In this article, I describe the lclogit2 command, an enhanced version of lclogit (Pacifico and Yoo, 2013, Stata Journal 13: 625–639). Like its predecessor, lclogit2 uses the expectation-maximization algorithm to fit latent class conditional logit (LCL) models. But it executes the expectation-maximization algorithm’s core algebraic operations in Mata, so it runs considerably faster as a result. It also allows linear constraints on parameters to be imposed more conveniently and flexibly. It comes with the parallel command lclogitml2, a new stand-alone command that uses gradient-based algorithms to fit LCL models. Both lclogit2 and lclogitml2 are supported by a new postestimation command, lclogitwtp2, that evaluates willingness-to-pay measures implied by fitted LCL models.
Lclogit: A Stata Command for Fitting Latent-Class Conditional Logit Models via the Expectation-Maximization Algorithm
