Multiple Imputation Methods Addressing Planned Missingness in a Multi-Phase Survey

• Activity Number: 314 - SPEED: Missing Survey Data: Analysis, Imputation, Design, and Prevention
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 9:25 AM to 10:10 AM
• Sponsor: Survey Research Methods Section
• Author(s): Irina Bondarenko* and Yun Li and Paul Imbriano
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Multiple imputation; responsive survey design; propensity score; combining data from multiple sources; multi-phase survey; calibration

Abstract

Over the past decades the medical field developed rapidly. Emergence of new tests and technologies altered treatment and care patterns. Surveys collecting data over a long time have to respond to these changes. Responsive survey design refers to design with multiple phases, with each phase implementing a different survey. The changes in these surveys include the emergence of new survey questions, or refinement of existing ones, and hence results in data missing by design We propose a tailored multiple imputation approach that allows to combine different phases of survey data into a single data set for analysis purposes. The Individualized Cancer Care Study is a multi-phase study conducting surveys to examine breast cancer treatment experiences and decision making in year 2013-2015. We use this study to demonstrate our multiple imputation approach which consists of three stages: imputation of common variables, stratifying subjects on their propensity being surveyed on the later phase, and imputation of phase-specific variables and imputation of missing values arisen from combining different but related questions between phases.