to add for future plot extensions. 


Dear Linda and Bengt, I have a question about GMM. I want to add a timevarying factor, because the subjects were measured at different times in their life. I also want to abstract classes from the data. Is this possible? Because when I try to put timevarying variables (the different time points) in the model, Mplus says I must say TYPE = RANDOM. And for the mixture analyses with classes i must say TYPE = MIXTURE. Is this not possible? Thanks in advance, Nanda 


Say TYPE=MIXTURE RANDOM; See the table on pages 531 and 532 of the Version 6 user's guide to see how the TYPE settings can be used together. 

Youngoh Jo posted on Monday, December 12, 2011  6:27 pm



I just wonder whether I can use GMM with timevarying covariates measured in different time points repeatly. If possible, could you help me get resources for the technique or command syntax? 


Extend Example 6.12 to mixture. 

Youngoh Jo posted on Thursday, May 17, 2012  6:38 pm



Dear Muthen, When I interpret the results of GMM with timevarying and timeconstant variables, what do I need to use as oveerall model fit index? Thanks, 


There is no overall model fit statistic when means, variances, and covariances are not sufficient statistics for model estimation. You can compare nested models using 2 times the loglikelihood difference which is distributed as chisquare. 

Back to top 