A0417
Title: Efficient nonparametric inference of causal mediation effects with nonignorable missing confounders
Authors: Wei Li - Renmin University of China (China) [presenting]
Abstract: Causal mediation analysis is considered with confounders subject to nonignorable missingness in a nonparametric framework. The approach relies on shadow variables that are associated with the missing confounders but independent of the missingness mechanism. The mediation effect of interest is shown to be a weighted average of an iterated conditional expectation, which motivates the sieve-based iterative outward (SIO) estimator. The rate of convergence and asymptotic normality of the SIO estimator are derived, which does not suffer from the ill-posed inverse problem. Essentially, it is shown that the asymptotic normality is not affected by the slow convergence rate of nonparametric estimators of nuisance functions. Moreover, the estimator is demonstrated to be locally efficient and attains the semiparametric efficiency bound under certain conditions. The efficiency loss attributable is accurately depicted to missingness, and scenarios are identified in which efficiency loss is absent. A stable and easy-to-implement approach is also proposed to estimate asymptotic variance and construct confidence intervals for the mediation effects. Finally, the finite-sample performance of the proposed approach is evaluated through simulation studies, and it is applied to the CFPS data to show its practical applicability.
