Women are severely marginalized in software development, especially in open source. In this article we argue that disadvantage is more due to gendered behavior than to categorical discrimination: women are at a disadvantage because of what they do, rather than because of who they are. Using data on entire careers of users from GitHub.com, we develop a measure to capture the gendered pattern of behavior: We use a random forest prediction of being female (as opposed to being male) by behavioral choices in the level of activity, specialization in programming languages, and choice of partners. We test differences in success and survival along both categorical gender and the gendered pattern of behavior. We find that 84.5% of women’s disadvantage (compared to men) in success and 34.8% of their disadvantage in survival are due to the female pattern of their behavior. Men are also disadvantaged along their interquartile range of the female pattern of their behavior, and users who don’t reveal their gender suffer an even more drastic disadvantage in survival probability. Moreover, we do not see evidence for any reduction of these inequalities in time. Our findings are robust to noise in gender recognition, and to taking into account particular programming languages, or decision tree classes of gendered behavior. Our results suggest that fighting categorical gender discrimination will have a limited impact on gender inequalities in open source software development, and that gender hiding is not a viable strategy for women.
展开▼
