The Simpson's Paradox also comes into play here.
It is perfectly possible for 1 group to be (apparently) discriminated in the bulk data, while the reverse is happening in individual data. E.g. a university showing a male bias overall, yet each department shows neutral, or even a female bias.
This makes bulk patterns particularly troublesome to work with. Men and women want different things from work. Men are disproportionately discouraged from having a work life balance, while it's far more acceptable for women to not maximise their earning potential.