Usually by using the most generous estimates that come from looking at the statistical differences between stats like actual reports and results from anonymous studies, that's just 2 examples, I'm sure staticians use way more data, and then turn that into "for every reported x, there's y unreported" then you just apply the pre-algebra and compare the adjusted number to which ever stat for women (the unfair answer here is then not adjusting for unreported in women, which would probably be a lower number than men but still)