Thus the "assuming equally qualified candidates" in my original post.

That can be measured (imperfectly, but well enough), by filtering only for qualified candidates and then comparing the rate at which both A and B are hired and the rate at which they appear in the filtered list. E: This of course requires the filtering to be done _only_ with knowledge of a candidates skills/accomplishments, and association with demographics (including name, location, etc.) to be done only after the sorting.