The issue of whether to adjust the 2000 Census for undercount (and overcount) was the subject of intense public debate over the past four years. The Census Bureau spent enormous resources in an effort to reach a well-founded scientific conclusion on whether their methods could be used to adjust the census for redistricting and other purposes not related to reapportionment. (The Supreme Court prohibited the use of adjustment for the purposes of reapportionment in 1999). In the end, the results of their research were too inconclusive to recommend adjustment because the unanswered questions introduced enough “reasonable doubt” to call the accuracy of adjustment into question. Therefore, none of the 2000 census data presented here or in any Census Bureau product have been adjusted for undercount or overcount.
The Executive Steering Committee for Accuracy and Coverage Evaluation Policy at the Census Bureau was responsible for studying the issue of adjustment. On March 1, 2001, this group of senior statisticians and demographers recommended to the Director of the Census Bureau that adjustments for undercount and overcount not be incorporated into the redistricting data files. The Accuracy and Coverage Evaluation (A.C.E) program could not conclude, with a high level of certainty, that the adjusted census results would be more accurate than the unadjusted results by the April 1, 2001 legally-mandated deadline for release of the PL 94-171 redistricting data files. This recommendation was endorsed by Commerce Secretary Donald Evans on March 6, 2001. (The A.C.E. survey was conducted after the census on a national sample of households to determine who was counted and who was missed in the decennial census.)
Subsequent analysis of the adjustment issue by the Census Bureau was conducted to evaluate the utility of using adjusted data for purposes not related to redistricting, including future 2000 Census products, post-censal population estimates and control totals for future surveys. On October 17, 2001 the Census Bureau released A.C.E. Revised estimates and announced that it would use unadjusted census data for these purposes. Their analysis revealed that the A.C.E. survey overstated the net undercount by at least 3 million people, effectively compromising the accuracy of any census adjustment based on the survey. This error was traced most closely to the failure of A.C.E. to accurately account for erroneous enumerations in the census, many of which were duplicates. Unlike the earlier figure of 3.3 million or 1.2 percent net undercount, the national net undercount was determined to be near zero (.06 percent). In March of 2003, the A.C.E. Revision II estimates were released. This initiative further refined the methods used to determine census omissions and erroneous enumerations. The figures issued under A.C.E. Revision II have now determined that there was a net national overcount of approximately 1.3 million persons or 0.5 percent. Also, as part of Revision II, the Census Bureau determined that adjusted numbers will not be used as part of their intercensal estimates program because too many questions about the quality of the data remain, especially at a subnational level.
There are several important issues to note. First, the overcount of 1.3 million persons is a net figure, the result of close to 6 million persons erroneously enumerated and 4.7 million persons omitted, for a total of close to 11 million gross errors. (Since there is often disagreement about what to count as a gross error, the 11 million may be viewed as approximate and dependent on the definition used.) Second, a small undercount for the nation tells us little about undercount in hard-to-enumerate areas, such as inner cities or very rural places, or about groups whose members may have been missed and require adjustment. The Revision II estimates showed a statistically significant national net overcount of 1.13 percent for nonhispanic whites and a significant net undercount of 1.84 percent for nonhispanic blacks. Estimates for other groups were not found to be statistically different from zero. Finally, the Census Bureau seems to have better reconciled the A.C.E. estimate of undercount with an independent estimate of the national population derived from demographic analysis (i.e., estimate based on births, deaths, medicare records, immigration and emigration) in Revision II. In past censuses, independent verification of the total population has been a principal means of verifying the integrity of the national population figure. However, many questions remain about the integrity of the demographic analysis for 2000. In "The Analysis of the 2000 Census: Interim Assessment," the Panel to Review the 2000 Census at the National Academy of Sciences indicated that difficulties in accounting for all immigrants entering the nation and problems in the classification of persons by race make demographic analysis problematic, such that "...demographic analysis should not be used as a standard for evaluating the census or the A.C.E. at this time."
Estimates of undercount for New York City, based on early Census Bureau work and released by the now defunct Census Monitoring Board, at one point had the city’s undercount at 140,000 or 1.7 percent of the population. After incorporating better estimates of omissions and erroneous enumerations in Revision II, the Census Bureau has established that New York City experienced a net undercount of about 36,000 persons or 0.4 percent of the 2000 population. Since this estimate was derived from a sample, it is subject to sampling variability (i.e., sampling error). When sampling error is taking into account, the estimate of 36,000 was not significantly different from zero. While the Census Bureau will continue to work on census coverage issues in preparation for the 2010 Census, this is likely the final statement on the 2000 undercount for New York City.