|Download paper||909 KB|
We propose a test of bias based upon patterns of judicial errors. We model the trial court as minimizing a weighted sum of type I and II errors. We define racial bias as a situation where the weight depends on defendant/victim race. If the court is unbiased, the error rate should be independent of the combination defendant/victim race. We test this prediction using an original dataset on all capital appeals in 1973-1995. We find that in the first and last stages of appeal the probability of error is 3 and 9 percentage points higher for minority defendants who killed white (versus minority) victims.