Survivor-Complier Causal Effects with Selection on Treatment

• In Event: Causal Inference

Abstract

Pre-treatment selection or censoring (`selection on treatment') is very common in instrumental variable studies: it can occur when two treatment levels are compared ignoring the third option of neither treatment, in `censoring by death' settings where treatment is only defined for those who survive long enough to receive it, or in general in any study where the treatment is only defined for a subset of the population. Unfortunately, standard analyses are biased in the presence of such selection. In this work we propose a novel estimand for these settings, called the survivor-complier causal effect. Although the effect is not identified under standard assumptions, it is possible to construct bounds. We derive these bounds, and propose a doubly robust and semiparametric efficient approach for estimating them; importantly, our methods allow for high-dimensional confounding adjustment, as well as valid inference even after employing machine learning.

Author

• Luke Keele, Georgetown University