"We took an average, j'accuse!" (Chronicle of Higher Education edition)

LinkedIn helpfully reported that one of my contacts was "in the news!" and pointed to this article that is mostly about Duke's Academic Council chair; it's the section on salary equity that has the person of interest.

In short, Duke admins are worried because the Chronicle of Higher Education is reporting a gender salary discrepancy that is at odds with the University's own study, and Merlise Clyde will investigate.

The problem with the Chronicle study is pretty simple to understand.  It simply took the average of male salaries and the average of female salaries and compared them, despite the fact that salaries vary greatly from discipline to discipline.  If you account for department-to-department variation in salary, as in the Duke study performed by the University's own Statistics department, there isn't a measurable difference in salary.

The reasons behind this apparent paradox are taught in the first weeks of Stats 101 at Duke**.  Perhaps the writers at the Chronicle would do well to attend.


** Statistics, or at least the 2nd edition, by Freedman, Pisani, Purves, and Adhikari that was used in some Stat 101 courses in the mid-90s, specifically mentions the famous UC Berkeley study in the 70's where, overall, about 44% of the men and 35% of the women applicants to grad school were admitted.  It looks like a clear case of gender bias until you break down the admission rate by department, and find that the admission rates by department were roughly equal, but more women were applying to departments that had lower admission rates.