Abstract:
|
In this talk, we study the nonparametric methods for missing data problems. We introduce the concept of pattern graphs--directed acyclic graphs representing how missing patterns are associated. A pattern graph represents an identifying restriction that is nonparametrically identified and is often a missing not at random restriction. We introduce a selection model and a pattern mixture model formulation using the pattern graphs and show that they are equivalent. A pattern graph leads to an inverse probability weighting estimator as well as an imputation-based estimator that can be nonparametrically estimated. We discuss the identification theory, asymptotic theory, and multiply-robustness properties of pattern graphs.
|