For those who haven't read the linked article before, it really is fascinating.
The relevant bits are two factors:
* Is the person sending the pull request someone already known to the project, or an outsider?
* Is the gender of the pull request's author easily determinable?
It would be unsurprising to find that "outsider" pull requests get merged at a lower rate. But "outsider" pull requests from women merge at different rates depending on the second factor: when it's easy to determine the gender of the author, a pull request from an "outsider" woman is significantly less likely to be accepted than a pull request from an "outsider" woman whose gender cannot easily be determined. Also, "insider" women and women whose gender is not easily determinable have higher rates of merged PRs than their male counterparts.
There's no reason to suspect that whether someone's gender is easily determinable by a PR reviewer (usually, via username, profile picture, other clues like linked blogs or social-media profiles) has an effect on the quality of their code. So while this of course doesn't prove gender bias (we'd need telepathy for that), it does very very very strongly suggest it.
This is also in line with a lot of prior research and reporting on blind interview and auditioning practices, so shouldn't be too surprising, but for some reason people love to insist that tech is somehow different from all those other fields where hiding someone's gender changed how they were evaluated.
The relevant bits are two factors:
* Is the person sending the pull request someone already known to the project, or an outsider?
* Is the gender of the pull request's author easily determinable?
It would be unsurprising to find that "outsider" pull requests get merged at a lower rate. But "outsider" pull requests from women merge at different rates depending on the second factor: when it's easy to determine the gender of the author, a pull request from an "outsider" woman is significantly less likely to be accepted than a pull request from an "outsider" woman whose gender cannot easily be determined. Also, "insider" women and women whose gender is not easily determinable have higher rates of merged PRs than their male counterparts.
There's no reason to suspect that whether someone's gender is easily determinable by a PR reviewer (usually, via username, profile picture, other clues like linked blogs or social-media profiles) has an effect on the quality of their code. So while this of course doesn't prove gender bias (we'd need telepathy for that), it does very very very strongly suggest it.
This is also in line with a lot of prior research and reporting on blind interview and auditioning practices, so shouldn't be too surprising, but for some reason people love to insist that tech is somehow different from all those other fields where hiding someone's gender changed how they were evaluated.