A robust and efficient approach to causal inference based on sparse sufficient dimension reduction (Q2414095)

This paper deals with the problem of causal inference in the presence of a large number of covariates. A new approach for inference about the average treatment effect based on a flexible modeling strategy incorporating penalized variable selection is proposed. More precisely, a two-step method is introduced, where first an estimator based on an efficient influence function is constructed and afterwards a new sparse sufficient dimension reduction method is applied, which allows to avoid restrictive parametric modeling assumptions. \par The authors discuss certain properties of the method, including consistency of the variable selection step, and prove asymptotic normality even in cases where the variable selection is inconsistent. The proposed method is illustrated in a simulation study and a biomedical application.