University of Michigan - Ann Arbor

1. Introduce formal sensitivity analysis for linear SCMs and reduce it to a non-zero constrained identification problem.
2. Introduce graphical process PUSH-FORWARD to simplify identification - which can they be solved by a number of algorithms
3. Develop graph-based constrained identification solving algorithm that returns sensitivity curve.

Contributions of Paper

Pearl, Judea, et al. “Sensitivity Analysis of Linear Structural Causal Models.”

http://proceedings.mlr.press/v97/cinelli19a/cinelli19a.pdf

Sensitivity Analysis of Linear Structural Causal Models

We define $G_O$ as model of the the original hypothesis. $G_A$ is the model made to account for confounders. $G_B$ is another variation of $G_O$.

Learn Before

Related