[Exofunctor]

This approach may work (I say "may" because most of the time it's not actually backed up by hard evidence free of confounding variables), but it's also lazy; there are vastly more accurate predictors of a priori success rate than race or gender. It's especially lazy for organizations that have access to those more accurate predictors, like universities, to use race and gender instead.

More accurate predictors include parental education, family wealth, social connections, etc. Mostly what people would describe as social class.

[burkaman]

You don't have to pick just one though, right? Gender clearly has an effect even if it's less predictive than other factors, so why not try to account for every factor you know of?