A Causal Framework for Observational Studies of Discrimination

In studies of discrimination, researchers often seek to estimate a causal effect race or gender on outcomes. For example, in the criminal justice context, one might ask whether arrested individuals would have been subsequently charged convicted had they different race. It has long known that such counterfactual questions face measurement challenges related omitted-variable bias, and conceptual definition estimands for largely immutable characteristics. Another concern, which subject recent debates, is post-treatment bias: many discrimination condition apparently intermediate outcomes, like being arrested, themselves may be product potentially corrupting statistical estimates. There is, however, reason optimistic. By carefully defining estimand—and by considering precise timing events—we show primary quantity interest can estimated under an ignorability hold approximately some observational settings. We illustrate these ideas analyzing both simulated data charging decisions prosecutor’s office large county United States.