B1267
Title: Identifiability of total causal effects from observational data
Authors: Emilija Perkovic - ETH Zurich (Switzerland) [presenting]
Markus Kalisch - ETH Zurich (Switzerland)
Marloes Maathuis - ETH Zurich (Switzerland)
Abstract: One of the most commonly used methods for estimating total causal effects from graphs learned from observational data is covariate adjustment. Previously, we developed a graphical criterion that is sound and complete for covariate adjustment in graphs learned from observational data. However, not all total causal effects are identifiable from observational data. Furthermore, not all total causal effects that are identifiable from observational data are identifiable through covariate adjustment. We outline how to improve the identifiability of total causal effects with covariate adjustment through the addition of background knowledge and discuss some preliminary results on the gap between covariate adjustment and identifiability of total causal effects from observational data.
