Theoretical benefits recommend segmenting choice models via latent class MNL rather than running HB to get respondent-level utilities and then clustering on those utilities. Many practitioners are either unaware of the recommendation to use LC-MNL or they think HB-then-cluster will work just as well. Testing three data sets wherein we know segment memberships, we want to determine whether the two approaches perform at parity or whether LC-MNL outperforms HB-then-cluster, in terms of (a) identifying the correct number of segments and (b) assigning the correct respondents to the correct segments.
