Can you use SAVE = cprobs in an LTA PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
 Dylan John posted on Friday, July 28, 2017 - 9:29 am
Is it possible to assign observations to their most likely "transition group" from timepoint 1 to timepoint 2?

To further clarify, I was able to use save=cprobs to assign participants to their most likely latent class in an LCA. Can I use the same syntax to assign participants to their most likely transition from timepoint 1 to timepoint 2?
 Bengt O. Muthen posted on Friday, July 28, 2017 - 5:46 pm
Yes, but this is not done automatically. Say that you have 2 times points and 2 classes. This gives you a 2 x 2 table of posterior probabilities for a subject. From that you can compute the transition probabilities (they are conditional probabilities) and then see the most likely transition.
 Dylan John posted on Friday, July 28, 2017 - 5:57 pm
Ok, so I will get the transition probability that the case is in classA, classB...class n at timepoint 2, based on time 1. Then I assume I will assign to a transition group based on whichever probability is the highest?

And as I understand it,save=cprobs will give me the probability for each of the possible classes they could transition into?
 Bengt O. Muthen posted on Saturday, July 29, 2017 - 1:42 pm
Yes on both.
 Dylan John posted on Saturday, July 29, 2017 - 2:13 pm
Many thanks!
 Dylan John posted on Tuesday, August 22, 2017 - 4:44 am
If I used your method where the logits are added into the LTA to account for measurement error, will the transition group variable in the save=cprobs outfile take this into account? Can I export this file to SAS and still have the measurement error accounted for?
 Bengt O. Muthen posted on Tuesday, August 22, 2017 - 6:06 pm
 Jon Heron posted on Wednesday, August 23, 2017 - 11:34 pm
Just to clarify (as I'm currently in email contact with Dylan)

The model in SAS would still need some way of accounting for the uncertainty in class assignment. Any model that treats the modal-assignment as a perfect manifest variable will lead to bias unless entropy is close to 1.0.

I'll wager that if you import modal-assignments into Mplus and run a model using the bias-adjustment logit constraints before exporting the modal-classes again then import and export classes would match (barring any changes induced if you have partially missing respondents).
 Dylan John posted on Thursday, August 31, 2017 - 9:38 am
Clarification here would be appreciated
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