Estimation in Hierarchical Models with Incomplete Ordinal Response and Ordinal Covariates

• Activity Number: 361
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Yong Zhang*+ and Trivellore Raghunathan
• Companies: University of Michigan and University of Michigan
• Keywords: Missing Data ; Hierarchical Models ; EM Algorithm ; Multiple Imputations

Abstract

Hierarchical models are often used when data are observed at different levels and the interaction effects on the outcomes between variables measured at different levels are of interest. Missing data can complicate analysis using hierarchical models and can occur at all levels, in both outcomes and covariates. Ignoring the subjects with missing data usually leads to biased estimates. We use a combination of the EM algorithm and multiple imputations to develop approximate MLE of the parameters in hierarchical models, assuming missing at random and ignorable missing mechanism. In this paper we consider an ordinal response with missing values as well as continuous and ordinal covariates with missing values at each level. Simulation study is used to demonstrate that our proposed method has desirable repeated sampling properties. The method is also applied to a survey data.