Hello,
I am currently running a latent class analysis for my MaxDiff survey. The results of the LCA point to a 2-class solution, e.g., indicated by an "elbow" in the curves of the model-fit indices (AIC, CAIC, BIC, ABIC).
Prior to the LCA, I have also run a HB estimation which I use to specify part-worth utilities for my sample - I would use the mean zero-centered raw utilities here to enable a subsequent analysis of utility scores for different segments based on HB estimates.
Now I am wondering which utility scores I should use for calculating average scores for the two latent classes. I can think of the following approaches:
1. Exporting the segment memberships of the LCA to the data set including the utility scores derived from the HB estimation. Then, calculating the mean zero-centered raw utilities for both latent classes based on the HB utility scores.
2. Exporting the zero-centered (logit-scaled) raw scores at an individual level derived from the LCA as well as segment memberships. Then, calculating the mean zero-centered raw utilities for both latent classes based on these LCA raw utility scores.
3. Re-run HB estimation with segment memberships as co-variates.
Could you please help me to figure out which approach is best suited or if any of the described approaches might not be correct / less recommended.
Thank you and best regards