Generalized Kernel Ridge Regression for Long Term Causal Inference: Treatment Effects, Dose Responses, and Counterfactual Distributions

  • 2022-01-13 18:51:56
  • Rahul Singh
  • 1

Abstract

I propose kernel ridge regression estimators for long term causal inference,where a short term experimental data set containing randomized treatment andshort term surrogates is fused with a long term observational data setcontaining short term surrogates and long term outcomes. I propose estimatorsof treatment effects, dose responses, and counterfactual distributions withclosed form solutions in terms of kernel matrix operations. I allow covariates,treatment, and surrogates to be discrete or continuous, and low, high, orinfinite dimensional. For long term treatment effects, I prove $\sqrt{n}$consistency, Gaussian approximation, and semiparametric efficiency. For longterm dose responses, I prove uniform consistency with finite sample rates. Forlong term counterfactual distributions, I prove convergence in distribution.

 

Quick Read (beta)

loading the full paper ...