Amelie Schiprowski

Diversity goals are typically defined at the level of an organization but implemented through decentralized decisions by individual evaluators observing only a small subset of the candidate pool. In this paper, we analyze how organizational diversity concerns interact with the small-sample nature of candidate selection. We use data on more than 40,000 admission interviews of a merit-based scholarship program, where each evaluator assesses a randomly assigned subset of candidates. Conditional on own ability, a candidate's selection probability is substantially reduced by higher-ability candidates assigned to the same evaluator. This effect is consistently twice as large within than across diversity-relevant candidate characteristics including gender, migration background, and socioeconomic status. To structure the analysis of potential mechanisms, we develop a theoretical framework showing that these effects can arise both from a group-specific learning mechanism and from the application of soft quotas within small candidate samples. Additional  analyses indicate that learning alone cannot fully account for the empirical patterns, pointing to a role for "small-sample quotas".