Home > Events > 2019 Clogg Lecture with Michael Sobel

2019 Clogg Lecture with Michael Sobel

Main Content

Michael Sobel is a professor at Columbia University, Department of Statistics.
24 April 2019
102 Thomas Building
Add event to calendar

Michael Sobel

Public Talk

Date: April 24, 2019
Time: 5:45 PM
Location: 102 Thomas Building

"Between Causation and Association: The Role of Sex, Race and Political Party in EEOC Litigation Outcomes, 1996-2006"

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.

Scientific Talk
Date: April 25, 2019
Time: 11:00 AM
Location: 406 Oswald Tower

"Identification of Treatment Effects in Fixed Effects Models for Longitudinal and Clustered Data:  Problems, with Illustrations from Demography''