Using aggregate measures like that rolls up the wrong numbers into a final metric. It tells me that the men who were underpaid had a lesser gap between what th

Using aggregate measures like that rolls up the wrong numbers into a final metric. It tells me that the men who were underpaid had a lesser gap between what they were being paid and what they should be paid, or that the men who were overpaid were overpaid to a greater extent. They needed to compare the number of men who were underpaid as a proportion of all men to the number of women who were underpaid as a proportion of all women. Using bum-pulled numbers, your company has 1000 employees, 667 are men, and 333 women, and they all do pretty much the same job. Your payroll is $100M, and 70% of it goes to the male employees. The mean pay for men is $105k, and the mean pay for women is $90k. Mean pay is $100k--let's say this is fair pay for the work. Overall, the men are overpaid, and the women underpaid. But the mean is a fool's game; the median is more telling. Now say 445 men get $90k, and 222 get $135k. 2/3 of men are underpaid. Now let's say 222 women get $101k, and 111 get $68k. 1/3 of women are underpaid. On a per-person basis, a man working for the company is twice as likely to be underpaid as a woman, but the amount by which they would be underpaid is much less. Going by the aggregate numbers, one could say "men at Company X are overpaid, compared to the women" and then any given underpaid man might not even realize they are being underpaid, thanks to the misleading statistic. When one of the $68k-paid women complains (naturally, as the worst-off of the bunch), and the company analyzes its own payroll at a more detailed level, it realizes that men are more likely to be underpaid than women (and also that increasing its payroll by 8% would make no-one underpaid).