Dear all,

I was estimating a conditional logit model with alternative-specific coefficients for an alternative specific variable (let's call it a; i.e. I split up the "generic" coefficient of an alternative-specific variable) and some other case-specific/ individual-specific variables (no panel or other multilevel data, just a choice model on plain old cross-sectional data!):

Code:
clogit chosen `constants'   a_j1 a_j2 a_j3  a_j4   g_j1 g_j2 g_j3, group(id)
whereby j=4 is the reference alternative and `constants' a local macro for the constants.

My main interest lies on a case-specific variable, let's call it g, and i want to get average predicted probabilities of Y=j (choice set of e.g. 4) at specified values of s.

After mlogit, I would, for instance, type:

Code:
margins, at(g(1(1)20))  predict(outcome(2))
to obtain the average using the observed values for the covariables.

How do I do this after - clogit - (with the default baseoutcome, in this example j=4, for simplicity)? From what i understood so far, the option - , predict(pu0) - should be adequate for my purposes. However, how do I "tell" Stata to treat "g_j1 g_j2 g_j3" as one variable within the margins command?


Note: I would also appreciate comments on how to do this in Stata 16 as i might get access to it soon!