That has been proven false, negligible or within noise over and over.

The version stubbornly pushed is simply that median_salary(all women) != median_salary(all men).