Using Administrative Records for Imputation in the Decennial Census

• Activity Number: 356
• Type: Topic Contributed
• Date/Time: Wednesday, August 10, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): James Farber*+ and Deborah Wagner
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: Missing data ; matching ; statistical modeling

Abstract

Administrative records provide a potential source of comprehensive, inexpensive, and accurate data for a number of uses in U.S. Census Bureau programs. One promising application is imputation of missing data in the decennial census. The deterministic algorithms traditionally used for imputation rely heavily on the similarity of housing units in small geographic areas. As the country has diversified, however, the assumption that neighbors are alike has grown increasingly tenuous. Hence the U.S. Census Bureau undertook research into alternative methods that may provide more accurate imputed data. This paper discusses two new imputation methodologies based on administrative records. One method is direct assignment, in which administrative data imputes missing data on a matching address or person record. The second method builds a statistical model that relates administrative data to data on the census so predictions from those models fill in the missing census data. These methods were implemented on a set of truth decks and the results evaluated for accuracy and feasibility.