Modeling Incomplete Longitudinal Bounded Outcomes: An Application Study

• Activity Number: 649
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Nai-Wei Chen* and Yong-fang Kuo and Kyriakos S. Markides
• Companies: The University of Texas Medical Branch and The University of Texas Medical Branch and University of Texas Medical Branch
• Keywords: incomplete response ; bounded distribution ; longitudinal data

Abstract

Aging research often involves unavoidable incomplete data measured or observed longitudinally on individuals. Inappropriate handling of the presence of nonresponse or attrition may result in biased or inefficient inference. In this study, we attempt to fill in an extension to the situation in which there are non-ignorable or no sequential ignorable missing bounded outcomes, such as MMSE scores (range: 0 to 30), observed longitudinally incorporating with a missing categorical covariate at baseline, such as ADLs. The primary objective is to tackle and alleviate the complexity of model estimation for normal or skew-normal censored distributed outcomes through MCMC sampling techniques combined with a nonparametric Bayesian imputation for incomplete covariates. We apply the proposed method to a data set arising from H-EPESE study to assess the effect of physical activity on cognitive functioning in late life.