A1832
Title: A LATE-targeting experimental design under multivalued treatment
Authors: Tetsuya Kaji - University of Chicago (United States) [presenting]
Abstract: Experiments with multivalued treatments are appealing when experimentation is costly. Yet, under imperfect compliance, the identification and even the definition of compliers is unclear. First, the impossibility of identifying margin-specific local average treatment effects (LATEs) in a standard randomized experiment, even under a strong generalization of one-sided compliance, is established. Then, a novel experimental design is proposed that achieves identification of margin-specific LATEs under a mild generalization of one-sided compliance.
