The cliffhanger census: the 2000 census could be the best in American history, or the worst - Census Bureau proposes using different sampling methods

American Demographics, Jan, 1998 by Brad Edmondson

The 2000 census could be the best in American history, or the worst.

The Census Bureau wants to use sampling methods to account for millions of Americans who don't fill out census forms. Newt Gingrich and other Republican House leaders are fighting this plan with everything they've got. In public, they argue that the U.S. Constitution describes the census as an "actual enumeration"--a headcount, not a poll. In private, they worry that the sampling plan could add millions of people to the count for low-income neighborhoods. These people tend to vote Democratic, and this could hurt Republican chances when the boundaries of Congressional districts are re-drawn in 2002.

It's a clash between the art of politics and the science of statistics. Most statisticians agree that the Census Bureau's plan would produce a more accurate count. They estimate that the 1990 census, with no sampling, failed to include about 10 million Americans while it counted another 6 million twice or in the wrong location. Planners, marketers, and others whose jobs depend on accurate demographic information ought to be speaking out for the 2000 plan. But this battle has been raging for two years, and most census data users have said nothing. Why not?

Few people understand the complexities of the Census Bureau's plan well enough to defend them, while a headcount is easy to understand. And when you try to explain something as complex as census sampling, most people's eyes glaze over. Americans may love statistics, but they find statistical methods boring.

Even heavy users of demographic data are sometimes unaware that the Census Bureau is their ultimate source. Companies like Claritas and CACI make millions by selling current-year estimates and five-year projections of the population in small geographic areas. The decennial census is as important to these estimates as flour is to bread. But the customer sees Claritas and CACI as the source, not the Census Bureau.

Database marketing also hurts the census's political support. Many marketers think they no longer need demographic data. Instead, they buy lists that combine the customer records and mailing addresses of individual households. But Metromail, Polk, and the other companies that sell these lists enhance them and check their accuracy by comparing them to the census. A more accurate census would make database marketing more accurate, too.

A compromise version of the 2000 census plan squeaked through Congress in November, after a prolonged battle. In the spring, the bureau will "rehearse" one census that includes sampling in Sacramento and another census without sampling in rural South Carolina. Both plans will be watched by a monitoring board of four Democrats and four Republicans. Meanwhile, Republicans will challenge the sampling plan in court, at government expense.

If the 2000 census includes sampling, it could be the most accurate count in history. Without sampling, it would be among the worst. For more information, contact the Census 2000 Initiative; telephone (202) 434-8756, or e-mail Henry Griggs at hgriggs@ccmc.org.