Multiple Imputation of Disease Status Based on a Two-Phase Sample Design

• Activity Number: 333
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Steven G. Heeringa*+
• Companies: University of Michigan
• Keywords: Missing by design ; Multiple imputation ; Missing information ; Prediction ; Dementia ; Alzheimer's Disease

Abstract

The Aging, Demographics and Memory Study (ADAMS) was based on a stratified, probability subsample of 856 persons age 70 and older selected from the full panel of the Health and Retirement Study (HRS). In addition to replicating the full battery of survey-based health and cognitive test measures employed in the biennial HRS, the ADAMS data collection involved detailed in-home psychometric testing, neuropsychological evaluations, and physical measures that were used by a panel of medical experts to assign disease classification (dementia, Alzheimer's Disease) to each consenting subject. This paper investigates model-based methods for imputing dementia status to all HRS panel members age 70+ based on the ADAMS observations of the relationship of between assigned disease classifications and the putative risk factors observed in both ADAMS and the full HRS panel interviews.