| I'll bite. It can happen if the target diversity ratio is different from that of the candidate pool. For example, trying to have software engineer distribution that matches that of the general population, or even just college graduates. Let's assume men and women are equal in the distribution of their ability to perform tasks required as a software engineer, i.e. same mean, same standard deviation, etc. Let's also assume the female candidate pool targeted by affirmative action is smaller than that of male candidate pool, i.e. more male applicants than female, hence underrepresented relative to general population. Let's say the hiring bar is such that top 10% from either gender meets the bar, since the distributions are the same. Now, if male pool has 1000 applicants, and female pool has 400 applicants, this would mean there are 100 qualified male candidates, and 40 qualified female candidates. If the company wants 60/40 representation, with the understanding that 50/50 desirable but not probable, as long as the company only needs 100 hires, the company can hit the target without lowering the bar (60 male + 40 female). However, if the company wants to hire more employees and still maintain diversity target, it would be necessary to hire female candidates that are not within the top 10%. Again, this is assuming the target diversity is _not_ the same as the candidate pool, but rather some other demographics that the company is trying to match (like local population demographics). Since Google gets so many qualified applicants (and even more applicants in general), most engineers are probably overqualified anyway regardless of gender, so unqualified female engineer is most likely not issue at all. However, more qualified male engineer would be rejected (in the case of 100 hires, 40 male engineers would be rejected vs none for female engineers). And this is probably where the reduced false negative rate for applicants that fall under affirmative action umbrella comes from in Damore's memo. |