Truths, Half-Truths, And the Census; In Describing Us, the Count Has Its Limits

Every 10 years, the United States Census Bureau unleashes a swarm of numbers that affect everything from the political power of states to the municipal budgets of cities and the clout of ethnic groups. And every 10 years, anyone who knows anything about how those precise-sounding numbers are arrived at knows there is no way they can be absolutely true.

Yet the numbers are put to every imaginable use: for bragging, belittling, planning bus routes, closing schools, marketing fast food. The farther the numbers migrate from the bureau’s offices, the less anyone lets on that the census is no more than what one former bureau director calls it: an estimate of the truth.

”Anyone who’s producing these numbers knows exactly all the assumptions, all the uncertainty,” said Wendy N. Espeland, a sociologist at Northwestern University who studies commensuration, the turning of things into numbers. ”They know where the bodies are buried. But the way numbers travel, they leave the people who produced them and they get harder and more real the farther away they go.”

The shortcomings of census data were brought home last week when the bureau released its count of the many Hispanic groups in New York City. The bureau appeared to have underestimated the number of Dominicans, Colombians and perhaps others as a result of rewording a question on the 2000 census form.

Meanwhile, census officials were scrambling to fend off interest in a seeming surge in the nationwide number of gay and lesbian households, based on comparisons of 1990 and 2000 census data. The numbers were not actually comparable, the officials hastened to point out, because the bureau changed the procedures it uses to ”edit” the data.

And in March, when the bureau announced that New York City’s population had topped eight million, half the apparent growth turned out to be the product of better counting. New York, unlike many places, had gone to the trouble and expense of updating the bureau’s address list and had found several hundred thousand missing households. Many had probably been there all along.

”The census is actually probably precise to the millions,” said Theodore M. Porter, a professor of history of science at the University of California at Los Angeles and author of ”Trust in Numbers: The Pursuit of Objectivity in Science and Public Life” (Princeton, 1995). ”There are about six digits, maybe even seven, that are in some sense meaningless.”

The census is a gargantuan undertaking. It tries to take stock of what Margo Anderson, a population historian at the University of Wisconsin at Milwaukee, calls ”one of the most diverse and dynamic populations in the history of the world” — a moving target of several hundred million people, many arriving and departing, some speaking no English, some with no known address, others with two or three.

If it is impossible, as demographers say, to count the national population to within a margin of error of one or two percentage points, the possibility of error is even greater for New York City, with its large immigrant population and minority communities; its high housing costs, which cause illegal units and doubling-up; and its abundance of nontraditional households.

The census itself is what Professor Anderson calls ”a blunt instrument.” The field work is done by temporary workers, not trained survey researchers. The forms are filled out hastily at home. Though the questions are intended to be simple and clear, people still misunderstand, make mistakes. And when they are asked to write in an answer, rather than put an X in a box, things become even worse.

In the 2000 census, people who checked the box marked ”some other race” were then asked to write in which one. Answers included Bolivian, Bushwacker, Cosmopolitan, Aryan, Jackson White.

Several censuses ago, the bureau asked about ancestry right after asking a question about ability to speak English. The number of people reporting English ancestry came out suspiciously high. ”If you’re asked if you speak English and you say, ‘Yeah, I speak English pretty well,’ and you don’t know exactly what ‘ancestry’ is, and you’ve just been reminded that you speak English, you put down English as your ancestry,” said Reynolds Farley, a sociologist at the University of Michigan. ”Particularly if you’re in a hurry. It looks as though that is what happened.”

Some simple questions are not simple to answer. What is the regular place of residence for a college student or a person in a commuter marriage? On race, not everyone shares the government’s definitions. Eugene Ericksen, a professor of sociology and statistics at Temple University who has studied census errors, said as much as 25 percent of people asked to answer a series of questions about their income leave out some or all of the answers.

Faced with incomplete responses, the bureau sometimes resorts to what is called imputation: by looking at similar people with similar characteristics, it comes up with a plausible response. In areas with high refusal and low return rates, it has been known to happen that nearly everyone in a tract has ended up with the same job and distance traveled to work.

In 1990, a person who shared a household with someone of the same sex and also reported being married posed a problem because the bureau’s system did not recognize same-sex marriages, said Robert Kominski, an assistant chief in the bureau’s population division. To make the responses consistent, he said, the bureau changed either the person’s sex or his or her relationship to the other person.

The most problematic census data concern census blocks, the smallest areas for which the bureau releases numbers, anywhere from a handful to a few hundred people. To protect the confidentiality of people in those areas, the bureau swaps some of their characteristics — say, their race and age — with those of people in an adjoining census block, said Kenneth Prewitt, a former director of the Census Bureau and now dean of the graduate faculty at New School University.

When blocks are added up into tracts, the picture is accurate. But the fact that there are so many potential errors in small-area data — used in redistricting, by local governments and by business — was one reason that Mr. Prewitt proposed to the Office of Statistical Policy at the Office of Management and Budget last week that the government consider no longer releasing block-level decennial census data.

”It would create a higher level of statistical sophistication to recognize that we shouldn’t be making public policy on the basis of data that the agency itself knows to have high error margins,” he said.

If Americans take census numbers seriously, it is understandable. The count has financial implications and an immediate effect on American political life. Professor Anderson likens it to an election: ”We count the population in April, report it in December, then we move political seats around. So we take away political power from certain pieces of geography and give it to others.”

But census data users want precision. They want to hear that 37 percent of the population has a particular characteristic — when the true number is that number, give or take some error. They do not want estimates. How many representatives would New York City get in Albany, Mr. Farley wondered, if its population count was ”somewhere between 7.8 and 8 million?”

Mr. Prewitt said, ”If we said, ‘Between 34 and 40 percent of the American public have the following characteristics,’ the people who want it to be the higher number would say 40 percent and the people who want it to be lower would say 34. You would have a whole politics built up around the two ends of the distribution.”

Mr. Prewitt wistfully suggests a nationwide numeracy campaign. The country talks about improving literacy, he says. But most of the public conversation is about numbers: statistics, trend lines, social indicators. Perhaps the country should take numeracy as seriously as literacy if it wants intelligent public discourse.

”Using census numbers, are you better off than if you didn’t have them?” he said. ”Yes. You either argue about numbers or you argue about anecdotes. And I think the public discourse is better when the argument is about numbers. But we should be sophisticated enough to understand that the number itself is an approximation.”

He added, ”I say the census is an estimate of the truth. There is a truth; the census is an estimate. It happens to be the best estimate that we have. It’s not as if you could come up with a better one. You’d better go with ours.”