IsoLATEing: Identifying Counterfactual-Specific Treatment Effects with Cross-Stratum Comparisons

Abstract

Instrumental variable (IV) estimates may be difficult to interpret when there are multiple alternatives to treatment. I consider identification of counterfactual-specific local average treatment effects via interactions of the instrument and stratifying controls. Deriving general IV estimands, I establish identification under mean-independence of complier effects visa-vis the strata. I then extend IV to a two-step procedure valid under conditional independence. The theory is illustrated in the context of GED certification, where individuals may otherwise graduate high school or drop out. I also show how it can be used to address differential attrition with an application to the Oregon Medicaid experiment

Keywords: local average treatment effects, compliers, returns to schooling, Medicare, attrition

JEL Classification: C26, C36, I11, I21