B1255
Title: Inference for a log-concave counterfactual density
Authors: Charles Doss - University of Minnesota (United States)
Ted Westling - University of Massachusetts Amherst (United States)
Daeyoung Ham - University of Minnesota (United States) [presenting]
Abstract: The problem of causal inference is considered based on observational data (or the related missing data problem) with a binary or discrete treatment variable. The counterfactual density estimation is studied, which provides more nuanced information than counterfactual mean estimation (i.e., the average treatment effect). The shape-constraint of log-concavity (an unimodality constraint) is imposed on the counterfactual densities. Then doubly robust estimators of the log-concave counterfactual density are developed (based on an augmented inverse-probability weighted pseudo-outcome). The consistency in various global metrics of that estimator is shown. Pointwise confidence intervals are developed for the counterfactual density. The confidence intervals can be used to test whether two densities are equal at a given point.
