Using the Fraction of Missing Information to Tailor Survey Design

• Activity Number: 16
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Rebecca Andridge*+
• Companies: The Ohio State University
• Keywords: Multiple imputation ; Fraction of missing information ; Missing data ; Survey data

Abstract

The fraction of missing information (FMI) has recently been proposed as an alternative to the response rate for monitoring the quality of survey data and for tailoring survey design. In this paper we discuss the use of FMI as an indicator that can identify which nonrespondents should be targeted in order to reduce variability in estimates. Two-stage multiple imputation (Harel, 2007) is one method for estimating the relative contribution to FMI of individual observations. We investigate analytically and through simulation its properties as a method for identifying which nonrespondents have the largest impact on variability in outcomes. Since two-stage multiple imputation is computationally intensive, we also investigate the use of subject-level between-imputation variance as a proxy measure for contribution to FMI, a simpler measure to obtain. These methods are illustrated using data from the Ohio Family Health Survey (OFHS).