It looks like a measurement issue. The problem we want to eliminate is sexism and racism. Counting the number of times people are actually affected by a negat

It looks like a measurement issue. The problem we want to eliminate is sexism and racism. Counting the number of times people are actually affected by a negative -ism is difficult. We don't know what people are thinking unless they tell us. Bigots have an incentive to lie about their reasons and non-bigots can be subject to unconscious biases. On the other hand categorizing people by race/gender is easy. There is always some wiggle room, but there is a lot less ambiguity when answering "is this employee a black woman" compared to "was this employee hired because she is black". Once you have a set of (mostly) reliable statistics then you can start looking at trends over time, patterns by industry, etc. These numbers can't tell you anything about racism/sexism on their own, but they can help show the general direction the population studied is going and if there are any outliers who might deserve more attention.