Description:

Tetsuya Kaji (UChicago Booth): "Assessing Heterogeneity in Treatment Effects"

Abstract: Treatment heterogeneity is often assessed with quantile treatment effects, which lack clear interpretation without rank preservation. We propose three new bounds on heterogeneity that complement quantile treatment effects. First, we bound on the average treatment effects for subgroups defined by Y0 < c for a user-specified value c. This is useful when we are interested in the treatment effect on one tail of the outcome, such as in poverty assistance programs, health insurance, and educational aids. Second, we establish bounds on the proportions of winners for the same subgroups. This bound is useful, e.g., when a policymaker wants to assess the share of winners from a policy change or when doctors want to make sure no patient is harmed by the treatment. Third, we derive a second-order stochastic dominance bound on the distribution of individual treatment effects. This is helpful when we are interested in a non-utilitarian welfare that is concave in the treatment effects.