Searching for Donors: Finding an Imputation Strategy

• Activity Number: 401
• Type: Topic Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Michael Hogye*+
• Companies: U.S. Department of Agriculture
• Keywords: imputation ; donor ; editing

Abstract

For the 2002 U.S. Census of Agriculture, USDA/NASS employed an array of new tools developed for this purpose. This included the use of donor imputation to adjust records that could not be corrected through other editing strategies. While the goal of incorporating Fellegi-Holt principles proved to be overly ambitious, significant progress was made in broadening the scope of available donors. Large investments in automation of data-editing provided complementary benefits to the imputation process. On one hand, efficient data retrieval and powerful processing made possible the creation of large donor pools and extensive donor searches for nearest neighbors. On the other hand, interactive data-editing capabilities allowed a nationwide network of analysts to correct records for which donor imputation proved problematic. This paper reviews the imputation techniques used, lessons learned, and improvements being considered for imputation in the 2007 Census. Specific issues include the construction of initial donor pools; selection of appropriate similarity measures; scaling of continuous imputed values; and balancing of computational demands with improvements to data quality.