Conditional Logistic Regression (Discrete Choice Model) with a Latent Class Component that Allows for Covariates

[Jonathan Boyajian]

Hello,

I have standard discrete choice data and I would like to fit a conditional logit model with a latent class component to the data. The LCLOGIT package will run this model; however, it does not allow for the inclusion of predictors of the latent class variable.

Does anyone know of a Stata package or plugin that can fit a conditional logit model to discrete choice data with a latent class component that also allows for the inclusion of covariates for the latent class variable? I know I can use 2-step procedures to test covariates of the latent class variable, but I would really prefer to estimate all parameters at once in a single model. Any direction would be greatly appreciated. Thanks!

Jonathan

[Marcos Almeida]

Hello Jonathan,

Welcome to the Stata Forum.

I wonder if SEM (Structured Equation Models) - precisely the generalized SEM (gsem) - wouldn't fit your needs.

Best,

Marcos.

[wbuchanan]

I'm not sure if it'd be possible to do this in a single step. Mplus has some "wrapper" type functionality for proximal and distal covariates used with latent classes, but they all use varying types of multistep procedures (from what I remember of the manual). Essentially, you need to fit the measurement model first to, then constraints are added to prevent the covariates from changing the classification in the latent class variable and then differences in covariates across the classes are tested. They are a bit more measurement focused, so there may be other solutions from the econometrics space, but I wouldn't be too optimistic about finding a single step solution like that from my readings.

[Jonathan Boyajian]

Thanks for your replies. I will see if anyone else has specified these models in an SEM framework and, if so, give GSEM a shot. I also read that gllamm might be able to handle the problem. If not, I will settle on a 2-step solution!