An Inferential Procedure in a Linear Regression Model Under a Model-Free Missingness Mechanism: Without and with Penalization

• Activity Number: 281 - The State-of-the-Art Developments with Nonignorable Nonresponse
• Type: Invited
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Survey Research Methods Section
• Author(s): Jiwei Zhao*
• Companies: State University of New York at Buffalo
• Keywords: nonignorable missing; missingness mechanism; penalization; U-statistics; model-free

Abstract

In many scientific studies, the primary interest is to conduct a hypothesis testing or construct a confidence interval for the parameter of interest. The existence of missing data hinders the researchers to correctly achieve this goal. Although one can adopt a method where the modeling of the missingness mechanism is incorporated; however, this model is unfortunately untestable in applications and any misspecification could yield loss of power or incorrect coverage. In this paper, we propose to overcome this difficulty by imposing a model-free missingness mechanism assumption while still achieving the goal of model identifiability. We first develop the asymptotic representation of the estimator when the penalization is not needed. When the penalization is not avoided, we propose to apply the perturbation method for the variance estimation. Comprehensive simulations studies and an application to a study on orthopaedics are conducted to demonstrate the usefulness of our proposed methodology.