What Do Names Reveal? Impacts of Blind Evaluations on Composition and Quality

Concealing candidate identities during evaluations, or "blinding", is often proposed as a tool for combatting discrimination. I study how blinding impacts candidate selection and quality, and the forms of discrimination driving these effects. I conduct a natural field experiment at an academic conference, running each submitted paper through both blind and non-blind review. Four years after the experiment, I collect proxy measures of paper quality—citations and publication statuses—for each paper and link it to the experimental data. I find that blinding significantly reduces scores for traditionally high-scoring groups, and consequently alters the composition of applicants who are accepted to the conference. Despite these compositional changes, blinding does not worsen the conference's ability to select high-quality papers. I develop a model of evaluator discrimination that allows me to rationalize these effects and decompose non-blind disparities into two distinct forms of discrimination: accurate statistical discrimination and bias.