Assessing Several Hot Deck Imputation Methods for Several Economic Programs

• Activity Number: 80
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Laura T. Bechtel*+ and Roberta Gibson and Yarissa Gonzalez and Alison Gajcowski and Matthew Nelson
• Companies: U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau and U.S. Census Bureau
• Keywords: hot deck imputation ; simulation ; cross-validation

Abstract

Economic programs conducted by the U.S. Census Bureau often use ratio imputation models to impute missing or erroneous values. These methods are designed to yield "correct" tabulation at the cost of failing to preserve the underlying distribution of the microdata. In this paper, we evaluate hot deck imputation for potential use in three economic programs. Hot deck imputation procedures classify units into disjoint groups based on variables assumed to be correlated with the missing values, then match donor values from respondents to nonrespondents within the classification group. We consider three different hot deck methods. To compare the statistical properties of the totals generated using each hot deck method, we conduct simulation studies with populations modeled from the sample data, contrasting the hot deck results to those obtained using each program's existing imputation methodolo