Kenneth Prewitt (New School University)
With an introduction by Sidney Verba (Harvard University)
Kenneth Prewitt has had several interlocking careers: as an investigator and teacher at Washington University in St. Louis, Stanford University, and the University of Chicago, and as one of the major social science administrators in this country—director of the National Opinion Research Center at the University of Chicago, twice president of the Social Science Research Council, and (between presidencies of the SSRC) senior vice president of the Rockefeller Foundation. More than anyone else, he is a person committed to the social sciences as an institution—to how they are organized, how they carry out their research, and how they serve this country. In fact, Ken is so smart about these matters that I was astonished when I heard that he had agreed to serve as director of the Bureau of the Census, a position made for losers.
The 1990 census had been very difficult. The 2000 census was faced with even more complicated issues and was already being attacked in Congress. I said to myself, "Perhaps he's been smart for a long time, but now he's finally lost it." Why did he accept the position? Ken is not only a committed social scientist; he is also a committed citizen who believes in serving the needs of society. One way to serve those needs was to make the 2000 census as sound and as objective as possible—and that is precisely what he did. He brought the census in more accurately than ever before, and he did it under budget. I am enormously grateful for what Ken has accomplished, and I believe that the nation should be as well.
The decennial census is the longest continuous scientific project in American history. It is also the largest applied social science project undertaken in this country. If you view it as a scientific project, then of course its importance is in what it tells us about our population and housing characteristics. But it is a misunderstanding of the first order if you treat the census primarily as a scientific project with a demographic payoff—because in America’s political history, the special status of the decennial census derives from its political purpose: its predetermined role in the reappointment of congressional seats, in the federal spending formula, and in the enforcement of civil rights laws.
The roots of its political purpose are well known. The decennial census is mandated by the Constitution so that the seats in the House of Representatives will be apportioned among the states of the union according to their respective numbers. The men who wrote our Constitution were superb political engineers. They may have borrowed their theoretical ideas, but it was their designated task to institutionalize solutions to the great problems of government bequeathed to them by political philosophy. Among the most challenging were the issues posed by federalism and colonialism. How could the distribution of powers under federalism protect local rights yet provide appropriate central authority? A bicameral legislature offered the compromise. In the Senate, each state would be equal in voting power, but in the House of Representatives, votes would be allocated in proportion to the population as counted in the census.
Why a census every ten years? To solve the problem of colonialism. Here was a new nation with vast territories that it intended to occupy. The restless population was already crossing the Appalachians, spreading westward into the Ohio Valley and down the Mississippi River to the Gulf of Mexico. There was even speculation about reaching the Pacific Ocean. What was to be the status of these soon-to-be-acquired territories? Would they be annexed as colonies of the original thirteen states or be accepted on an equal footing with them? Drawing from Montesquieu, the founders held that the republic could not also be a colonial power, and thus the territories would join the union as equal states. By measuring population growth and its geographic dispersion, the decennial census served to regulate the pace at which the western and southern territories were added as new states.
The census was political from the very beginning and remains so. Although the science of measurement is used to complete the task as accurately as possible, the central purpose of the census remains: to shift power from one set of interests to another. When the figures for Census 2000 were released, politicians and pundits alike opined that the western states would now get a more favorable hearing in the national government because of shifts in congressional representation. If the census is both political and scientific, how shall we accomplish the one without compromising the other?
To comment on this complicated issue, I want to begin with two shadows that fell over the Census Bureau after the 1990 census. First, that census was declared to be an operational failure; second, the bureau was criticized for devising a methodology that invited political tampering. If widely believed, the heavy charges of incompetence and corruptibility laid on the Bureau would compromise the credibility of census counts and, potentially, all other major government surveys. Two types of errors were cited: undercounting or missing people, and overcounting or counting some people twice. In this discussion, I will focus on what has become known as the "differential undercount."
An undercount of some magnitude has occurred in every US decennial census since 1790, when Thomas Jefferson conservatively estimated an undercount of about 200,000 on a base of 3.9 million. George Washington concurred that a combination of citizen resistance and flaws in the enumeration procedures produced an official count that fell far short of the true number of residents in the new country. In no decennial census since has the enumeration process been able to account for every resident in the United States, nor has it done so in any other nation. Demographers everywhere assume that a census is an approximation of the true count—perhaps an overestimation, but much more likely an underestimation. Insofar as the benefits of a census are allocated on a share basis—as is true if the benefit is a fixed number of congressional seats, now 435, or a fixed amount of federal funds—an undercount distributed equally across geographic units and population groups will result in equitable outcomes. Inequity emerges only if some areas or groups are counted at lower rates than others—that is, if the undercount is differential, which we know to be the case in the US census.
Although census professionals had long assumed that the undercount rate differed from one demographic group to another, it was not until World War II that the US had its first systematic measure of differential undercounting. In the early 1940s the government initiated mandatory, universal sel-ective service registration. Although obviously not its intention, this universal registration provided statisticians with two independent counts of males between the ages of 21 and 35: the count recorded in the 1940 census and the count of those registered for the military draft. Comparison of these counts provided the first reliable measure of how many persons, at least in that demographic group, had been missed in the census.
Save for one factor, this finding would have attracted little attention beyond demographers and statisticians concerned with improving census practice. What attracted wide interest was that African American males of draft age had been missed at much higher rates than white males. Here was the first systematic evidence of a differential undercount.
In subsequent years the Census Bureau has tried to deal with the undercount, but despite improved census practices, the differential has persisted. The first important national examination of the problem was a 1967 conference on "Social Statistics and the City," held by the Joint Center for Urban Studies of MIT and Harvard. That conference produced the by-then-obvious conclusion that "miscounting the population could unconstitutionally deny minorities political representation or protection under the Voting Rights Act. It could also deny local jurisdictions grant funds from federal programs." These themes are now commonplace in public discussion about census design issues, repeated in editorial pages, congressional testimony, scholarly studies, civil rights newsletters, and Census Bureau documents.
With significant issues of fairness and civil rights at stake, how could the Census Bureau respond? The recommended methodology is known as dual system estimation. It is similar to capture/recapture studies in wildlife research. For example, you can net trout in a pond, tag them, and put them back in the water. You can then repeat the process a day later and calculate, on the basis of how many previously tagged trout are recaptured in the second round, the total trout population. If one considers the census as the capture phrase, a random follow-up survey is equivalent to the recapture phase, permitting the calculation of undercount rates for many more population groups in greater geographic detail. Despite methodological challenges, the Census Bureau and panels of the National Aca-demy of Sciences and the American Statistical Association all agree that appropriate techniques are available for dual system estimation. The decision about whether to apply that methodology to Census 2000 has not yet been made.
Whether dual system estimation is implemented or not, the more important issue is what the two charges of incompetence and corruptibility meant for Census 2000. The first charge was partly self-inflicted because criticism from newspapers, the oversight committee of Congress, and others was accompanied by criticism from the Census Bureau itself. While the public accepted that judgment—and the concept of a "failed" census—the l990 census was not, in fact, a failure in any serious sense. It was and is being used for all its intended purposes, including reapportionment, redistributing, and the allocation of federal funds. One measure of its accuracy is that in 1990, as opposed to 1980, the census reduced overcounts in a number of areas.
There remains the charge of political corruptibility and its relation to the persistent undercount. In 1990 the Census Bureau recommended the use of dual system estimation to reduce the differential undercount. Rejecting the recommendation, Robert Mosbacher, then Secretary of Commerce, expressed the concern that adjustment would open the door to political tampering with the census in the future. This passage, as best I can determine, is the first instance in American political history in which a high government official gives voice to the speculation that the nonpartisan, professionally managed Census Bureau might choose a data-collection methodology so as to favor one political party over another.
Over the next several years, the conditional warning about what could happen turned into the accusation that the proposed methodology for Census 2000 was, in fact, designed to achieve a partisan outcome. The Republican National Committee declared, "The Clinton Administration is implementing a radical new way of taking the next census that effectively will add nearly four and one-half million Democrats to the nation's population." Political fights over the census have occurred throughout history, but they have been regional—between slave and non-slave states, between conservative southern and midwestern industrial states—and they have focused on the use of census results. In the 1990s the focus shifted to census methodology. In my view, both sides of the aisle now believe that the Census Bureau is capable of, and would design, a census that could have a known partisan outcome. For the past three or four years, every vote on the census in the US Congress has divided along party lines.
Because the 1990 census was thought to have been an operational failure, an oversight apparatus unprecedented in this country's fiscal history was put into place. It included nearly two hundred full-time employees drawn from four congressional committees, the General Accounting Office (an investigatory arm of Congress), the office of the Inspector General (an investigatory arm of the Executive Branch), and a specially appointed Census Monitoring Board, as well as a half-dozen major standing advisory committees, including one appointed by the National Academy of Sciences. Justified on operational grounds, the apparatus soon became a focus of political oversight as well. To deal with this scrutiny, the Census Bureau invited transparency and actively cooperated with these groups, providing information on what it planned to do, how decisions would be made, and the actual steps that were taken. Its efforts to respond took countless hours of senior management time, but they did blunt the charges of incompetence and partisanship.
Political fights about census methodology will continue to interfere with efforts to achieve a better census. We need more science in the process rather than less. For example, to fully apply dual system estimation, it would be useful to provide more time to correct the count than current law allows. The census calendar of nine months for apportionment counts and twelve for redistricting data was set in 1932, when the nation was much easier to measure. Despite the complexity of the technical process and the pending decision about whether adjustment is or is not going to be made, the Census Bureau must have redistricting data out by April 1. If the decision could be postponed for another six or nine months, the process could be made more precise.
In the most immediate sense, the Census Bureau must continue to demonstrate that its traditions and its competencies are not consistent with advancing a political agenda. Throughout its history, it has been resolutely professional and apolitical. Its peer community is the professional statistical community worldwide, and it earns respect in that community by providing accurate data. The process of internal deliberation and external transparency will continue through every step of the implementation of Census 2000 and will be available for public review. The bureau will take every step possible to demonstrate that, in its design and implementation, statistical adjustment is based on the best technical judgment available and never involves partisan consideration.
I want to conclude with an observation about the meaning of Census 2000. When its history is written, the issues surrounding sampling and other aspects of measurement theory will be a footnote—albeit an important footnote—to the real story of this count: multiracial identity. With "Question 8: What is this person's race? Mark one or more," we turned a corner about how we think about race in this country. Census 2000 identifies five discrete racial groups: white; African American, black, or Negro; Asian; Native Hawaiian or other Pacific Islander; and American Indian or Alaskan Native. It also allows respondents to select an "other" category, making a total of six. There are 63 possible combinations to how the race question can be answered. And if these 63 are subdivided by Hispanic and non-Hispanic groupings (which are treated by the census as ethnic rather than racial distinctions), there are 126 categories.
There is no way to measure race. Race is not a scientific construct but a political one. During the nineteenth century, the census counts helped put in place a discriminatory set of social policies. In the second half of the twentieth century, the census has been a tool to undo that discrimination. It is unlikely that more than a small percentage of the population will describe themselves as multiracial in Census 2000. But this expected change in self-identification has long-term and unpredictable consequences for race-conscious social policy. Laws prohibiting racial or ethnic discrimination in such areas as education, housing, and employment assume a small number of fixed racial or ethnic groups. With the proliferation of different multiracial groups in society and the general blurring of racial boundaries, the future of enforcing such laws in unclear. The task of the Census Bureau will be to provide the most accurate data for those who will determine the parameters of social justice.
Introduction © 2001 by Sidney Verba. Communication © 2001 by Kenneth Prewitt. Photo © 2001 by Mark Morelli.
This presentation was given at the 1841st Stated Meeting, held at the House of the Academy in Cambridge on January 10, 2001. Mr. Prewitt's remarks were based in part on his paper "The US Decennial Census: Political Questions and Scientific Answers" (Population and Development Review, March 2000).