Preference Heterogeneity in Admission to Universities and the Effects of Minority Reserves: Improving Gender Balance in STEM Programs (Job Market Paper) - This paper studies the limits of minority reserves for gender-balancing in post-secondary admissions when university programs have heterogeneous preferences. Using data from the centralized post-secondary admission system in Taiwan, we estimate preferences for both applicants and programs, and explore how the preferences relate to the gender gap in science, technology, engineering, and mathematics (STEM) fields. To account for applicants' and programs' decision timing, we introduce a modified version of the truth-telling assumptions for estimating preferences under a deferred-acceptance mechanism. On the program side, we find substantial heterogeneity across programs in gender preferences in their ranking of applicants. On the applicants' side, the largest difference between gender lies in male applicants' preference for programs where students are strong in math, and the preference is higher for those with lower math scores. Female applicants' emphasis on math is weaker and show less heterogeneity. Counterfactual simulations show that reserved seats for women in STEM programs lead to limited changes in female representation.

   The Effects of Brief Juvenile Detentions on Recidivism: Evidence from Low-Risk Youths (with Diego Amador) - Many youths accused of delinquent conduct are detained as their cases get processed by juvenile courts. Using a decade of detailed administrative data for all initial detention decisions made in one of the largest counties in the US, we find that these short-term detentions of low-risk youths lead to a sizeable increase in the likelihood of rearrest. We also find that these effects are concentrated on youths arrested for non-violent, less serious offenses and are unrelated to the actual amount of time spent in detention. To obtain our estimates, we implement the double/debiased machine learning estimator developed by Chernozhukov et al. (2018), which relies on selection on observables as the key assumption. Sensitivity tests developed by Masten et al. (2020) show that our estimates are robust to plausible levels of violations of this assumption.