WiP Seminar: Giacomo Opocher

Abstract

I study the substitutability between individual and aggregate-level data in making causal inference on the effect of a treatment whose unit of reaction is the individual. I show that, under random assignment, one-sided non-compliance, and constant treatment effects, the share of individuals assigned to the treatment can be used as an instrument for the share of treated to deliver a consistent estimator of the local average treatment effect. However, if we allow for heterogeneous treatment effects (e.g. allowing for individuals to be of a high- or low-treatment-effect-type) consistency is not preserved, unless we further assume that there is no differential compliance depending on the type.