[strken]

It seems more like a faulty null hypothesis. Instead of assuming zero correlation between ability and pay, some people start from the assumption of perfect correlation.

With regards to ability, we seem to intuit the existence of an unmeasurable "ability factor" that underlies real, measurable metrics of performance. I have no idea how Google measure this kind of thing or any statistics qualifications, but a very dumb first attempt might be to get all the engineers to recommend five top co-workers, break the recommendations up into male/female groups and discard cross-group recommendations, then look at the attributes of the most recommended co-workers in each group.