I developed a GMM model for longitudinal psychological distress data that includes monthly assessment outcomes for 1 year for a population of 107 people. From this data Mplus fit statistics as well as theory point to a 4 class arrangement:
Upon closer visual inspection of the classes and individual trajectory assignments however, I can see a large minority of individuals that appear to be assigned to the wrong class (i.e., trajectories that clearly decrease in distress with time but are assigned to the chronically high or even resilient classes). While the most of the individuals appear to be accurately assigned there are enough that appear to be incorrect that I am concerned that something has gone awry. Is there a generally accepted approach to dealing with this issue?
Any further reading or specific guidance you can provide are much appreciated.
This could be related to entropy or your small sample size. It could also be that the intercept is more important to the class formation than the slope. If you would like further input, send your output, plot file, and license number to firstname.lastname@example.org.