A new study that used AI to analyze 2M+ contributions by ~365K developers on GitHub finds users with white-sounding names may have more success on the platform (Issie Lapowsky/Protocol)

new 2m 365k githublapowskyprotocol Software developers with white-sounding names may have more success on GitHub than developers whose…

new 2m 365k githublapowskyprotocol

Software developers with white-sounding names may have more success on GitHub than developers whose names are perceived as Black, Hispanic or Asian-Pacific Islander, according to a recently published study.

The findings, which were published earlier this year in IEEE Transactions on Software Engineering, raise important questions about the consequences of a lack of diversity on GitHub and in the open-source software community in general.

Researchers at the University of Waterloo analyzed more than 2 million contributions, or “pull requests,” made by 365,607 developers on GitHub. Using an AI tool called NamePrism that analyzes people’s names for their perceived race and ethnicity, the researchers found that being perceived as white on GitHub generally increases a developer’s odds of having their ideas accepted. Compared to developers perceived as Hispanic or Asian-Pacific Islander, it increases those odds between 6 and 10%.

“Theoretically this is the one place where there’s the possibility of a full meritocracy. You don’t see a person in open-source software. It’s unlikely you’ve met them or have an opinion of them. You know, at best, their name,” said Mei Nagappan, an assistant professor of computer science at the University of Waterloo who co-authored the study.

The fact that racial bias may still exist, even in this environment, is concerning, given the influence open-source communities like GitHub have on product development, Nagappan said. “If we don’t listen to diverse voices, then it becomes software built by and for a very homogenous population,” he said.

Not only that, but GitHub has become a sort of portfolio for software developers, meaning this bias could have an adverse affect on developers’ careers. “If you have contributions accepted to even one of the big projects, then as a newcomer, you could translate that to a successful career in a company,” Nagappan said.

GitHub did not respond to Protocol’s request for comment, and Nagappan said the goal of the research isn’t to target GitHub in particular, but to address concerns in the open-source community more broadly. Nagappan said these findings build on prior research, which has found that developers on GitHub who are perceived as women have lower acceptance rates. Acceptance rates have also been found to vary by developers’ country of origin.

He notes that the NamePrism tool his team used isn’t perfect at predicting people’s race and ethnicity. The researchers only assigned a race or ethnicity to developers when the tool had a high level of confidence. For all others, they classified the developer’s perceived race as “unknown.”

While the Waterloo researchers steered clear of attributing this phenomenon of racial bias on GitHub to any specific cause, they did find that the majority of developers contributing ideas on GitHub as well as the majority of people responding to those contributions have names that the researchers estimated were white. What’s more, they found that developers who are perceived as Black, Hispanic and Asian-Pacific Islander are more likely to have their pull requests accepted when the people responding to them are part of the same racial or ethnic group.

To remedy this potential bias, the researchers suggest that GitHub adopt a single or double-blind structure, similar to how research is assessed in the world of academia. Another suggestion would require multiple people to assess a given contribution, so that no single individual’s biases interfere.

The question of how perceptions of race impact people’s online interactions isn’t unique to GitHub. Last year, Airbnb launched a research project called Project Lighthouse that also aimed to analyze how racial discrimination manifests on the platform, including the role that people’s names play in skewing other users’ perceptions.

Related Posts