Searching for Donors: Refining an Imputation Strategy

摘要

For editing data from the 2007 U.S. Census of Agriculture, USDA/NASS built upon the foundation it laid for the 2002 Ag Census, when it ventured for the first time into editing and imputation on a large scale. Institutional acceptance of automated data correction was hindered in 2002 by problems inherent in establishing an entirely new enterprise-level processing system. Changes for 2007 were thus focused on making the system more responsive and reliable. This paper is an overview of improvements facilitated by development of a new imputation subsystem. New tools were provided---within the framework of decision logic tables (DLTs) built by NASS---not only to make donor data available directly to DLTs, but also to improve donor selection, especially through stratification. A variety of SAS techniques achieved better speed in identifying and retrieving nearest neighbors.