The ultimate impact of labor market discrimination depends not only on whether employers discriminate but also on jobseekers' strategic responses to (anticipated) discrimination. We ran three field experiments with 2,200 jobseekers to understand these strategic behaviors in the context of Rio de Janeiro's favelas. In this sample, over 80% of jobseekers overestimate anti-favela discrimination, as we measured in a new audit study. We partnered with a private firm with real job openings to estimate how jobseekers' anticipated discrimination affects job application behavior and interview performance. Interview performance is 0.13SD higher for jobseekers randomly told their interviewer would know only their name, as opposed to their name and address. In contrast, average job application rates are unaffected by (i) removing the need to declare address at the application stage and (ii) information that we did not find evidence for discrimination in our audit study. The subset of white jobseekers stands out as an exception: they apply more when they can hide their address and benefit more from hiding their address at interviews. Our findings show experimental evidence that anticipated discrimination may affect jobseekers' search, and suggest these effects may be larger for in-person interactions.

Go to paper

We study whether tweets about racial justice predict the offline behaviors of nearly 20,000 US academics. In an audit study, academics that tweet about racial justice discriminate more in favor of minority students than academics that do not tweet about racial justice. Racial justice tweets are more predictive of race-related political tweets than political contributions, suggesting that visibility increases informativeness. In contrast, the informativeness of tweets is lower during periods of high social pressure to tweet about racial justice. Finally, most graduate students mispredict informativeness, more often underestimating than overestimating, reducing the welfare benefits of social media.

How does present bias affect welfare when agents want to coordinate over time? To answer that, I analyze a dynamic coordination model under quasi-hyperbolic discounting. I document a novel mechanism through which present bias can be adaptive, i.e., it can internalize the social cost of coordinating on a new action, say going from coordinating on using Twitter to using Threads. Agents migrating from Twitter to Threads ignore that their choice imposes negative externalities on those still using Twitter. So, to achieve efficiency, regular exponential discounters should ask for a higher relative quality of Threads before adopting it. In turn, present biased agents overvalue the externalities they currently receive from Twitter since externalities from Threads can only come in the future, after others adopt it. Hence, present bias leads agents to ask for more quality before migrating to Threads, preventing paths of inefficient coordination. Furthermore, small amounts of present bias always prevent society from taking inefficient paths.