Primary Submission Category: Machine Learning and Causal Inference
Nonparametric Identification and Estimation of Average Treatment Effects with a Latent Exposure
Authors: Ying Zhou, Eric Tchetgen Tchetgen,
Presenting Author: Ying Zhou*
In many practical scenarios, the direct observation of exposure variables is not feasible, hindering the ability to draw valid causal inferences. To address this issue, we propose a method that employs two proxy variables of a binary latent exposure, enabling the identification of the average treatment effect (ATE). We then derive the efficient influence function for the ATE, and construct an efficient nonparametric estimator. A significant obstacle to our approach is the estimation of nuisance functions involving the latent exposure, which prevents the direct application of standard machine learning algorithms. To resolve this, we introduce a novel EM-like algorithm, thus adding a practical dimension to our theoretical contributions. This methodology can be adapted to analyze causal functionals beyond ATE, provided two proxies of the latent exposure are available.