Census 2000--Science Meets Politics

Abstract

T he Supreme Court has ruled that it is illegal to use statistical sampling to provide population counts for congressional reapportionment. Should scientists other than statisticians, demographers, and political scientists bother to follow the strange politics that have engulfed Census 2000? Yes, as a warning about how partisanship can affect a scientific agency. The census began as the result of the great political compromise that founded the United States. If congressional power was to be apportioned geographically (satisfying small states) and proportionate to size (satisfying large states), regular measurement of the number and distribution of residents was a necessity. In the 19th century, census results regulated the pace for admitting new states to the union, as partisan interests took turns vying for advantage. In 1920, the census measured a population shift from rural to urban states so alarming to Congress that it postponed for a decade its constitutional obligation to reapportion. But past politics focused mostly on how census results were to be used. Today's battle has been broadened to focus on how data will be collected. Technical decisions (for example, on how many census forms to mail) normally made by career professionals are today routinely reviewed by an oversight apparatus intent on influencing how the census will be conducted. Regrettably, the division of views falls along partisan lines. Even the Supreme Court divided along conservative and liberal lines in its recent decision. Science by partisan vote is a risky enterprise. The Census Bureau takes as much professional pride in accurately reporting how inaccurate it is as it does in trying to be completely accurate in the first place. We know that we miss people in our decennial count, and we know that this undercount is unevenly distributed demographically and geographically. This knowledge comes from our own post-census quality survey. The quality survey we recommend for Census 2000 is under attack because of its potential use for drawing legislative districts in 2001. Such use is an altogether appropriate issue for resolution in the political-legal arena. But other critical uses should not be compromised. To discard the quality survey would take away the science by which we assess our performance; we could not then tell the public whether census procedures in 2000 improved on those used in 1990. Planning for the next census would be crippled. Moreover, for at least a decade, statistical surveys of any kind would be based on an imperfect sampling frame and error-prone sampling weights. Just as there can be reasonable differences among lawyers about the interpretation of the census statute, there can be reasonable differences among politicians and among statisticians about the operational feasibility of the sampling design recommended by the bureau. But no statistician can accept the widespread misrepresentation of modern sampling methods. There can be no national statistics on unemployment, housing, transportation, health, or education without sampling. We can accommodate the Court ruling that Census 2000 is not to include statistical adjustment for the apportionment count, but without sampling the Census Bureau cannot fulfill its broad statistical responsibilities to policy-making, to business planning, and to scientific research. Also at issue is the damaging and unfounded claim that the bureau opted for sampling so it could then manipulate numbers for partisan purposes. To seek tactical advantage in a political dispute by questioning the integrity of a scientific agency is troubling. When national leaders say that the bureau cannot be trusted, they invite public doubt about other statistics, such as the consumer price index, poverty trends, unemployment rates, even measures of gross domestic product. When partisanship intrudes into the conduct of science, when widely accepted scientific methods are deliberately misrepresented, when scientific agencies are casually accused of dishonesty, a shadow falls across all of U.S. science.