Research

Conducted an econometric analysis of racial wage disparities using 2015 IPUMS ACS microdata. Estimated log wage models with demographic, education, occupational, and quadratic age controls, incorporating race–education interaction terms and restricting occupational categories to ensure comparability. Applied a twofold Oaxaca–Blinder decomposition to separate explained and unexplained components of the wage gap, with careful interpretation of structural differences and model assumptions. The workflow included percentile trimming to mitigate outliers, mean-centering of education variables, and robustness checks to assess stability of results.

Analyzed 2022 traffic stop data from Ventura County using administrative records from California’s RIPA dataset (~87,000 stops). Implemented the Veil of Darkness test, exploiting variation in officer visibility around sunset with time-of-day and date fixed effects to assess whether stop composition changes discontinuously when visibility shifts. Additionally applied Becker’s outcome (hit-rate) test using linear probability models to compare search productivity across racial groups. While search rates differed across groups, neither the visibility test nor hit-rate analysis provided evidence consistent with systematically lower suspicion thresholds for minority drivers under the assumptions of these frameworks.

Analyzed 2024 Orange County District Attorney administrative records (~69,800 case-defendants) to evaluate whether diversion outcomes differ across racial and ethnic groups after accounting for observable case severity. Estimated a linear probability model (LPM) using race indicators, felony exposure, number of charges, prior allegations, and gender as controls. Coefficients were interpreted as conditional percentage-point differences in diversion probability relative to White defendants. While observable case characteristics significantly predicted diversion outcomes, racial coefficients did not indicate evidence of disadvantage for Black defendants under the model specification. Results are interpreted as conditional associations rather than causal effects, with discussion of omitted variable bias and unobserved case factors.