coverage and contain costs, it would be necessary to use statistical estimation to account for the hard-to-enumerate population.
It was expected that demographic analysis would continue, as in the past, to provide a benchmark for the 2000 census results for age, race, and sex groups at the national level; however, demographic analysis could not provide reliable estimates for subnational geographic areas or for race and ethnicity groups other than blacks and nonblacks. Little research attention was devoted to demographic analysis in the 1990s.
Research concentrated on refining the methodology for dual-systems estimation (DSE) using the results of an independent population survey (P-sample) matched with a sample of census enumerations (E-sample) to estimate the population. There was also research to compare DSE with a method called Census Plus, in which enumerators would revisit a sample of households to obtain a roster of household members and immediately reconcile that roster with the roster from the census. When weighted, the union of the two rosters (after making any deletions or additions) would provide an estimate of the population.
A major test in 1995, which incorporated sampling for nonresponse follow-up, Census Plus, and refined DSE procedures, determined that Census Plus was itself subject to undercounting. Hence, subsequent research focused on further improving the DSE methodology, particularly the ability to conduct the independent survey and other DSE operations on a schedule to permit using the results to adjust the state population counts for reapportionment by December 31, 2000. This use of DSE for census adjustment was termed Integrated Coverage Measurement (ICM). The Census Bureau proceeded with plans for ICM but—as discussed in Section 3-C.4—a 1999 Supreme Court decision forced the move from ICM to a more limited-scale Accuracy and Coverage Evaluation (A.C.E.) Program.
This brief review indicates the substantial amount of research that the Census Bureau devoted to planning innovations in major areas of census operations for 2000. The questionnaire and mailing package research was particularly effective in that the 2000 design incorporated most of the results of this research (e.g., multiple mail-