CFE 2019: Start Registration
View Submission - CMStatistics
Title: A simple and efficient estimation of average treatment effects in the presence of unmeasured confounders Authors:  Zheng Zhang - Renmin University of China (China) [presenting]
Abstract: A critical condition in the treatment evaluation literature is that, conditional on all confounders, participation decision is independent of potential outcomes. If some of those confounders are not observable, then conditional on the observable confounders, participation decision is no longer independent of potential outcomes and consequently the average treatment effect (ATE) is not identified without further assumption. Indeed, the literature establishes that ATE is not identified even if standard instrumental variables are available. Two proposals are suggested. The first one assumes that the unmeasured confounders are not interacted with the treatment in potential outcomes or with the instrument in participation decision. Under this additional restriction, ATE is identified. The other assumes that the complier's treatment status is monotone in instrument. Under this additional restriction, the local average treatment effect (LATE) for the compliers is identified. A simple and efficient estimation of ATE and LATE is proposed which does not estimate the influence function parametrically, thereby is more robust than the existing methods.