WASHINGTON -- To cut costs and improve accuracy, the Census Bureau said yesterday that it would actually count only 90 percent of the United States population in 2000 and rely on statistical sampling methods to determine the number remaining.
The plans, announced at the Commerce Department, mean that for the first time, the official tally of the U.S. population, done every 10 years and used to apportion seats in the House of Representatives, will be based in part on a scientifically determined estimate rather than the actual head count conducted through a massive direct mail campaign.
Census Bureau officials say the revised method is needed to keep costs down and to avoid a repeat of the 1990 census, which missed record numbers of people that traditionally had been hard to count, mainly members of ethnic and racial minorities.
"What we intend to do to meet our twin goals of reducing costs and increasing accuracy is to make a much greater use of widely accepted scientific statistical methods, and sampling is first and foremost among them," said Martha Farnsworth Riche, director of the Census Bureau.
Scientific sampling, commonly associated with public opinion polls, involves questioning a randomly selected group of people and using the information derived from them to count or describe a much larger group.
The Census Bureau has used sampling for years to determine such national characteristics as the poverty and unemployment rates. But it has never used it to determine the official population.
In 1990, the government refused to adjust the census results to correct an acknowledged undercount of about 4 million people, mostly black and Hispanic residents of large cities. That refusal prompted a lawsuit by a coalition of big cities, led by New York. The case is pending before the Supreme Court.
Critics of the bureau's use of sampling argue that is is unconstitutional because the Constitution calls for an "actual enumeration."
But decisions in lower federal courts have approved the use of sampling as long as it supplements, and does not supplant, an actual count.
Census Bureau officials say sampling is needed because of the increasing number of people who fail to return their mail-in forms.
In 1990, only 63 percent of American households sent the form back, compared with 85 percent in 1970, the first year the census was conducted by mail.
The Census Bureau has to dispatch temporary employees, called enumerators, to households that do not return the forms, greatly increasing the cost.
Officials say that in the 2000 census, mail-in forms and enumerators will be used until 90 percent of the households in a county have been counted.
Once that response level is achieved, a statistical sample of 10 percent of the remaining households will be selected, and enumerators will be dispatched to count them. The results will be used to estimate the total number of those originally missed.
"Better to hound one in 10 than 10 in 10," said Everett M. Ehrlich, undersecretary of commerce for economic affairs, who oversees the Census Bureau.
The bureau also plans to simplify its forms, reducing the number of questions to about eight on the short form, down from 17 in 1990.
In addition to mailing the forms, the bureau for the first time will allow people to pick them up at government buildings and stores. Also, the bureau will set up a toll-free phone number to allow people to call in their responses.