2019 Clogg Lecture with Michael Sobel

Google map

Public Lecture

102 Thomas Building

Wednesday, April 24 at 5:45 PM

A longstanding question in American literature on judicial decision making is how case outcomes depend on judge attributes such as race, sex, and ideology. The literature is inconclusive at best. Typically, using case covariates and judge attributes as predictors, statistical models of various outcomes are estimated, and the coefficients associated with the attributes interpreted as effects. However, researchers do not regard the features of an attribute as treatments, but as proxies for unmeasured variables that vary by feature: thus, this interpretation is incorrect, and it is necessary to reformulate the question of interest and develop a suitable methodology to address it. The primary concern in the literature is that judges with different features of an attribute, for example, male and female judges, will handle cases differently. But studying how judges with a given feature handle cases to which they are assigned may not be indicative of how judges with a different feature would handle these cases. Ideally, one wants to compare judges with different features on a common set of cases, taking also into consideration within feature heterogeneity in outcomes. We propose several estimands based on this idea. For each case, we define potential outcomes for every judge eligible to hear the case, and we use these to define a unit (case) causal comparison that compares judges with different features; the unit causal comparison is then used to define average and percentile causal comparisons. An estimation strategy based on a Bayesian hierarchical model for award amounts in cases filed by the Equal Employment Opportunity Commission between October 1, 1996, and September 30, 2006, is used to estimate these quantities. We find little support for the notion that non-white judges favor plaintiffs more than white judges; similar comments apply with respect to female and male judges, and to judges appointed by Republican and Democratic presidents.