A1620
Title: On the graphical rules for recovering the average treatment effect under selection bias
Authors: Haidong Lu - Yale University (United States) [presenting]
Yichi Zhang - Yale University (United States)
Abstract: Selection bias is a major obstacle to valid causal inference in epidemiology. Over the past decade, several graphical rules based on causal diagrams have been proposed as sufficient identification conditions for addressing selection bias and recovering causal effects. However, these rules are typically tied to specific identification strategies and estimators. Two important cases of selection bias cannot be addressed by these existing rules and their associated estimators: one in which selection is a descendant of a collider between the treatment and the outcome, and another in which selection is affected by the mediator. To address selection bias and recover the average treatment effect in these settings, alternative graphical rules are proposed and identification formulas are derived using g-computation and inverse probability weighting (IPW) based on single-world intervention graphs (SWIGs). Simulation studies evaluate the performance of the proposed estimators, particularly in settings where traditional crude selected-sample analyses (i.e., complete-case analyses) yield conclusions that contradict the truth.
