When performance cannot be measured precisely, evaluators often use additional information to form more accurate assessments. This paper studies how evaluators incorporate evaluatees' past performance into subjective evaluations. Using pitch-level data from Major League Baseball (MLB), I examine the extent to which umpires' ball-strike decisions depend on batters’ prior plate discipline, measured by their history of swing and take decisions. I find that umpires are more likely to make calls that favor batters with higher prior plate discipline and that their calling accuracy is modestly higher for these batters. The influence of prior plate discipline becomes stronger both as an umpire accumulates more observations of a given batter and as the perceptual signal about pitch location becomes noisier. I also show that reliance on past performance information varies with asymmetries in the costs of decision errors. Estimates from an empirical Bayesian model indicate that plate discipline effects reflect not only rational updating but also systematic bias relative to a Bayesian benchmark. I provide further evidence that players adjust their strategies in response to umpires’ decisions that draw on prior plate discipline. Taken together, these findings highlight the importance of using information about past performance in a context-dependent way to improve the efficiency of evaluations and limit the perpetuation of inequality rooted in initial performance gaps.

Low take-up of benefits among eligible individuals plagues public programs. This paper examines the impact of laws mandating improved information sharing on benefit recipiency. We study employer notice laws, which require employers to provide employees with information about unemployment insurance at the time of separation. Using cross-sectional and panel data, we find that individuals residing in states with notice laws have about 9 percentage point higher rates of applying for and receiving unemployment insurance benefits, and are 6 to 10 percentage points less likely to cite lack of information about eligibility or the application process as reasons for not applying. By leveraging variation in the timing of notice law adoption—driven by policy changes in Massachusetts in the 1990s and nationwide during the Covid-19 period—we find that adoption of notice laws increased recipiency rates by similar magnitudes.

This paper investigates whether matching female students with female homeroom teachers narrows the gender gap in STEM track choice. By leveraging the quasi-random assignment of teachers and students to classes in Korean high schools, I find that female students assigned to female homeroom teachers are about 13 percent more likely than male students to report an intention to major in STEM in college. By contrast, I find no evidence that male students benefit from being matched with male homeroom teachers. The effect of female teacher-student gender matching is stronger in single-sex classrooms, among students with higher prior math achievement, and among those from low-income households. These patterns suggest that limited information about STEM is an important driver of female students' lower propensity to pursue STEM majors. Consistent with this mechanism, female students matched with a same-gender homeroom teacher report college counseling as more helpful, and this effect is stronger when gender-matched homeroom teachers specialize in STEM-related subjects.

This study examines how the adoption of artificial intelligence (AI) shapes human capital investment decisions, with a particular focus on educational attainment and adjustments in college major choice. Using data from the 2021 Annual Business Survey (ABS), I construct industry-level AI adoption rates in the United States as the share of firms in each industry that report adopting AI. Exploiting a shift-share design and instrumenting U.S. exposure to AI with industry-level exposure to AI in the European Union, I find preliminary evidence that greater AI adoption increases college enrollment: a one standard deviation increase in AI adoption raises the share of students enrolling in college by 4.3 percent relative to the sample mean. I further examine how exposure to AI technologies affects students' major choices, how these effects vary across sectors, and how they reshape the occupational and skill distribution.