There are other reasons why people would be paid less, other than race and gender, but when aggregated might appear to show bias. Imagine that there is a compa

There are other reasons why people would be paid less, other than race and gender, but when aggregated might appear to show bias. Imagine that there is a company that is entirely fair and pays people solely based on their seniority at that company. Currently, that company only has employees of race/gender combination A, with various tenures. If a new employee of race/gender combination B joins the company, the employer will look as if they're underpaying race/gender combination B, even though race/gender combination A and B would be paid the same, given the same seniority. Because the aggregation discards useful information that would explain the disparity, it can induce in error. There are also other effects such as Simpson's paradox [0] that show disparity when data is aggregated, but has a different effect direction when analyzed in more detail. Reporting this information is an improvement over no reporting, but I'm not particularly optimistic that it will always be looked at with the proper statistical background to interpret it correctly. [0] https://en.wikipedia.org/wiki/Simpson's_paradox https://en.wikipedia.org/wiki/Simpson's_paradox