Title: Bounds on causal direct and indirect average treatment effects in the presence of noncompliance
Authors: Xuan Chen - Renmin University of China (China) [presenting]
Abstract: Causal mediation effects are examined in an experimental setting with treatment noncompliance. A binary instrument is used to address noncompliance. Given a particular mechanism variable, we decompose the local average treatment effect (LATE) for individuals who would comply with the assigned treatment (i.e. compliers) into the complier natural indirect (or mechanism) average treatment effect (CMATE), and the complier natural direct (or net) average treatment effect (CNATE). CMATE refers to the part of the LATE that works through the mechanism variable, while CNATE refers to the part that works through all other channels. Without strong assumptions for point identification, we employ the principal stratification approach to derive nonparametric sharp bounds on CMATE and CNATE by imposing weak monotonicity assumptions on mean potential outcomes within or across principal strata, which are subpopulations defined by the joint potential values of the treatment receipt and mechanism under each value of the treatment assignment indicator. We also obtain sharp bounds on the local mechanism and net effects for different strata of compliers. To illustrate the identifying power of our bounds, we analyze the part of the effects of a training program on participants' future earnings and employment that works through the attainment of a vocational degree.