Primary Submission Category: high-dimensional causal effects estimation
Causal Inference with High-dimensional Discrete Covariates
Authors: Zhenghao Zeng, Edward Kennedy, Sivaraman Balakrishnan,
Presenting Author: Zhenghao Zeng*
When estimating causal effects, covariate adjustment is often applied in estimating the outcome model and propensity score. The desired properties of the estimator are typically based on fast convergence rates of these nuisance function estimators, for which additional structural assumptions (e.g. smoothness) are usually required on the nuisance functions. However, in real applications, researchers may only have access to discrete covariates (with potentially a large number of levels). In this setting, commonly used structures such as smoothness fail to hold and the behavior of the estimator has not been well-understood. In this work, we study estimation of the average causal effect in a model where the covariates required for confounding adjustment are discrete but arbitrarily high-dimensional. Specifically, we develop point estimation theory for two causal estimands: average treatment effects (ATE) and variance-weighted average treatment effects (WATE). We consider commonly used estimators and examine conditions required for consistently estimating the functionals of interests. The results are illustrated via simulation studies. Importantly, we also characterize minimax lower bounds of the target functionals.