Most existing causal inference literature focuses on prospectively evaluating the total effect of single or multiple causes on an outcome. In fields such as law and public health, however, it is often essential to retrospectively find the true causes given the evidence, which often includes the observed outcome variable. In this talk, I will review recent advances in causal attribution analysis for retrospective settings with multiple causes. Focusing on the case of two binary causes and a binary outcome, I will discuss how posterior probabilities can be used to characterize the interactive causes of an observed outcome. I will also present a posterior responsibility attribution framework designed to quantify the share of responsibility attributable to each cause. To illustrate, I will discuss some practical examples, such as apportioning responsibility for lung cancer between smoking and asbestos exposure.
