Does cmclogit automatically account for group fixed effects?

[hankyul yoo]

Dear friends,

I am trying to analyze my discrete choice experiment data with the cmclogit command. My data has around 400 respondents who chose from 7 choice sets. There are 4 alternatives in each choice set.

My quesiton is, if I use cmclogit, would Stata automatically assume I want to control for fixed effects within a respondent?
I read that cmclogit basically is the same as clogit, with a difference in data format.
And clogit is said to account for fixed effects for each group in the data. So I am curious if cmclogit would account for respondent fixed effect by default.


Best,
Han

[Hong Il Yoo]

hankyul yoo: No, -cmclogit- doesn't account for respondent fixed effects. -clogit- is an interesting command because you can use it to estimate two totally different types of models: McFadden's conditional logit model (-cmclogit-) and fixed effects logit model (-xtlogit, fe-). Even though these models happen to have algebraically similar likelihood functions, allowing one to use -clogit- to estimate both, it remains the case that they are very different models.

Most DCE studies use mixed logit models (see, for example, Stata's -cmxtmixlogit- command as well as community-contributed packages -mixlogit- and -lclogit2-) to account for unobserved heterogeneity between respondents. Note that mixed logit models are random parameter models, not fixed effects models.