A Simulation Study To Evaluate The Robustness Of Recent Methods For Preparing Variance Estimates In The Presence Of Hot Deck Imputation

• Activity Number: 16
• Type: Contributed
• Date/Time: Sunday, August 11, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Michael Sinclair*+ and Nuria Diaz-Tena and Lap-Ming Wun
• Affiliation(s): Mathematica Policy Research, Inc. and Mathematica Policy Research, Inc. and Agency for Health Care Policy and Research
• Keywords: Variance Estimation ; Hot Deck Imputation ; Simulation

Abstract

Recent work on variance estimation in the presence of hot-deck imputation had lead to re-sampling methods for estimating both the sampling variance and the variance due to the imputation process. These procedures are typically conducted by incorporating adjustments to the imputed values during a re-sampling or replication-based variance estimation technique (e.g., jackknife, bootstrap, or balanced half-sample procedures). These methods have been shown to provide unbiased estimates of the combined variance providing the methods are applied to a univariate statistic in which the classing variables used in the hot-deck procedures are known for all units in the sample. Unfortunately, in most survey situations, hot deck procedures are applied to many variables which may be combined to form other statistics, such as sums or ratios. Furthermore, the imputations are often conducted in stages using imputed values in the classing sets. In this paper, we will present the results of a limited simulation study to examine the properties of these methods when applied to a sum of the imputed variables and when the classing set contains imputed variables.