CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Graphical criteria for efficient covariate adjustment in causal linear structural equation models Authors:  Leonard Henckel - ETH Zurich (Switzerland) [presenting]
Emilija Perkovic - ETH Zurich (Switzerland)
Marloes Maathuis - ETH Zurich (Switzerland)
Abstract: When the underlying causal graph is known, there exists a sound and complete graphical criterion for when a covariate set allows for unbiased causal effect estimation. However, typically a large number of sets fulfils this criterion. Restricting ourselves to the causal linear structural equation model setting, we introduce graphical criteria that allow one to identify in many cases which of two valid adjustment sets provides the smaller asymptotic variance. Even though this result only induces a partial ordering, it can be used to identify a set that always provides the optimal asymptotic variance. Alternatively, when this set cannot be used, the graphical criterion can also be used for strictly beneficial pruning.