Bayes & correlation between latent an... PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Joseph E. Glass posted on Thursday, May 10, 2012 - 5:32 pm
Dear Mplus team,

I am attempting Bayes H0 imputation of a longitudinal CFA model involving latent (categorical indicators) and observed variables (also categorical) at three time points. I am using type=complex to account for clustering. For the latent factor indicators, I am imposing cross-time constrains on loadings and thresholds.

When I specify correlations between one of the latent variables and an observed categorical variable, the analysis will not run unless I use algorithm=gibbs(rw). However, the model seems to be more susceptible to convergence problems with gibbs(rw), and it is quite slow.

Do you have suggestions for diagnosing convergence problems with the gibbs(rw) algorithm? I have increased iterations without success. There are factor indicators with low endorsement proportions.

Thank you,
Joe
 Linda K. Muthen posted on Thursday, May 10, 2012 - 7:52 pm
TYPE=COMPLEX is not available with Bayes.

Why do you correlate the latent variable and the observed categorical variable.
 Joseph E. Glass posted on Thursday, May 10, 2012 - 10:15 pm
Thank you for your response Linda. I assumed that type=complex was functional because the analysis proceeded and the cluster variable was reported in the analysis summary in the output.

I am correlating the latent variable and the observed categorical variables to represent within-time correlations in a panel model. The latent and observed variables reflect separate constructs measured at three time points. So there will be autoregressive paths for each construct across time, predictive regressions among them across time, and within-time correlations among them. Is this approach allowed in Mplus?
 Linda K. Muthen posted on Friday, May 11, 2012 - 12:36 am
We don't have COMPLEX with Bayes. We have TWOLEVEL. I think if you check your output you will see a message saying the estimator was changed to the default estimator.

Please send your output and license number to support@statmodel.com so we can see your MODEL command.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: