01-20-2017 04:58 PM
I am looking for a little help in implementing Vermunt (2010)'s 3 step approach to latent class modeling with covariates. My model is too complex to incorporate all my covariates in a one step approach so I need to take a "classify-analyze" approach however I do not want to underestimate the effect of the covariates by not correcting for the classification error.
Step 1 - fit LCA model
Step 2 - assign classes based on posterior probabilities and calculate classification error
Step 3 - fit LCA model with covariates using the assigned class membership as the (only) indicator and treat calculated classification errors as known error probabilities.
Does anyone know how to define fixed-value constraints on the model parameters in PROC LCA? The article indicates that I should be treating the calculated classification error (obtained in the second step) as known in the third step.
Thanks in advance!