Census takers grapple with the 2 to 4 million-people question. Cities and states don't want repeat of undercount

America is coming to its census. And while the decennial tally by the US Bureau of the Census is still more than two years off, already the battle lines are drawn over whether, and how, the 1990 census can be made more accurate than the 1980 count. At stake are enormous sums of money and considerable political clout. Federal and state governments give away tens of billions of dollars based on population data provided by the census, and some cities got less than what they felt was their fair share after the 1980 count.

Since seats in the United States House of Representatives are allocated on the basis of population, some states with large metropolitan areas felt they had to give up congressional seats unfairly. State and local election district boundaries also are redrawn according to census data.

After the 1980 figures were released, the Census Bureau faced more than 50 suits by cities and states alleging erroneous counts.

By its own reckoning, the bureau missed from 1 percent to 2.2 percent of the population in 1980 (that's about 2 million to 4 million people). It is believed that more than half of those missed were black. Hispanics, too, may have been significantly under-counted. And though the bureau cannot determine the whereabouts or economic status of those missed, other observers think it very likely most were poor people living in major cities.

What could be done? The Census Bureau began a lengthy search for ways to make its figures more accurate. Seven years later, in 1987, it proposed a ``post-enumeration survey,'' an after-census census that would involve a scientific sample of 300,000 households (out of a projected 106 million by 1990). Names from the sample households, it was assumed, should match those in the actual census. The names that did not match would indicate the degree of overall error, and the characteristics of those missed indicate error in counting specific groups. The census would be adjusted accordingly.

But late last October the US Department of Commerce, of which the Census Bureau is a part, announced that no such post-census would be taken.

Steven Feinberg, a professor of statistics at Carnegie-Mellon University who was on a National Academy of Sciences panel that reviewed and approved the methodology of the survey, says that decision was based on ``politics.'' He explains: ``Nobody who's a Republican expects to gain from the [proposed] adjustment,'' since inner-city blacks and Hispanics are likely to vote Democratic.

But Peter Bounpane, an associate director of the decennial census, says the survey was canceled because of ``technical difficulties.'' Not everyone at the bureau thought the technical hurdles to such a survey could be overcome, he says. And ``almost all'' felt that it could not be done in time. By law, the president must have the final count by Dec. 31, 1990.

There were other difficulties, says Mr. Bounpane. Taking the census is difficult enough without another ``huge'' enumeration. Changing the numbers through some mathematical methodology might erode confidence in the census and smack of ``manipulation'' to some. Creating two numbers might invite still more controversy: If the numbers had been adjusted once, people may reason, why not adjust them further?

``When we think we know we're making the census better, we'll make the adjustment,'' Bounpane says. ``At the time, we didn't think that was the case. Not everyone was in agreement that that was the case.''

Barbara Bialer, for one. The former associate director for statistical studies and methodology quit the Census Bureau in December, saying the decision to drop the post-census survey had been politically motivated. ``I have no documentation,'' she said in a telephone interview. ``It was personal - my own feeling about the decision.''

What bothered her, Ms. Bialer said, was the fact that the bureau ``had taken such great pains to have a procedure everyone was on board with,'' a procedure that was abruptly ``thrown out the window.'' The National Academy of Sciences panel had approved the adjustment methodology. And she herself had been persuaded that the procedure was a good one after having testified against adjusting the 1980 census numbers.

It is not too late to reverse the decision announced by the Commerce Department in October, says Bounpane, though ```too late' is coming up soon.'' A bill now in Congress would require that the adjustment be made.