The Administrative Records Experiment in 2000: an Application to Population Count Estimation via Triple Systems Estimation

The U.S. Census Bureau has a long trackrecord of efforts to assess the accuracy of decennial census counts, and has, with the help of others, investigated alternatives to traditional census taking such as the use of administrative records. In past censuses and census tests, the Bureau has also used administrative records to supplement traditional counting methods and to explore the possible uses of multiple systems estimation. This paper describes the use of a three-way cross-classification of the the Administrative Records Experiment (AREX) 2000 bottom-up (BARCUF) data file, Census 2000, and the Accuracy and Coverage Evaluation Survey (ACE) to create block-level triple systems population estimates within the five AREX 2000 counties. We present a statistical approach in which the AREX 2000 and Census 2000 results are considered the primary population lists available for all blocks, and the ACE results are treated as an auxiliary random sample of blocks from a third population list.