2000 Census

The Census Bureau now plans to conduct the 2000 Census using a combination of mailed questionnaires and followup statistical sampling methods to achieve a more accurate count of the U.S. population than that obtained by the 1990 Census. However, the Bureau and Congress have not agreed on the methodology for conducting the 2000 Census and as of July 1, 1997, the Bureau had yet to share with Congress detailed results of its analysis supporting the use of statistical sampling. As a result, some Members of Congress have expressed concern about the Bureau's plan. With a census dress rehearsal scheduled for 1998, time and resources will be wasted if the Bureau tests a plan that Congress later rejects.

GAO noted that: (1) since GAO's October 1995 testimony, the Census Bureau has continued with the planning of the new design initiatives that GAO and others have suggested, which are aimed at increasing the mail response rate; (2) this is important because it will reduce the need for follow-up visits by enumerators, the most costly, difficult to manage, and error-prone operation in the census; (3) the Bureau has continued to develop its plan for dealing with those who do not respond by mail and for checking the quality of the results it gets from mail responses and from visits by enumerators; (4) its current plan, which was put forth in March 1997, is to statistically sample those who do not respond to its mail survey; (5) the Bureau believes that this approach offers the best combination of reduced costs, improved accuracy expected at various geographic levels, and operational feasibility; (6) the Bureau developed accuracy estimates by simulating what the results of the census would likely be under various design options; (7) after the Bureau made several modifications, GAO concluded that, if the Bureau's methods and assumptions were properly applied, the final data produced in June 1997 should be generally reasonable for use in projecting the likely effects of the Bureau's proposed sampling and statistical estimation initiatives; (8) because of concerns about the potential effects of sampling at the local level, GAO requested that the Bureau provide more detailed data on error rates at the census tract level; (9) the Bureau's simulations showed that the March 1997 sampling design plan would likely produce more accurate population estimates in two-thirds of the census tracts than using a conventional design once again; (10) as of July 1, 1997, the Bureau had not shared the detailed results of its analysis with Congress, nor had it yet fully demonstrated the operational feasibility of its current plan to Congress; (11) the Bureau is planning a dress rehearsal for the 2000 Census in 1998 to demonstrate and test its design features; and (12) it is important for Congress and the administration to reach agreement on the design as soon as possible before the dress rehearsal so that: (a) the Bureau can test what it plans to implement in 2000; (b) Congress and the Bureau can discuss the operational feasibility of the plan in terms of the dress rehearsal results; and (c) Congress and the Bureau can determine whether the dress rehearsal outcomes are sufficiently similar to the results of the Bureau's research and simulations.

Recommendations

Our recommendations from this work are listed below with a Contact for more information. Status will change from "In process" to "Open," "Closed - implemented," or "Closed - not implemented" based on our follow up work.

Director: Team: Phone: