Estimating Treatment Effects in the Presence of Noncompliance and Nonresponse: The Generalized Endogenous Treatment Model

Abstract

If ignored, non-compliance with a treatment and nonresponse on outcome measures can bias estimates of treatment effects in a randomized experiment. To identify treatment effects in the case where compliance and response are conditioned on subjects' unobserved compliance type, we propose the parametric generalized endogenous treatment (GET) model. GET incorporates behavioral responses within an experiment to measure each subjects' latent compliance type, and identifies causal effects via principal stratification. We use Monte Carlo methods to show GET has a lower MSE for treatment effect estimates than existing approaches to principal stratification that impute, rather than measure, compliance type for subjects assigned to the control. In an application, we use data from a recent field experiment to assess whether exposure to a deliberative session with their member of Congress changes constituents' levels of internal and external efficacy. Since it conditions on subjects' latent compliance type, GET is able to test whether exposure to the treatment is ignorable after balancing on observed covariates via matching methods. We show that internally efficacious subjects disproportionately select into the deliberative sessions, and that matching does not break the latent dependence between treatment compliance and outcome. The results suggest that exposure to the deliberative sessions improves external, but not internal, efficacy.

Keywords: Experimental Methods, Casual Effect Estimation, Principal Stratification

JEL Classification: C90, C93, C11