Network analysis of NIH grant critiques

Abstract

Network analysis has widespread applications for studying many social phenomena. Our research is focused on investigating why highly qualified women and racial/ethnic minorities tend to fare worse in peer review processes, such as for scientific grants, which limits their participation in research careers. Our prior work shows that gender and racial bias can be detected in reviewers' narrative critiques, but our work has yet to harness the power of varied learning algorithms for text analysis. To this end, we show preliminary evidence of the usefulness of network algorithms to study reviewers' written critiques of grant applications submitted to the U.S. National Institutes of Health (NIH). We construct word co-occurrence networks and show that network measures vary by applicant sex.