That's the fundamental problem. We are ruling out possible results in advance, based on arbitrary social conventions, not data.

This is fundamentally unscientific and will lead to stagnation at best, regression at worst, and progress only in the unlikely case. At some point we will have to reconcile the fact that the same mathematical and scientific techniques which we correctly apply to other sets of data produce unacceptable results any time models are conditioned on demographic data.

What then?