Title: Common support diagnostic for heterogeneous treatment effect
Authors: Jennifer Hill - New York University (United States) [presenting]
Abstract: Robust estimation of average causal effects requires strong assumptions that are often untestable. The common support assumption is testable in theory however in with high dimensional confounders this can become computationally challenging. When the goal is estimation of subgroup effects or individual-level treatment effects identifying sufficient common support is even more difficult. Bayesian nonparametric strategies for identifying common causal support for heterogeneous treatment effects will be presented. These have the advantage of more robust estimation of targeted treatment effects (for subgroups and individuals) and superior uncertainty quantification as compared to common frequentist alternative. Moreover, these approaches have the advantage of a natural and effective strategy for identifying observations and subgroups for whom common causal support is likely to be satisfied.