Model Modification Indices PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Andy Ross posted on Thursday, May 18, 2006 - 8:17 am
Dear Prof Muthen

I am running a latent class analysis including covariates.

When requesting modification indices i only seem to get them for path types that have already been specified, and not others.

For example, when i specify a direct effect between a covariate and an indicator the output will suggest further direct effects between that particular indicator and additional covariates. However there are no suggested direct effects for other indicators. Yet when i do specify one, again others are then suggested in the output.

Is there a way around this - can i ask for all modification indices relating to all indicators regardless of whether a particular path has been specified?

With many thanks

Andy
 Linda K. Muthen posted on Thursday, May 18, 2006 - 9:46 am
The reason this happens is that the matrix for direct effects is not opened if it is not part of the MODEL command. You can get modification indices by saying

y1-y4 ON x1-x2@0;

By fixing all of the direct effects to zero, the matrix will be opened and you will get modification indices.
 Alexandre Morin posted on Wednesday, February 06, 2008 - 7:22 am
Hi,

I did finally converge on a final (?) model in a latent profile analysis. I tried various parametrisation of indicators variances within classes. Finally, I retained a 5 classes model with the default parametrisation of between class equality of variances. I would like to know what you would recommend looking at in the output to see whether it would be adviseable to free the variance of one (or some) indicator(s) within some of the classes. I know that MI are not recommended in mixture models. I could look at residuals, but what would be a "big" residual ?

Thank you
 Bengt O. Muthen posted on Wednesday, February 06, 2008 - 6:27 pm
You could take the solution with class-invariant variances and classify people by most likely class - and then look at each variable to see how much sample variance you have in each class and therefore see if one or more classes stand out as having much more/less variance than other classes. Then modify the model accordingly to see how BIC changes.
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: