A Machine-Learning Investigation into Publication Bias in Author Ethnicity and Gender

Evidence and reports document that unprofessional peer reviews disproportionately harm underrepresented groups in STEM. Unfortunately, these reviews are not publicly available and examining such reports would take full-time study by many people over a long time period to discern the impact in the various subfields of physics. Lattice QCD is at the intersection of theoretical physics and computer science, two fields for which diversity in ethnicity and gender are in poor shape. Analysis of publications in this field could provide indirect evidence of any hidden issues. Toward this end, we analyze papers that are classified as primary hep-lat, studying whether there is any race or gender bias in the journal-publication process. We implement machine learning to predict the race and gender of authors based on their names and look for measurable differences between publication outcomes based on author classification. We would like to invite discussion of how journals can make improvements in their editorial process and how institutions or grant offices should account for publication differences in gender and race.