Unconditional Estimating Equation Approach for Missing Covariates Under Ignorable Missingness

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Lin Lu*+ and Peiyong (Annie) Qu and David S. Birkes
• Companies: Quintiles and University of Illinois at Urbana-Champaign and Oregon State University
• Keywords: missing data ; GEE ; unconditional

Abstract

We present an unconditional estimating equation approach to handle missing data that is missing at random. In a typical generalized estimating equation approach, covariates are generally treated as fixed variables. In our approach we treat them as random. We construct estimating equations simultaneously associated with both response and covariate variables. This enables us to handle missing covariates and consistently obtain missing information through known information. One advantage of our approach is, in contrast to the weighted estimating equation method, that it does require modeling the missing mechanism. Our approach also does not require modeling the conditional distributions that the multiple imputation method requires in order to impute missing values. In addition, it does not require fully specifying the likelihood function, but only requires the first few moments of response