Research

Matching algorithms, such as deferred acceptance, are commonly used throughout the world in important markets because of their proposed incentive compatibility. Through two new re-analyses of existing data I find that while these algorithms are theoretically 'strategy-proof,' they are not 'gender-proof'. Many women do not apply to their most preferred jobs or colleges, leading to under-representation at top programs. I document this gender difference in misreporting using data from Chinese high school students and U.S. medical students. In both settings, female applicants are more likely than male applicants to rank less prestigious options at the top of their preference lists. In an experiment with 1,488 subjects, I aim to disentangle two leading explanations for this gap: confusion and social concerns. Across two treatments I find no evidence that confusion drives the gender difference. Participants deciding on behalf of others are more likely to apply to the most preferred options,  suggesting something other than confusion drives the deficit, while the intervention  directly aimed at alleviating confusion does not improve applications for women. However, two interventions designed to increase the social appropriateness of applying to a prestigious option—encouragement and a minimal affirmative action policy—both lead to meaningful improvements. These findings suggest that market design should incorporate the molding of social norms to increase the representation of female applicants in top programs.

Men and women often make very different economic decisions. These gender gaps often disappear, however, when they are choosing for other men and women. What if a decision-maker were to choose for an as-similar-as-possible other before choosing for themselves? Would the previous decision -- perhaps less prone to errors in confidence or to social and visceral forces – provide clarity on how they would make the decision free of such biases? Would this help close gender gaps? We test this mechanism across three contexts – whether to compete, to do non-promotable work, and to contribute ideas -- collecting data to pinpoint mechanism.