Hello!
I'm estimating a heckman selection model "by hand" i.e. not with the heckman command.
Reasons for this:
- I need fixed effects in the first stage and while i.xxx is working, fixed effects in probit models are econometrically a bit messy
- there are people that think that the linear probability model is a good idea
I'd be happy to not have to debate this!
Now I want to test the two criteria for my selection variable(s): being strong in stage1 and being uninformative in stage2.
I'm a bit unsure which tests to use for this. Online investigations increase my confusion because people keep calling selection variables instrumental variables and many well-published papers don't report formal tests on the first stage.
Relevance:
- LR Test does not work with clustering
- ranktest: works but does not allow indicator variables that I'd like to put in my first stage and typically there's a good reason when stata doesn't allow things. But the F-statistic is nice.
Exogeneity:
- can I formally test this? Do I need more than one pure selection variable for testing it?
Any advice is happily taken!
Many thanks
I'm estimating a heckman selection model "by hand" i.e. not with the heckman command.
Reasons for this:
- I need fixed effects in the first stage and while i.xxx is working, fixed effects in probit models are econometrically a bit messy
- there are people that think that the linear probability model is a good idea
I'd be happy to not have to debate this!
Now I want to test the two criteria for my selection variable(s): being strong in stage1 and being uninformative in stage2.
I'm a bit unsure which tests to use for this. Online investigations increase my confusion because people keep calling selection variables instrumental variables and many well-published papers don't report formal tests on the first stage.
Relevance:
- LR Test does not work with clustering
- ranktest: works but does not allow indicator variables that I'd like to put in my first stage and typically there's a good reason when stata doesn't allow things. But the F-statistic is nice.
Exogeneity:
- can I formally test this? Do I need more than one pure selection variable for testing it?
Any advice is happily taken!
Many thanks
