It could be well into the 21st century before female computer scientists annually publish as many research articles as their male counterparts, an analysis published today concludes. If current trends in publishing continue, women in biomedical research are likely to reach parity sooner, possibly by 2050.
The study, which appears on the preprint service arXiv, used a large data set and statistical methods to estimate the portion of papers published by women in those fields, yielding a measure of progress in efforts to eliminate historical patterns of gender inequality. "Although gender balance is improving, progress is slower than we had hoped," write Oren Etzioni and co-authors at the Allen Institute for Artificial Intelligence in Seattle, Washington.
Using a tool called Semantic Scholar, developed by the institute, the researchers examined nearly 3 million journal and conference papers in computer science published between 1970 and 2018. They also analyzed more than 11 million biomedical papers that appeared during that period in the 1000 most-cited journals in the Medline database maintained by the U.S. National Institutes of Health in Bethesda, Maryland.
Determining the gender of authors required estimation because some first names, such as "Taylor," are used by people of all genders. The researchers ran the first name of each author through an online database, Gender API, which predicts the probability that a first name belongs to a man or woman based on known associations between first names and gender in various countries as shown by government data and social media profiles. The researchers applied those probabilities to calculate the share of all papers published each year by women. The analysis counted all authors equally, regardless of what order each was listed.
The study's conclusion—that publishing parity in computer science will be reached only around the year 2137, a few generations from now, and perhaps even later—is an extrapolation from past growth rates.
"We hope that these findings will motivate others in the field to … consider ways to improve the status quo," Etzioni and his colleagues write in their study.
They also examined the extent of cross-gender collaborations in authorship of computer science articles, and found reason for pessimism: These collaborations are not increasing as quickly as the data indicate they could, given the growing number of women in the field, they said. "Although both men and women are more likely to collaborate with authors of their own gender, the degree of same-gender preference is declining among female authors but increasing among male authors," the study found.
The findings are consistent with those reached by a similar study published in 2018 in PLOS ONE, which examined many more fields of science, 115 in all. In the large majority—87—women comprised significantly fewer than 45% of authors, according to the analysis by Luke Holman and colleagues at the University of Melbourne in Australia.
They found that changes in author gender ratios tended to be slowest in disciplines with large numbers of men, among them computer science and physics. A possible explanation, they said, was that such fields have biases that affect the relative publication rates of men and women.