Science 22 May 2009:
Vol. 324. no. 5930, pp. 1008 - 1009
DOI: 10.1126/science.324_1008

NEWS FOCUS
THE 2010 CENSUS:
Science/AAAS | Science Magazine: Sign In
Constance Holden

It is getting increasingly difficult to count a mobile population, but an "adjustment" based on postcensus sampling is politically unacceptable.


Face to face. The U.S. census didn't use the mails until 1970.
CREDIT: © THE METROPOLITAN MUSEUM OF ART
[Larger version of this image]

Next April, the United States will embark on by far the most expensive decennial census the country has ever seen: It will cost about $15 billion—double the cost of the 2000 census—to count an estimated 309 million U.S. residents. Billed as the country's "largest peacetime mobilization," it will send out 1.2 million temporary workers in early May to knock on the doors of about one-third of the population: the people who haven't sent in their census forms. Despite the best efforts of these workers, who may go back to the same addresses as many as six times, many people will be missed.

How to deal with this undercount has occupied statisticians for decades. It's also of perennial interest to politicians because the census is used to reapportion seats in the House of Representatives. The Obama Administration has repeatedly said it has no plans to conduct an "adjustment" in the 2010 census count, as many statisticians have advocated. But that hasn't stopped several House Republicans from agitating about such a possibility, which they fear would benefit Democrats, and they have accused the Administration of attempting to politicize the census.

The flames have been fanned as President Barack Obama tries to get his Commerce team in place. In February, alarm calls went out when a White House spokesperson seemed to say that Obama planned to have the census director report directly to the White House, bypassing the commerce secretary. That fuss is one reason Commerce nominee Judd Gregg, a Republican senator from New Hampshire, withdrew his name. Former census director Kenneth Prewitt, the top choice to head the bureau, dropped out soon afterward as the political heat rose.

The new census nominee, sociologist Robert Groves of the University of Michigan, Ann Arbor, has an impeccable reputation in the statistical community. But he has been greeted with suspicion by Republicans who identify him with past attempts to adjust, or correct, the census. He spent a good deal of this month, as well as part of his confirmation hearing on 15 May, reassuring legislators that an adjustment is not on the table and that it would be too late to try in the 2010 census even if he wanted to. "Any major change would bring so much risk that the benefits would have to be clearly established," he said.

The adjustment in question is also referred to as "sampling," which Republican critics say violates the Constitution's call to count every citizen rather than just sample the population. Many statisticians argue that casting the debate this way distracts from real concerns about the future of the census in a high-tech and increasingly mobile and diverse society. In fact, the technique in question does not substitute sampling for counting. Rather, it's a methodology for assessing the accuracy of an inevitably incomplete count.

Missing millions

The undercount problem was first recognized after the 1940 census, when the Defense Department was registering men for the draft. "More young black males registered for the draft than the census bureau thought were in the country," says historian Margo Anderson of the University of Wisconsin, Milwaukee. It's more of a political issue in the United States than it is elsewhere, she adds, because here it is used to determine how states are represented in Congress. Furthermore, as census figures came to be used for more and more purposes, such as federal revenue-sharing, there was "huge pressure on getting census figures right," says former House staffer David McMillen, now at the National Archives and Records Administration.

By the late 1980s, Republicans were worrying that any adjustments to the undercount would benefit Democrats because the people who end up not being counted—the poor and minorities—tend to vote Democratic. In 1990, for example, according to a 2004 report from the National Academy of Sciences (NAS), there were 16.3 million erroneous or duplicate counts and 20.3 million people weren't counted at all. Blacks—young black males in particular—were missed at a rate four times as high as that for whites that year, according to the U.S. Census Bureau. Affluent whites, on the other hand—such as people with multiple homes or students counted both by schools and by their parents—are disproportionately found in overcounts. In 2000, NAS says 15.9 million people were missed, whereas the overcount was 17.2 million.

By the time planning was under way for the 1990 census, adjustment had become a very contentious issue. The methodology in question is called dual system estimation (DSE). It uses a post-enumeration survey conducted independently from the census but very soon afterward. As former census director Prewitt explains, DSE is based on "capture-recapture" principles for counting wildlife. First you count as many as you can of a population and tag them (the census). Then in a separate operation (the post-enumeration survey), you capture a representative sample of the population and see how many have already been tagged (counted). From the ratio of tagged to newly captured, you can make a revised, and presumably more accurate, estimate of the total population.

In 1990, President George H. W. Bush's census director, Barbara Bryant, planned to use a post-enumeration survey of 300,000 households as the basis for an adjustment of the final census count. Census nominee Groves, who was associate director of statistical design from 1990 to 1992, favored the approach, as did most of the bureau's senior staff, according to McMillen. But Secretary of Commerce Robert Mosbacher Sr. quashed the plan.

The bureau again planned an adjustment for the 2000 census. This time it was challenged by a case that went to the Supreme Court in 1998. In the suit, Department of Commerce v. U.S. House of Representatives, Indiana Republican Representative Gary Hofmeister claimed that if the plan went forward, Indiana would lose a seat that would presumably go instead to a predominantly Democratic state. The court ruled that any adjustment of the final census count based on DSE could not be used for apportionment—that is, reallocation of the fixed number (435) of House seats among the states. It did not rule out DSE for other uses, including congressional redistricting (redrawing House district lines within states) and allocation of federal funds.

Using a post-enumeration survey as an accuracy check on the census is seen as an important way for the census bureau to evaluate its data, but to date it has not been used to adjust the final count. And it's not clear it ever will be. According to Hermann Habermann, deputy director of the U.S. Census Bureau from 2002 to 2006, statisticians inside and outside the bureau have said that there are too many possible flaws in the procedure to use it for that purpose. In 2003, the Census Bureau said there would be no adjustment for other purposes either, such as redistricting or the allocation of federal funds. Since then, officials have repeatedly stated that no adjustment is planned for the 2010 census. It's not even "technically feasible" at this late date, says Prewitt. The census "is a rocket that's on the launch pad, and they're about to ignite it. They can't redesign rocket fuel at this stage." Asked at his confirmation hearing about an adjustment in the future, Groves said, "I have no plans to do that for 2020." In his prepared material, he added, "I believe the Supreme Court ruling stands as the guidance on this issue."

None of this has stopped Republicans from fretting about the possibility that the procedure will be revived in 2010. As one House Republican staffer says, "If they wanted to, they could probably do it." In a statement last month, House Minority Leader John Boehner from Ohio accused Groves of having "advocated a scheme [in 1990] to ... manipulate Census data, rather than simply conducting an accurate count of the American people." Warned Boehner: "We will have to watch closely to ensure the 2010 census is conducted without attempting similar statistical sleight of hand."

A moving target

Aside from the adjustment issue, other improvements to the census process are sorely needed, say statisticians. There has been one major development: Next year, for the first time, citizens won't have to grapple with the Long Form, a lengthy questionnaire that used to go to one in six respondents; fewer and fewer people were bothering to fill it out. It's been replaced by the American Community Survey (ACS), a "rolling" survey that covers 300,000 people each month.

Although the ACS is generally hailed as an advance in obtaining thorough and timely information, it doesn't compensate for the fact that U.S. residents are getting ever harder to count. With traditional households few and far between, the census must count an increasingly mobile population; one where many people are suffering dislocation because of the economic recession or other factors; one with millions of illegal immigrants as well as others who prefer to avoid contact with the feds; one where even rich people can be hard to count because of the increase in hard-to-access gated communities.

With the long form out of the way, "it means you can think of doing the census in perhaps radically different ways," says statistician Stephen Fienberg of Carnegie Mellon University in Pittsburgh, Pennsylvania. The census is based on a system of physical addresses; its purpose is to locate an individual at "a particular geographic spot" at a particular time, says Anderson. But as the population becomes harder to track down, she says the postal system—used for the first time in the 1970 census—is already becoming "archaic" as a means to deliver the form. And modern modes of communication, the Internet and mobile phones, have the disconcerting feature of not being tied to any location. "The real question, I think, is, can you move to a person-based instead of a physical household census," says Fienberg.

As for reducing the undercount, the main strategy being explored is the use of "administrative records"—that is, tax data, Social Security numbers, driver's licenses, and even data from utilities to supplement census forms, an idea that's been around at least since a 1972 report, America's Uncounted People—the first of a long line of census reports from NAS. But heavy reliance on such a strategy is a long way off, not only because of privacy and security concerns but also because of the country's "decentralized statistical system," Anderson says.

Centralizing records such as driver's licenses would be in line with an even more radical step: a national registry. Prewitt is a proponent. In some other countries, particularly in northern Europe, traditional censuses have been eliminated, says Prewitt, as citizens are required to tell the government every time they move. "Without a national registry, we will seriously undercount," he says, pointing out that if we miss half the estimated 12 million immigrants now in the country illegally, that's 6 million right there.

Unfortunately, says Anderson, "unresolved debates" from past censuses "are obfuscating [and] drowning out other issues we should be talking about."