FOR IMMEDIATE RELEASE WEDNESDAY, MARCH 12, 2003 Mark Tolbert III CB03-CS.02 Public Information Office (301) 763-3691/ 457-3620 (fax) (301) 457-1037 (TDD) e-mail: <2000usa@census.gov> Statement of Census Bureau Director C. Louis Kincannon on Accuracy and Coverage Evaluation Revision II Context In conjunction with Census 2000, the Census Bureau conducted an Accuracy and Coverage Evaluation (A.C.E.). The Census Bureau intended to use the estimates produced by the original A.C.E. to adjust the census for both undercounts (omissions) and overcounts (erroneous enumerations). In March 2001, the Census Bureau recommended against census adjustment because the results of the initial A.C.E. were dramatically inconsistent with Demographic Analysis, another independent measure of census coverage. During the summer of 2001, the Census Bureau re-evaluated the results of this original A.C.E., and in October 2001 once again decided against adjusting the decennial census. In this instance, we cited the small size of the estimated net undercount, and that there were only four estimates of coverage developed in the course of the second evaluation and these were at the national level. During the course of the research for the October 2001 decision, serious concerns emerged regarding the number of duplicate and erroneous census enumerations that were found during the course of the research. In response, the Census Bureau researchers conducted extensive, groundbreaking research throughout 2002. The Census Bureau, in effect, "re-opened" the A.C.E. and without reentering the field, completed a new coverage evaluation and adjustment using enhanced and sometimes new procedures that were targeted to address specific error in both the census and the coverage evaluation survey. The Census Bureau conducted this research, A.C.E. Revision II [pdf], to provide a more in-depth analysis of, and where possible corrections to, the estimates of census net coverage. In addition, this research has allowed us to understand more fully the properties of the census, how to account properly for census coverage, and how to improve our planning for 2010. The Findings from A.C.E. Revision II A.C.E. Revision II revealed a great deal of additional information about the accuracy of the census and about the accuracy of our measures of coverage evaluation. The decennial census is a complex process, involving millions of operations and decisions that contribute to the overall coverage and accuracy of the census, and it is not without error. Among the specific issues for which we must do additional research are residence rules, the use of proxy data, as well as the quality of proxy data, and procedures for identifying duplicates. A.C.E. Revision II results contain the following key findings: - Rather than an estimated net undercount of 3 million persons, there was an estimated net overcount of 1.3 million persons. (This would have been an estimated overcount of 3 million persons, except for the application of an adjustment for correlation bias. Correlation bias has been observed and measured in the past but was never included in estimates of net census coverage.) - This includes a net overcount for Non-Hispanic White that is statistically significant, as well overcounts in the Non-Hispanic Asian and American Indian on Reservation populations that are not statistically significant. - There was a net undercount, however, for Non-Hispanic Black that is statistically significant. A.C.E. Revision II results also indicate net undercounts for the Hispanic, Hawaiian and Other Pacific Islander, and the American Indian off Reservation populations that are not statistically significant. - There is also an estimated net overcount for children aged 0-9 that is statistically significant according to A.C.E. Revision II. - There were at least 5.8 million duplicates that are now measured in Census 2000. The March 2001 estimates dramatically underestimated these duplicates, as well as duplicates in the survey. Decision and Future The Census Bureau believes that the estimates produced from A.C.E. Revision II represent the most accurate assessment of Census 2000 coverage to date, replacing the flawed results of the March and October 2001 A.C.E. processes. However, we still have technical concerns about the limitations of the methodology and the data produced by A.C.E. Revision II. In fact, due to our concern regarding the effects of the adjustment for correlation bias and the effect of synthetic adjustment on smaller geographies, as well as other limitations, the Census Bureau reaffirms its prior decision (October 2001) that the intercensal population estimates will be based on the official Census 2000 results rather than on a population base adjusted using revised estimates of census coverage. Concern and discussion of these limitations, as well as the limitations of the census itself, was part of an extensive technical and scientific process. We reached a consensus in making the decision not to adjust the intercensal estimates through open discussion and debate of the technical merits of A.C.E. Revision II. This process also reaffirms our confidence in the decisions made in March and October of 2001. It is our consensus that there are serious considerations that must be addressed in order to assess accurately coverage for the decennial census and that those considerations cannot be addressed in time to meet the deadline for redistricting. Finally, the Census Bureau will now move forward from Census 2000. We will use this research to improve the coverage of the decennial census and coverage evaluation program for 2010. We will always bear in mind that the decennial census belongs to the people of the United States and we will seek the counsel of our stakeholders, especially our peers in the scientific and policy community, as we plan for the future. Attachment [pdf]