STRATEGY FOUR: USE STATISTICAL METHODS

In 1990, respondents who did not return their census forms by mail cost the Census Bureau more than those who did, as temporary census workers were needed to conduct personal visits with nonresponding households. Statisticians agree that incorporating widely accepted statistical methods into Census 2000 will produce better results at less cost. So for Census 2000, we will make every attempt to find everyone. Some households, however, will neither mail in nor phone in their response. So we will use personal visits to obtain responses from the remaining addresses, to ensure that we directly contact at least 90 percent in each census tract.

Using sampling to gather information on nonrespondents will ensure that Census 2000 is built on a solid core of responses. It will ensure that we can complete our personal visits with no loss of accuracy but with substantial savings of time and money. Sampling will allow us to make scientific estimates of the population for the final 10 percent of the housing units.

Our experience in the last six decennial censuses has demonstrated that having responses from 100 percent of the housing units does not ensure inclusion of 100 percent of the population. People are left out for many reasons, and our objective is to account for everyone.

To check the quality of our work and to reach our goal of accounting for 100 percent of the population, we will take a second independent sample--of about 750,000 housing units--of the total population and conduct a second interview. We will use the information from the second interview as the basis for quality checking all our results: the results from the mail returns, the Be Counted program, telephone interviews and possibly the Internet, and personal followup visits. This quality check survey will lead to a "one-number" census and will eliminate the need for subsequent adjustment of the decennial count. We will use demographic analysis to validate the results.

By using both kinds of sampling--that is, sampling for nonresponse and quality check sampling--the accuracy of Census 2000 will be very high for all states, congressional districts, and other populous areas.

There always will be some uncertainty surrounding population totals for some smaller areas, such as census blocks, census tracts, or small communities. Unlike previous censuses, for Census 2000, we will know the level of uncertainty for all areas. The totals for historically undercounted areas will be much better than with traditional methods.

Contact: 2000usa@census.gov