Judicial Errors: Evidence from Refugee Appeals

Different judges, doctors, loan officers, and patent examiners make different decisions, generating costly uncertainty over ultimate outcomes. In this paper, I use multiple-stage decision-making institutions to identify nonparametric bounds on disagreement between decision-makers. I bound disagreement to at least 17% of all Canadian refugee appeals, 150% larger than the estimate using existing methods and substantial relative to an average approval rate of 14%. I aggregate disagreement into judge-specific measures of quality, and find that quality improves with experience, declines with workload, and is higher for judges appointed under a nonpartisan regime. Finally, I adopt my method to test and reject the typical examiner-assignment monotonicity assumption.