A Penalized Likelihood Approach for Selection Model with Nonignorable Missing Data

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Chi-hong Tseng*+ and Robert Elashoff and Gang Li
• Companies: UCLA and UCLA and University of California at Los Angeles
• Keywords: Missing not at random ; longitudinal data ; pesudo likelihood

Abstract

Selection model is a popular statistical model to handle non-ignorble missing data. It specifies an overall outcome model and a missing mechanism model that describes how the missing data depend on the unobserved outcome. However, because the covariates in the missing model is unobserved, the information contained in the data on the parameters is often limited. The corresponding likelihood surface is often flat, and the parameters can be estimated poorly. A penalized likelihood approach is proposed to address this problem and numerical simutaions are carried out to demonstrate the performance of this method.