[anthonyb]

Mean SATs, not their variance. I don't think that overcomingbias link is really adding anything to the discussion.

The paper that roguecoder referenced is just pointing out that SAT scores are not a perfect predictor, and adjusting the intake based on gender is probably a good idea if you want to maximise the real effect (academic performance), rather than the predictor (SAT scores).

[yummyfajitas]

The Robin Hanson blog post points out the exact same thing. Why is a positive correction for women useful, but a negative one "not adding anything to the discussion"?

If you want to use gender as a predictor, it could be positive or negative.