Type I and Type II Error Probabilities in the Courtroom

Abstract

We estimate the likelihood of miscarriages of justice by reframing the problem in the context of misclassified binary choice models. Our estimator is based on new nonparametric identification results, for which we provide methods to empirically test the key identifying assumptions and alternative identification schemes for when these checks fail. Using case-level data from Virginia, we find blacks have both a higher probability of conviction when innocent and a higher probability of acquittal when guilty, relative to whites. We go on to show that this seemingly contradictory result is, in fact, consistent with a model where blacks are discriminated against at both the arrest and the conviction stage of the judicial process.