Primary Submission Category: Proximal Causal Learning
A generalized front-door method when the mediator is confounded
Authors: Helen Guo, Beatrix Wen, Ilya Shpitser,
Presenting Author: Helen Guo*
Unobserved confounding is a fundamental obstacle in causal inference problems. In the graphical modeling literature, a general theory has been developed that allows identification in the presence of hidden variables, with some limitations. In particular, Pearl’s celebrated front-door criterion only allows identification in the presence of treatment outcome unobserved confounding when a mediator variable exists that captures all causal influence from the treatment and outcome, and does not itself suffer from unobserved confounding.
In this paper, we propose a proximal generalization of the front-door criterion, allowing both arbitrary treatment/outcome confounding, and unobserved confounders of the mediator, provided informative proxies for the latter type of confounders are observed. In addition to deriving new identification strategies in this setting, we provide a Neyman orthogonal estimator for the resulting functional under one of these strategies with desirable efficiency properties, and evaluate its performance through simulations.