Gender plays a complicated role in the hiring of computer science tenure-track faculty members, of which on average only about 15% are women, according to a study presented today at the peer-reviewed International World Wide Web Conference in Montreal, Canada, and posted on the arXiv preprint server in February. Gender bias in hiring is not blatant, the authors found, but gender-associated differences in productivity, postdoctoral experience, and institutional prestige of degree-granting institutions—which are likely due to bias against women during the training process—largely account for the observed gender imbalance in computer science faculty hiring networks.
“This is evidence that gender doesn't stand by itself; it's baked into all these other processes,” says Brian Keegan, a research associate at the Harvard Business School who was not involved in the study.
Despite significant attention to hiring more female faculty members in science, technology, engineering, and mathematics, they remain underrepresented at the top levels of academia. But faculty hiring is a complex and sometimes idiosyncratic process, and the specific underlying reasons for the continuing disparity are unclear. “One of the difficulties in this area [of gender and faculty hiring] is figuring out causality,” says senior author Aaron Clauset, a computer scientist at University of Colorado (CU), Boulder. “Gender, productivity, and postdoctoral training are all related to each other, so trying to tease out how important … a driving factor gender is is quite difficult.”
To address this question, the authors used a hand-curated dataset of 2659 tenure-track computer science faculty members at 205 Ph.D.-granting departments across North America, which they had previously developed for a 2015 study of faculty hiring networks. The dataset included the years the faculty members were first hired as assistant professors, which ranged from 1970 to 2011; the prestige of the institutions where they got their degrees and were hired; whether they had postdoctoral experience; their publication history; and their gender. When the researchers created a model that took all these factors into account to predict the real-world hiring decisions, they found that institutional prestige and productivity alone were sufficient for accurate predictions. In other words, gender as a separate factor did not improve the model's accuracy. So, all other things being equal, gender did not seem to matter during the faculty hiring process.
But the researchers are quick to point out that all other things are not equal. Women in computer science Ph.D. programs operate in cultures that often fail to be inclusive, so when they get to the faculty hiring process, they have already been disadvantaged in their training. Using “unbiased” measures like productivity and prestige can make it look like the decisions are gender-blind, but gender has in fact played a role all through the training process and is therefore already baked in. For example, the authors found that for the assistant professors who started their positions after 2002, women were less productive than men. “The origin of this productivity gap seems unlikely to be related to inherent differences in talent or effort,” the authors write, “and may instead be related to differential access to resources and mentoring, greater rates of hostile work environments or sexual harassment, differences in self-perceptions, or other gender-correlated factors.”
This result offers a contrast to a controversial paper published last year that found that based on one-page personal narratives or CVs with similar qualifications and either a typically male or female name, reviewers were twice as likely to choose to hire the woman. The authors of that study concluded that there was no longer any discrimination against women in the faculty hiring process, but others found this approach problematic. “It treats gender like it’s in a vacuum, like there is no baggage that goes along with being a woman or a man,” says Jane Stout, director of diversity research at the Computing Research Association in Washington, D.C. Perhaps women are not being discriminated against when reviewers consider resumes, “but what have women and men had to do differently to get to this specific situation right now [of being considered for a faculty job]?"
The authors of the current study hoped to get around this objection by focusing on the outcome of the entire hiring process, not just a single step such as reviewing CVs. “If you are looking at different parts of a complex system, a system-level approach can be really beneficial in helping to put individual pieces together,” says study co-author Daniel Larremore, an applied mathematician at the Santa Fe institute. “We wanted to deconstruct [faculty hiring] … in terms of things we could measure, like publications, information about the prestige of the doctoral program, and so on.”
However, one of the things that is missing from their model is the effect of the proportion of women who are already on the faculty at a given institution, Larremore adds. Departments that already have female faculty members are likely to have a climate that is friendlier to women, he explains, and therefore attract more women. In contrast, all-male departments might have a harder time recruiting women.
The researchers also found that women and men were about equally likely to be hired at the most prestigious institutions, but fewer women than men were hired at less prestigious universities. The most prestigious ones tend to out-hire schools in the tier immediately below them because they have more hiring power, the researchers suggest. Due to the paucity of women with Ph.D.s in computer science—in 2011, just 20% of computer science Ph.D. recipients were female—efforts by the top-ranked departments to improve their gender ratios might come at the expense of institutions below them, explains lead author Samuel Way of CU Boulder. “In some ways, it's like a zero sum game, because there are so few women in computer science,” Clauset adds. “When the small number of women choose to go to one place, it means that other places don't get to hire them.”
Overall, the gender breakdowns of computer science Ph.D. recipients and new faculty hires “are very close,” Clauset says, which “suggests there isn't much slack in the system for intervention at the faculty hiring stage. … If we want to increase the number of women in computing, the data suggest we should really focus on increasing the number of women who get Ph.D.s in computer science.”
Finally, the authors forecasted when computer science might reach gender parity in faculty hiring at the assistant professor level: It’ll take approximately 60 years. “I think that’s really depressing,” Clauset says. “If we as a community are not happy with this rate, perhaps we as a community should get together and do something to change it. This is a conversation that I hope that can get started as a result of having real quantitative data.”