A Simple Test for Prejudice in Decision Processes: The Prediction-Based Outcome Test

We propose a novel implementation of the outcome test for diagnosing prejudice in decision processes, the Prediction-Based Outcome Test (P-BOT). We motivate our approach with a model of prejudice in pretrial detention decisions. The main empirical challenge when implementing outcome tests is the identification of marginal individuals. Our method uses the predicted release status to identify marginally released defendants. Concretely, we provide sufficient conditions under which a ranking of the propensity score among released defendants identifies those who are more likely to be at the margin given their observables, and propose a set of diagnostics to empirically assess the plausibility of the identification assumptions. Some appealing features of the P-BOT are that (i) it does not require instruments nor random assignment of judges, (ii) it is robust to standard omitted variable bias, and (iii) it is very easy to implement. However, its performance depends on the availability of good predictors, something that can be assessed by the econometrician. We use the P-BOT to test for prejudice in pretrial detentions against the main ethnic group in Chile, the Mapuche, using nationwide administrative data. We find strong evidence of prejudice against Mapuche defendants and show that the discrimination patterns are likely to be non-binary. We assess the relative performance of the P-BOT and alternative approaches to test for prejudice, and discuss the test’s interpretation in more general versions of the model, sketching a new taxonomy of prejudice.