I am fitting a four class mixture model (LCA) on 4 repeated ordinal measures (3 categories per measure). My complete-case sample size is 7016 and 75 of the 81 possible patterns are present in my dataset.
My entropy results appear to be inconsistent. The global entropy measure is 0.699, however the matrix showing Class Probabilities for Modal-Class Membership by Latent Class has diagonal elements 0.903, 0.735, 0.777 and 0.835.
I would have expected the entropy to be a weighted average of the four diagonals however it isn't even within their range of values.
I have managed to recreate the diagonal elements by hand by examining the posterior probabilities for each response pattern, but using the same method for the model entropy yields a figure of 0.841.
I have searched the MPlus forum to no avail so i would appreciate some comments on how the model entropy is calculated and how one might achieve such an 'out-of-range' value.