SUTVA short for Stable Unit Treatment Value Assumption requires that potential outcomes for one individual are independent of every other individual.

University of Michigan - Ann Arbor

The Blessings of Multiple Causes - Wang, Blei 2019

Deconfounder Algorithm

The goal of Causal Inference is to describe the distribution of potential outcomes to each state of a causal factor. Multiple Causal Inference does this for multiple causal factors / vector of factors.

Multiple Causal Inference

Argument for Deconfounder

Defintion : SUTVA

Identifying Average Causal Effects of all Causes

Identifying Average Causal Effects of Subsets of Causes

Ignorability

Identification of conditional mean potential outcome

Lemma 3

Lemma 4

Learn Before

Related