Can Racially Unbiased Police Perpetuate Long-Run Discrimination?

We develop a stylized dynamic model of highway policing in which a non-racist police officer exhibits a cognitive bias: relative overconfidence. The officer is given incentives to arrest criminals but faces a per stop cost which increases when the racial mix of her stops differs from that of the population. Every period, she observes the racial composition of jail inmates (generated from arrests made by her peers) and forms estimates about the crime rates of each race. In some settings, her overconfidence leads her to overestimate the crime rate of one race relative to another causing the long-run racial composition of the jail population to deviate from the "fair" one (one where the racial mix in jails is identical to that in the criminal population). We compare this to a situation where officers have detailed stop data on each race, similar to data being currently collected in many US states.