I am trying to estimate the dropout of a social program as a discrete time duration model. I've already performed preliminary cloglog estimations assuming multiple functional forms for the hazard function. However, it is possible that there might be some unobserved heterogeneity in particular for some groups of individuals (say municipalities).
I've seen that a shared frailty model could deal with this but I'm not sure how to run it on stata for a discrete model. On the other hand, I thought of including a dummy for each group, but in that case I would find myself dealing with a very large dataset that stata wouldn't be able to handle.