Shige Song posted on Thursday, April 19, 2007 - 6:26 am
We have two sets of class counts: those based on the estimated model, and those based on estimated posterior probabilities. The classification of individuals, however, is based on "their most likely latent class membership". Here is part of my output: --------------------------- FINAL CLASS COUNTS ... BASED ON THE ESTIMATED MODEL
CLASSIFICATION OF INDIVIDUALS BASED ON THEIR MOST LIKELY LATENT CLASS MEMBERSHIP
Class Counts and Proportions
1 77 0.02523 2 111 0.03637 3 1554 0.50917 4 1310 0.42923 --------------------------- Why the classification is so different from the two class counts (based on estimated model and based on posterior probabilities)?
Shige Song posted on Thursday, April 19, 2007 - 6:27 am
Also, Mplus "plot3" produces sample means or estimated means plot based on either estimated model or based on posteriori probabilities (probably the latter), but the generated data use most likely membership to group individuals. In my case, I want to present the figure and I want to do some analysis using the Mplus generated data, how do I talk about the discrepancy between these two classification system?
The discrepancy is a function of entropy. If entropy is high, the discrepancy will be lower. IF entropy is low, the discrepancy will be higher.
Shige Song posted on Thursday, April 19, 2007 - 8:35 pm
I use SAVEDATA with "save=cprob" option to output predicted factor scores as well as latent class posterior probability from my growth mixture model. I have two set of factor scores. For example, I have "i" and "s" in my model to represent intercept and slope factors, I have them in the generated data set; I also have a "ci" and "cs". They are very similar but not identical. My questions are: why there are two different set of factors scores, and what are they for?