EFried posted on Tuesday, April 17, 2012 - 6:58 am
Dear Dr. Muthén,
I would like to use MPLUS to validate/enhance the multivariate multilevel fixed-effects models I was running in R before because I couldn't solve some of the issues (e.g. covariance matrix, correlated errors over time, autoregressive effects etc.).
I have 5 ordinal dependent variables (questionnaire items), 5 measurement points, time-invariant and time-varying covariates. There are no 'groups' within the sample, but heterogeneity of slopes and intercepts between subjects is very high.
A very simplified version of the current model is: y ~ y_dummy + time + x + ( time | User ID )
(y_dummy coding the 5 different dependent variables; they have the same metric; the datafile currently has time*y_dummy=25 lines per person).
My question is what your recommendation would be in terms of how to do this in MPLUS.
I have not found multivariate multilevel SEM examples in your videos or papers linked to on the website, neither multivariate multilevel GMMs (I read the Khoo&Muthén 2000 paper, but that doesn't help me with the implementation of the models in MPLUS).
Would you be so kind and tell me ... 1) what model you would recommend and 2) where I can read up on that?
You can extend the parallel growth model found in Example 6.13. Parallel process growth modeling is treated in the Topic 3 or 4 course handouts and videos on the website. There may be references there.
EFried posted on Wednesday, April 18, 2012 - 5:26 am
Linda, thank you. Do you happen to know of any parallel process growth model that was published with more than 3 response variables? I am very skeptical whether models with 5 outcome variables would converge.
I don't know of any papers like this. I can't see any reason convergence would be a problem if the model and data are a match. I would start by finding a well-fitting model for each process before I put them together.