3-step mixture modeling with training... PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
 Veronica Cole posted on Thursday, April 19, 2018 - 2:57 pm
Dear Drs. Muthén,

I hope this message finds you well. I have a question about manual and automatic Modal ML in 3-step mixture models in the presence of training data, which I can't seem to get figured out.

We have a four-class solution, in which two of the classes (Classes 1 and 2) have complete training data (i.e., they are completely observed) and two of the classes (Classes 3 or 4) have no training data (i.e., they are completely latent). We are hoping to use these classes in a 3-step mixture modeling application. For our final model we will need to estimate the third step manually, so we wanted to test out a set of simpler models first.

Thus, we attempted to set up a manual latent class regression (i.e., latent classes as outcome) using the misclassification matrix provided by Mplus, and compare it to the automatic R3STEP results. I cannot seem to get the values of the regression coefficients (i.e., the logistic regression of class membership on the predictor of interest) to match up between the manual and automatic results. The class membership proportions match up, indicating that I put the logits in correctly, but the regression coefficients don't.

Thank you so much for helping me puzzle through this. I'm happy to provide any information that would be helpful!

Veronica Cole
 Tihomir Asparouhov posted on Thursday, April 19, 2018 - 5:26 pm
You can add two binary class specific perfect indicators (for the two known classes) instead of training data. If that doesn't work for you send your example to support@statmodel.com.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message