Hongbo Wang posted on Wednesday, July 16, 2008 - 5:46 pm
Dear Dr. Muthen: I now have results, more precisely posterior distributions of parameters, from an revised version of MIMIC model. The model assumes three factors, each of which follows a distributional assumption of having two normal mixture elements. Outcomes such as educational attainment (multi-nominal), drug use (probit), as well as depression scale (continuous) are determined by these three factors, conditional on covariates. The measurement system consists exclusively of linear equations for observed measures.
I want to use such results, together with the original data, to do simulation. I'm interested in how moving people over the distributions of factors will change their likelihood of, for example, going to college or using drugs. In terms of simulation itself, it involves first making random draws from the distribution of parameter sets (2 mean, 2 precisions, and 2 probabilities) for each factor, which is followed by making a random draw from each factor distribution that is formed by the drawn parameter set. Then I can compute predicted values for each individual randomly drawn from my sample. Is this doable in Mplus. Do you have any readily available examples that I can follow. I'm not too experienced with Mplus though.
Yes, you can follow examples in the Mplus User's Guide, chapter 11, which includes examples with mixtures. To have access to the data for each person, you can use the "external Monte Carlo" feature.
Hongbo Wang posted on Thursday, July 17, 2008 - 9:13 am
Dear Dr. Muthen: Thanks a lot for your response. To clarify, I've already get all the estimates (distributions of parameters). They're saved in text files. I have data saved in a separate ASCII file. Is it stil doable? Also, can Mplus fit models with correlated factors?