I am estimating distal outcomes in a growth mixture model, and I have used both the manual 3-step approach and the automatic auxiliary (dcon) options. I understand that these procedures differ in terms of general assumptions as well as handling of missing data, however, I have gotten drastically different mean estimates for each class with these two procedures. I do not have a substantial amount of missing data, so I don't think that different treatments of missing data could create such large differences in class mean estimates. Is there another reason I may be getting such different results with these procedures?
I really appreciate your help with better understanding these differences.
It may have to do with class formation changes. You should check that the class percentages are the same between the first and last steps. This should be the case with DCON but perhaps not in your manual approach.