This should handle all the missing data issues when going through patsy. I know this is mentioned in #1877. Probably elsewhere, but I have to run.
TST: Test GEE with missing.
STY: Remove docstrings from tests
TST: Uncomment and fix original missing data tests
ENH: Handle missing for extra data with formulas
Coverage increased (+0.06%) when pulling d5a971f on jseabold:fix-1877 into f65e8c0 on statsmodels:master.
I'm not really a fan of making these fundamental changes just before a release with no time to gain experience with how well it works.
But going through large part of the changes, I don't see any problem. But I also don't understand what all the implications are.
ENH: Workaround patsy NA handling. Closes #805.
TST: Regression test for ols formula and NA category
TST: Test that raise on NA with formula is possible
This is fixing a priority high bug release blocker. What else are we supposed to do?
ENH: Allow raise on missing with formula
DOC: Add NA-handling and patsy to release notes.
I'm going to go ahead and merge this shortly. Hopefully, any issues that the tests don't catch will come out in a release candidate or we can do a micro release as needed. In any event I think it's an improvement over the status quo of definitely doesn't work.