2000 CENSUS TAKING SHAPE WITH BATTLE OVER NEW METHOD.

Byline: Steven A. Holmes The New York Times

One neighborhood is urban, largely African-American and poor. The other is suburban, virtually all white, and middle and upper class. But they represent two sides of the same coin, the yin and the yang of the problems the Census Bureau experiences in trying to carry out a modern-day count of the United States' population.

Statisticians estimate that in 1990 the Census Bureau counted too few residents on the Hill, the low-income neighborhood on the south side of this state capital, while it tallied too many in New Cumberland, a small bedroom community across the Susquehanna River from Harrisburg.

Officials acknowledged the problems in the last census, when nearly 10 million people were missed nationwide and 4.4 million were counted twice or were made up by frustrated census takers who could not gain access to homes, according to the General Accounting Office, the investigative arm of Congress.

And the Census Bureau has a remedy for the next scheduled head count, in 2000. It is called sampling and is a method similar to that used in public opinion polls; it extrapolates the characteristics of a large group by surveying a representative part of it.

If the bureau can use sampling - a plan strongly opposed by Republican leaders in Congress - it may avoid the problems that stem from situations like that of Cory Layton, who lives on the Hill. Layton, who is 20 and married with two children, has not been getting along with his wife, Angel, who is 19. So he moved out of Hall Manor, the public housing project where they lived, and in with his mother, who lives in another part of the city.

If a census form were to arrive in the mail today, no one is sure who would list Layton. His wife said she definitely would not include him as living with her. And he strongly doubted that his mother would indicate he lived with her, if she bothered to fill out a census form at all.

``I don't care'' about being listed, Layton said, standing shirtless on a hot summer afternoon in the doorway of his wife's apartment. ``It's not affecting me.''

Layton might not believe that such an oversight affects him directly, but if the Census Bureau could correct for it through sampling, the resulting revisions in its 60,000 census tracts could have a high-stakes affect on the national political map. The census is used to apportion seats in the House of Representatives and to draw congressional and state legislative districts across the land.

And there is the rub. To Republicans in Congress, sampling has become a dirty word. They say it invites manipulation of figures for partisan purposes and insist that the Constitution requires an actual count of every resident.

What those Republicans rarely add is that many of those undercounted tracts, like the Hill in Harrisburg, are overwhelmingly Democratic and many tracts that were overcounted like the one in New Cumberland are solidly Republican. As a result, Republicans fear that adjustments for undercounts and overcounts could cost them congressional seats.

The Census Bureau said the problems that occurred in 1990 might get worse if Congress, as it has threatened, forbade the bureau to supplement its count with estimates based on samples of the population.

``Every indication since 1990 suggests that the census-taking environment is likely to be even more difficult in 2000 than it was in 1990,'' the bureau wrote in a report to Congress last month. The difficulties stem from a rise in the numbers of two-income couples, who are often too busy to fill out the forms; immigrants with limited use of English; and college students, who often are counted twice.

A plan by the Republicans to derail sampling earlier this year was put aside because it was stalling emergency flood relief for Midwestern states. But when Congress returns from its summer recess after Labor Day, the Republican leadership will revive the issue through a provision in an appropriations bill that forbids the Census Bureau to spend any money on preparations for using sampling in the 2000 census.

Census officials say sampling is needed to help smooth out the discrepancies it has found in places like Harrisburg. Using mathematical models of the types of communities where miscounts are likely, bureau statisticians estimate that the official count of 5,658 people in the census tract on the Hill was short by about 5.4 percent. In contrast, the 3,991 people officially listed in the New Cumberland tract was reckoned to be about 2 percent too many.

Layton is emblematic of the people unlikely to be counted: young inner-city males, especially African-Americans and Latinos, with only a tenuous hold on any permanent residence. In 1990 an estimated 12.7 percent of African-American men ages 25 to 29 were not counted, the highest undercounting of any group in the country.