A0345
Title: Pattern graph: A graphical approach to nonmonotone missing data
Authors: Yen-Chi Chen - University of Washington (United States) [presenting]
Abstract: The aim is to introduce the concept of pattern graphs--a directed acyclic graph representing how response patterns are associated. Pattern graphs provide an elegant way to model non-monotone missing data. We introduce a selection model and a pattern mixture model formulation using the pattern graphs and show that they are equivalent. Pattern graphs lead to an inverse probability weighting estimator and an imputation-based estimator for estimating a parameter of interest. Asymptotic theories of the estimators are studied, and we provide a graph-based dynamic programming procedure for computing both estimators. We introduce three graph-based sensitivity analysis and study the equivalence class under a generalized version of pattern graphs.
