|
Abstract:
|
The objective of many studies in health and social sciences is to evaluate the causal effect of a treatment or exposure on a specific outcome using observational data. In such studies, the exposure is typically not randomized and therefore confounding bias can rarely be ruled out with certainty. The instrumental variable (IV) design plays the role of a quasi-experimental handle since the IV is associated with the treatment and only affects the outcome through the treatment. A valid IV can be used to obtain a test of the null hypothesis of no treatment effect with nominal type 1 error rate. Beyond testing for the causal null, one may wish to obtain an accurate estimate of the treatment causal effect. In this paper, we present a novel framework for identification and estimation using an IV of the marginal average causal effect of treatment amongst the treated (ETT) in the presence of unmeasured confounding. For identification, we provide sufficient and necessary conditions and easy to use sufficient conditions. For inference, we propose three different semiparametric strategies: (i) inverse probability weighting (IPW), (ii) outcome regression, and (iii) doubly robust (DR) estimation
|
