Developing Replicate Weight-Based Methods to Account for Imputation Variance in a Mass Imputation Application

• Activity Number: 331
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Mark E Asiala*+ and Edward C Castro
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: variance estimation ; multiple imputation ; replicate weights

Abstract

The Census Bureau is currently developing a new imputation-based methodology to improve the estimates of the group quarters population for small areas. As a part of that work, a new variance estimation methodology was needed that would properly account for the imputation variance component in addition to the sampling variance component. Furthermore, any methodology would need to be incorporated into the American Community Survey (ACS) replicate weight variance methodology in order to minimize the impact on production tabulation systems. A benchmark was established using multiple imputation techniques for a fixed sample. We then compared alternative methods that attempted to incorporate the full measure of variance which incorporates the imputation variance into the replicate weights. Details on how the replicate weights are computed are described. This work led to a recommendation of a variance estimation methodology used in the 2011 ACS data products.