Estimating causal effects of treatments in randomized and nonrandomized studies.

Abstract

Rubin formalizes causal effects in terms of potential outcomes — what would happen to a unit under treatment vs control — clarifying when, and under what assumptions, observational studies can recover causal effects.