Statistics
I'm not a professional statistician but...
There were only 48 test subjects, 22 'control' who did the test in private (4 of whom identified as female), 26 were subjected to the stress test (6 women).
Those numbers don't sound too good to start with... but there's more:
Of the 22 private tests, 13 of the bottom 14 were male. Isn't that as significant as 13 of the top 14 stress tests being male? So if stress testing favours males, then does non-stress testing favour women? Clearly not, because only stress tests favouring men gets mentioned...
Crucially they did not test the same subjects in both environments, so I suspect that the results were more related to individual ability than any inherently sexual bias. The study just didn't have enough candidates (particularly women) to be statistically significant. I suspect that the study group was also likely to be biased because it was taken from current students with a knowledge of java and algorithmic representation who volunteered to be tested.
By the way, I don't think whiteboard coding is a sensible interview test. If you're going to make an interviewee use a whiteboard then make it interactive, see how they react to suggestions, modify solutions according to additional input, etc. You know, like how programmers actually use whiteboards...
