The economics and measurement of racial bias in law enforcement

This dissertation consists of three independent but related papers. The first paper develops a new empirical approach to measure racial bias which features fixed effects generated at an agency level to account for differences in the regional cost of assessments, as well as mitigate omitted variable bias. This empirical approach is applied to a unique dataset, which contains the results of all breathalyzer tests administered in Washington State from 2003 through 2006. Applying both the fixed effect test and conventional tests for racial bias to the Washington State Patrol dataset, I find evidence of racial inequality exhibited in assessments of both black and Asian motorists. However, the traditional test for racial bias suggests that police officers are racially biased in favor of Hispanic motorists. In contrast, using the fixed effects model, this outcome is no longer significant. This finding provides evidence that it is important to incorporate heterogeneous regional costs of assessment into models of racial bias. The second paper explores a significant problem of studying racial bias in law enforcement called the infra-marginality problem. That is, if the underlying distributions of the probabilities of guilt differ by race, the link between assessment outcomes and racial bias will be imperfect. In this paper I model the potential for the distribution of characteristics to differ by race, and then use simulations to compare how these differences in characteristics impact existing tests for racial bias. I find that individually the two most common tests for racial bias will be unreliable tests for racial bias; however, a test can be constructed which combines the two, and that test will be robust to the criticism of infra-marginality. The third paper of my dissertation employs a representative voter model to study the referendum that led to the construction of Qwest Field in Seattle. Voters "believed" the all-or-nothing threat that characterizes all sports facility votes. In addition, there are proximity values since the odds of voting yes decrease with distance from the proposed facility site and increase with population. All-in-all, the nine closest counties, also more highly populated and richer counties, had their way in this election