Building a Complete History for Respondents in Longitudinal Surveys Through Imputation

• Activity Number: 633
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Social Statistics Section
• Author(s): Yves Thibaudeau*+ and Sharon O'Donnell
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: intermitent missing data ; longitudinal survey ; attrition

Abstract

Data collection operations for longitudinal surveys are imperfect and often lead to situations where information is missing for some survey units at many points in time. This means there are either gaps in the sequence of longitudinal data expected from the survey, or attrition, meaning all the information for specific units is missing permanently after one point in time. In both cases, we can exploit the information from the units that are intermittently missing to produce whole-unit imputations that are partially adjusted for potential non-response biases. Because of their higher propensity for nonresponse overall, the intermittent responders share additional features with the units subject to permanent attrition, relative to compliant respondents. These similarities can be exploited through parametric modeling or non-parametric techniques, such as the hot-deck, to produce whole-unit imputations with mitigated biases. We examine some of these techniques (Benedetto and Stinson 2009) and proposed some new ones (Markov modeling O'Donnell and Thibaudeau 2012).