Swap viewpoint groups to make zero-variance tests against Stata more strict

[davidskalinder]

@sinanpl this one flips the viewpoint group in the baseline-adjustment test with zero-variance levels. The point of doing that is that the old viewpoint group was the one that had a bunch of zeros in the EXs, so the coefficient estimates weren't really being tested. This way around, there should be values for all the final estimates, so the test should be more strict.

This test does depend on how the viewpoint group is set, so it might need a tweak depending on what the viewpoint-choosing strategy turns out to be.

[davidskalinder]

Just added f67c8cd, which converts foreign_born to a logical instead of a factor. This makes the test stricter still since it now tests that everything correctly handles both logicals and factors.

(Note that the original foreign_born is an integer, and so will be treated by all the models as a numeric rather than as a categorical variable. Obviously that would be a strange way to run the model, but that behavior is consistent with the way nearly all regression software works, so I don't think we should change it.)

[davidskalinder]

Ah note that #61 includes this one, since I wanted to make sure that the changes there were tested against this stricter version.