Examining author gender data

![Figure][1] GRAPH: TERRY CLARK I previously reported results of a study on author gender in Science ([science.sciencemag.org/content/355/6323/329][2]) that was based on the examination of a random sample of approximately 2600 authors for which gender was inferred by painstaking analysis of websites and similar sources. Unfortunately, this approach does not scale well to large samples necessary for many analyses. We have since initiated systematic efforts to examine gender distributions of key populations of authors and reviewers for the Science family of journals using additional data and tools and plan to use this information to guide policy development and other appropriate steps to address any gender disparities. ![Figure][1] Acceptance rates for Reports by gender of first authorsGRAPH: J. BERG/ SCIENCE Since our initial study, we have collected gender and other demographic information voluntarily from approximately 5000 individuals (authors and reviewers). To extend this dataset further, we've used first name–based gender inference software that we validated and calibrated with accurate datasets, as described in the accompanying Sciencehound post ( ). The automated tool provides inferred genders for more than 70% of the authors of both published and rejected Science papers submitted from 2010 to 2017. Based on comparisons with the individual-provided dataset, these inferred genders are more than 93% accurate on an individual basis. Furthermore, the inferred gender information can be extended to populations with gender distributions that appear to be more than 98% accurate. The fractions of male and female authors for Reports submitted to Science are nearly constant over the 8-year period examined. Twenty-five ± 1% of the first authors are female, while 18 ± 1% of the corresponding authors are female, consistent with the results from our earlier analysis. These figures reflect the weighted averages across the different fields covered by Science. Separating submissions by field reveals that 30% of first authors of submissions in the life sciences are female, compared with 16% in the physical sciences and 22% in other fields. The values for corresponding authors are 19, 12, and 20%, respectively. The gender distributions for papers that were published can be compared with those for the overall submissions pool. The acceptance rates for Reports were not significantly different for female as compared to male first authors for papers submitted in 2016 and 2017, although significant differences were observed favoring male authors from 2011 to 2015. Further work is in progress to determine if these disparities are due to gender biases in reviews, editorial decisions, or other factors such as institutional differences and preferences. Significant differences in Report acceptance rates by gender of corresponding authors were observed in 2012, 2014, and 2015. We can also examine other article types. For example, the fraction of female corresponding authors for Perspectives (driven largely by editorial invitations to Perspective authors and, indirectly, peer reviewers) grew steadily from 19 to 26% from 2010 to 2017. The Science news team's analysis of data regarding its use of women and men as sources and in quotes in its stories reveals an increase in the fraction of quoted female sources from approximately 20% to more than 30% over the course of 2018. With these data and tools in place, we are now well positioned for further analyses and actions that address gender disparities. We plan to examine the gender distribution of our peer reviewers. Social science studies indicate that women and men tend to have similar gender biases, based on their perceptions of the gender distribution of the population that they are examining. Nonetheless, ensuring that gender distributions for reviewers approximate those for authors is good practice, and peer reviewers do get access to exciting scientific results and are often invited to write Perspectives. We plan to share these and other analyses and encourage others to perform and share similar examinations. [1]: pending:yes [2]: http://science.sciencemag.org/content/355/6323/329