Although it is a "no brainer" for "most educated people" that using sampling for the U.S. census in 2000 would produce a better count, you don't have to be a "Luddite or Republican" -- says Lawrence Osborne, in the New Republic -- "to wonder whether sampling is all it's cracked up to be."

• Mailing out forms to households with some immediate follow-ups should successfully count about 90 percent of the population.

• But instead of sending out more interviewers to find the missing people, sampling would extrapolate from the actual count, and select 750,000 households at random to compare with and adjust the count.

• The 1990 census did not count an estimated four million people, about 1.6 percent of the population, and a disproportionate percentage of the total were minorities, say sampling proponents.

While sampling would save money, and might marginally improve the total population count, "errors introduced by sampling nonrespondents" would overshadow the benefits if it is applied at the census tract level, according to a 1997 General Accounting Office report.

In other words, says Osborne, it cannot tell us where everyone is or in what density -- precisely the information that affects drawing electoral district boundaries and disbursing federal money.

Sampling's margin of error for a given block may reach 10 to 15 percent, according to one member of the Census 2000 Advisory Panel; thus a block with 100 residents might be determined to have a population of anywhere from 85 to 115, a difference of up to 30 people.

Thus sampling, which was originally proposed merely as an evaluation tool, is not sufficiently reliable to make it part of the census count.

Source: Lawrence Osborne, "Sample Error," New Republic, February 1, 1999.

For text of GAO Report GGD-97-142, July 14, 1997, "2000 Census: Progress Made on Design, but Risks Remain."