Title: Semiparametric inference for causal effects in graphical models with hidden variables
Authors: Razieh Nabi - Johns Hopkins University (United States) [presenting]
Rohit Bhattacharya - Johns Hopkins University (United States)
Ilya Shpitser - Johns Hopkins University (United States)
Abstract: The last decade witnessed the development of algorithms that completely solve the identifiability problem for causal effects in causal graphical models with hidden variables. However, much of this machinery remains underutilized in practice, owing to the complexity of estimating identifying functionals yielded by these algorithms. We describe simple graphical criteria and semiparametric estimators that bridge the gap between identification and estimation for causal effects of a single treatment on an outcome. We first discuss influence function-based doubly robust estimators that cover a significant subset of hidden variable causal models where the effect is identifiable. This allows us to incorporate flexible machine learning methods into causal inference pipelines that go beyond the standard, but often unreasonable, assumption of conditional ignorability. We further characterize an important subset of this class for which we demonstrate how to derive the estimator with the lowest asymptotic variance, i.e., one that achieves the semiparametric efficiency bound. Finally, we consider semiparametric estimators that resemble reweighed influence function-based estimators for any identified single treatment causal effect parameter. A Python software package named Ananke that implements these methods is available.