A growing literature uses random assignment of cases to judges to examine criminal sentencing. To extend this line of work, we directly examine how judicial “harshness” varies with the seriousness of criminal conviction. Using a model that respects the mix of cases and the noise produced by small caseloads, we find that case severity is best viewed as an endogenous outcome of bargaining. We also find that harsher judges have a higher share of cases failing to reach a plea bargain, but perhaps surprisingly, there is little evidence that large jumps in expected incarceration lead to differential plea bargain outcomes.

Using detailed administrative records, we find that, on average, women receive lighter sentences in comparison with men along both extensive and intensive margins. Using parametric and semiparametric decomposition methods, roughly 30% of the gender differences in incarceration cannot be explained by the observed criminal characteristics of offense and offender. We also find evidence of considerable heterogeneity across judges in their treatment of female and male offenders. There is little evidence, however, that tastes for gender discrimination are driving the mean gender disparity or the variance in treatment between judges.