Census 2000: An Overview
Kenneth Prewitt and Thomas A. Jones
U.S. Census Bureau
The decennial census is the longest continuous scientific project in U.S. history. It is also the largest applied social science project in our history. From an applied science perspective, the importance of the census is demographic--that is, how accurately it measures population and housing characteristics of the nation. But it is a misunderstanding of the first order to treat the census primarily as a scientific project with a demographic payoff. The special status of the decennial census in America's history derives from its political purposes--that is, its predetermined application in, especially, reapportionment and redistricting and to a lesser extent in federal formula spending and the enforcement of civil rights laws.
Although every decennial census is influenced by earlier censuses, especially operationally, to an unprecedented extent the 1990 Census set the stage for what has unfolded in 2000. The Census Bureau emerged from the 1990 Census under two shadows. First was the accusation that an important aspect of its recommended procedures could invite political tampering with the census counts(1). Second was the charge that the 1990 Census had been poorly conducted, was an operational failure. That these two charges were not based on evidence did not lessen their impact on the planning and execution of Census 2000. It necessarily became a goal of the Bureau to discredit both accusations. If either the charge of political manipulation or of operational failure was widely believed, the credibility of census counts would be seriously compromised. Could, then, the Census Bureau conduct the decennial census in a manner that erased the negative images that have shadowed it since the 1990 Census?
The Political Story
The decennial census is mandated by the Constitution, which stipulates that seats in the House of Representatives are to be "apportioned among the several States which may be included within this Union, according to their respective Numbers."(2) Of course something based on "respective numbers" requires a count; moreover, the count was to occur every ten years.
What was going on here? The accomplishment of those who wrote the Constitution was less in the originality of their political theories, which were largely borrowed, than in their state-building skills. Their challenge was to institutionalize solutions to the great problems of government that had occupied philosophy from ancient times. Two of these problems are of interest here: federalism and colonialism.
Federalism, though hardly a new political principle, had never been successfully institutionalized, at least on the scale envisioned for the United States of America. The problem to be solved by federalism was the distribution of powers in a manner that protected local rights and yet established a necessary degree of central authority. Part of the compromise solution was the bicameral legislature, in which for one branch of Congress the states would be assigned power proportionate to their respective population size. It was the census that made the "proportionate to size" principle workable.
But why a census every ten years? To solve the issue of colonialism. Theorists held that a republic could not also be a colonial power. Here, however, was a new nation with vast territories, rich in natural resources, that it intended to "colonize." What would the status of these soon-to-be-acquired territories be--would they be annexed as colonies or accepted on an equal footing with the original 13 states? Consistent with the principles of the new republic, new and equal states it was to be. The decennial census measured population growth and its geographic dispersion, thereby serving as the mechanism regulating the pace at which southern and western territories were added as new states. No sense can be made of the current census controversies without first appreciating that the census was designed as a political instrument to allocate power.
From 1790 through 1940, the census took place under the assumption that not everyone was included but without any systematic measure of the presumed undercount. After the 1940 Census researchers began to measure the undercount and quickly learned that it was differentially distributed across geographic areas and demographic groups. Attention focused primarily on Black-White differences (for reasons of data availability), and by the 1960s it was commonly assumed that there was a net undercount in the census that disfavored racial minorities.
The politics of race relations and the methodology of census-taking quickly converged with the 1965 Voting Rights Act and then with the steady expansion of federal formula spending for programs often targeted to those groups the census reported as undercounted.
With the stakes raised, the Census Bureau searched for a solution to the persistent differential undercount and concluded (with the support of many in the professional statistical community) that the best available alternative was dual system estimation. This would permit the Bureau to estimate on the basis of a census headcount, independently estimate on the basis of a subsequent sample, match the results, sort out the rate at which different groups were undercounted and overcounted, correct for these coverage errors, and report a new and more accurate count.
As most readers will know, dual system estimation, erroneously and misleadingly reduced to the term "sampling" in political debate, has been the subject of acrimonious and ill-informed partisan argument, budget games, presidential vetoes, and litigation that twice reached the Supreme Court--with the end not yet in sight.
In this political environment, what could the Census Bureau do? It could be as transparent as possible. In conducting the decennial census, for example, the Bureau prespecified its procedures, operations, and design choices far beyond normal practice and even beyond what was statistically prudent. From a political perspective, this prespecification increases congressional confidence that the design is without partisan political intent.
Transparency involves more than prespecification. It has involved cooperating with and even inviting a level of public scrutiny unprecedented in the agency's history, and probably unprecedented for any large-scale statistical operation. The Census Bureau made available a terabyte of real-time operational information; it provided briefings to congressional oversight committees and their staffs on nearly a weekly basis; it met frequently with a half-dozen advisory committees; it gave regular operational press briefings; and it was subjected to ongoing scrutiny by the Government Accounting Office of the Congress, by the Inspector General of its parent ministry, and by a special Census Monitoring Board that reported to the Congress and the Administration. Several hundred independent investigators, auditors, legislative staffers, and other overseers had access to all census operations.
Obviously, the level of oversight deflected management time and resources that otherwise would have focused on census operations themselves. As with prespecification, however, what might not make sense operationally was important in demonstrating that the census was conducted without partisan intent. That is, a statistical program that has critical partisan consequences must be designed and conducted in a manner to persuade partisan interests that it is politically neutral in its intent, even though its results are not neutral.
It is too early to know if the effort at transparency can erase the false charge that the Bureau would act in a partisan manner. As of this writing (January 2001) the basic apportionment counts have been reported (see below). Those counts, as required by the Supreme Court's statutory interpretation of the Census Bureau code, were generated without adjustment from dual system estimation, though they did involve a number of other statistical procedures, especially whole person imputation when a residence was known to be occupied but no census form was available.
The Census Bureau is now examining the quality of the census and the quality of the Accuracy and Coverage Evaluation (see below), and is applying dual system estimation. On the basis of this work, to be completed in late February, the Bureau will determine whether the accuracy of the census estimate used for apportionment can be improved. If so, the improved estimate will be used for three other critical Census Bureau products: the block-level counts used for drawing congressional district boundaries consistent with one person/one vote principles and for the enforcement of the Voting Rights Act; the census numbers used in the distribution of approximately $2 trillion in federal formula funds over the next 10 years; and the statistical controls that improve the accuracy of literally hundreds of sample surveys in the public and private sector.
A federal regulation specifies that senior career statisticians and demographers at the Census Bureau are to determine whether dual system estimation will improve the accuracy of the census numbers, with the final decision to be made by the Director. At this point there are two uncertainties. First, it is uncertain whether statistical adjustment will sufficiently improve the accuracy of the initial census counts to warrant its application. Second, it is uncertain whether the new administration will leave in place the arrangement under which the Census Bureau has the final say about adjustment. Then there is the more distant uncertainty of how the courts will rule if, as is likely, there is litigation no matter what is finally decided, by whomever. If the Bureau is allowed to render its best professional judgment, on the evidence then before it, science will have better served the nation.
[Editor's note: On February 16, 2001, the Bush administration transferred the power to adjust population figures produced by the 2000 Census to the Secretary of Commerce from the Census Bureau.]
The Operational Story
Census 2000 was a massive undertaking that involved years of planning (starting even before the 1990 Census was complete), testing, and preparations. Designing a census structure that would produce an accurate count of an estimated 275 million people living in households, in group quarters, or without a usual home was a complex challenge. Added to that challenge was the legal requirement to complete tabulations for use in apportioning the House of Representatives by December 31, 2000 (i.e., within nine months of Census Day), and counts for the states to use in redrawing congressional and legislative district lines by April 1, 2001.
The Census Bureau conducted a major census test in 1995 and a Dress Rehearsal Census in three locations in 1998, as well as a number of smaller tests throughout the decade. Major activities leading up to the census included determining content for both a short and long questionnaire, designing and printing the questionnaires and other forms, and establishing an infrastructure of 12 regional census centers, 520 local census offices, and 4 data processing offices. One of the key tasks was to compile some 120 million addresses--the Master Address File--used to determine whom to send questionnaires to and whether housing units had been accounted for in the census. One improvement for Census 2000 was to use the U.S. Postal Service address information throughout the decade to update our list. Another improvement was to provide state, local, and tribal governments an earlier and more effective opportunity to correct the Census 2000 address list for their areas. Congress passed legislation in 1994 to allow the Census Bureau to share its address list with these partners and to require the Postal Service to share its address information with the Census Bureau.
Building Support
In March 2000, the Census Bureau mailed or delivered questionnaires to each of the 120 million addresses on its list. Through a multi-faceted marketing program, the Census Bureau aggressively sought to encourage householders to complete and mail back their census forms. Since all addresses for which a questionnaire was not returned would have to be visited by census enumerators, good public cooperation would keep the nonresponse workload as small as possible, reduce the number of temporary enumerators needed, and reduce costs. Based on the experience of declining response rates over the preceding three censuses, the Census Bureau had anticipated that 61 percent of households would return forms in Census 2000.
Partnerships with state, local, and tribal governments, community and advocacy groups, the private sector, religious organizations, and educational institutions were key to building support and removing obstacles for the census. In all, the Census Bureau built over 141,000 partnerships for Census 2000. Paid advertising, designed to educate and motivate the public to respond and targeted to both a general audience and select population groups, was another important element of the census marketing program. The Census Bureau used paid advertising for the first time in Census 2000 and placed over $100 million in media buys. The Census Bureau also designed the questionnaires so that they would be easier to read and fill out and sent advance letters and reminder cards before and after the questionnaires were mailed out to increase response. Multiple ways to respond--the questionnaire, over the telephone, via the Internet, through "Be Counted" forms available at local sites, in English or other languages--gave the public more ways to include themselves in the census.
The Nation Responds
The aggressive marketing campaign paid off. Two-thirds of households answered the census, reversing the decades-long decline--a stubborn trend line--in response rates. The 67 percent final response rate was 6 percentage points higher than anticipated and 2 percentage points higher than the 1990 Census response rate. This notable achievement came despite declining involvement in community activities and despite organized complaints that the census long form was too intrusive. In fact, the response rate for the long form declined from 1990 and in Census 2000 was 11 percentage points below that for the short form; a higher increase in short form response rates more than made up for that.
The increase in response to the census was fueled largely by increases in minority communities, especially Hispanics. Thirteen of the largest 15 cities increased response. Over half of the states increased response over 1990. The higher than expected mail response rate meant that fewer housing units than expected would have to be personally visited during the next phase of the census--the nonresponse follow-up.
The Census Bureau began visiting 42 million addresses for which no questionnaire had been returned in late April and completed this critical operation in late June, slightly ahead of schedule. This nonresponse follow-up operation was the most serious operational challenge of the census because its success depended on hiring enough staff to conduct the work, on their meeting production goals, and on the public's willingness to open their doors and talk to the census enumerators. Because of our resourceful recruiting plan and full census funding that permitted us to offer attractive wages, we were able to recruit some 3.4 million job candidates and eventually hire 960,000 people over the course of the census; over 400,000 of these worked on the nonresponse follow-up operation. The census workers were dedicated, enthusiastic, and resourceful; they also braved tough neighborhoods and, in a few cases, tragic circumstances to get the job done. Because of the residual effects of our marketing program, the fact that we continued to advertise during the nonresponse follow-up, and the fact that we continued to involve our partners in census efforts, there was little outright hostility or resistance on the part of the public.
There were some concerns expressed that the Census Bureau had rushed to complete the nonresponse follow-up, but there were no facts to support that conclusion, and data showed that the Census Bureau did a good job of obtaining information directly from each housing unit. In only 3 percent of the cases did census enumerators, after having exhausted prespecified procedures to make up to six attempts to obtain an interview, get the information from another knowledgeable source, such as a neighbor or building manager. It is better to get information secondhand from knowledgeable sources than to get no information at all, and in most cases it is perfectly adequate information. In a very few instances, when procedures were not followed and we saw evidence of irregularities, we took appropriate corrective action.
Quality Counts Operations
After the completion of the nonresponse follow-up operation, the Census Bureau had accounted for every housing unit on its address list. We had conducted a "good census," completing every scheduled operation on time, achieving improved public response, having a successful hiring operation, and completing nonresponse follow-up within schedule. However, if we had stopped at the end of nonresponse follow-up activities, we would have provided an incomplete estimate of the population. Based on Census Bureau experience and using various quality indicators, the Census Bureau identified about 10 percent of the nation's housing units that we believed should be visited in a number of review, verification, and clean-up operations designed to improve coverage and the census estimate. We called these operations the "Quality Counts" program. The two largest operations were the coverage improvement follow-up (CIFU) and the coverage edit follow-up (CEFU). In CIFU, census workers went back to some housing units that had been identified as vacant by the enumerators in the earlier fieldwork and visited for the first time some addresses added too late to be included in earlier operations. In the CEFU, enumerators visited households that had more than six people (the census form only had room for six people) and households for which there was some question about the number of occupants.
Crunching the Numbers
By Labor Day, the Census Bureau had completed all field activities for Census 2000, including the Quality Counts program. That left four months to work with the huge data files from the census, running thousands of programs on them, to be able to release state population totals for apportionment by the legal deadline of December 31.
For more than a century, the Census Bureau has relied on technology to make the enormous task of tabulating the census numbers manageable. Herman Hollerith, a young mechanical engineer, developed the first Hollerith tabulating machine for use in tabulating the 1890 Census. The Hollerith tabulating machine and its descendants were used for the next 60 years until the Census Bureau shifted to the UNIVAC, the first commercial computer, which was developed to the Bureau's specifications for processing the 1950 Census.
For Census 2000, the Census Bureau used digital imaging and optical-character recognition technology for the first time to recognize handwritten answers instead of blackened circles. This was a vast improvement over tabulating machines and previous computer systems and allowed us to process the data faster and introduce a number of quality control features to be sure we had captured the data accurately. During the peak of questionnaire receipts, our data capture centers processed 3.3 million forms a day. Each bit of information on the captured census forms was sent over secured lines to the Census Bureau headquarters, where we performed quality control checks to ensure we had complete data and to allow us to remove duplicate questionnaires.
The Apportionment Counts
On December 28, 2000, the Census Bureau announced the state population totals for the purpose of apportioning seats in the House of Representatives and the number of seats each state will have beginning with the Congress that is elected in 2002. Earlier that same day, these apportionment counts had been delivered, as required, to the President, who, in turn, is required later to deliver them to the Congress. At the same time, we released the resident population of each state, the District of Columbia, and Puerto Rico. The apportionment counts include counts of federal civilian employees, U.S. military, and their dependents living abroad at the time of the census. The resident population for the states does not include these Americans overseas, and the District of Columbia and Puerto Rico are not part of the apportionment of seats in Congress.
As the Census Bureau reported on December 28, the resident population of the 50 states and the District of Columbia as of April 1, 2000, was 281,421,906. This represented an increase of 32.7 million during the 1990s, or 13.2 percent. The total resident population was more than 6 million above the most recent pre-census estimate (which was calibrated to the 1990 unadjusted census that did not include the measured net undercount of approximately 4 million).
The apportionment numbers are the first to be released in Census 2000 and the only ones that will be released before March 2000, when the Census Bureau will begin to deliver to states, on a flow basis, detailed small area data for use in redistricting.
The Accuracy and Coverage Evaluation (A.C.E.)
The resident population count of 281.4 million announced on December 28 is a net national estimate. It includes an as yet unknown level of two types of coverage error--persons missed in the census and persons erroneously counted twice. As indicated above, the method of assessing the magnitude of these two types of coverage error is dual system estimation. Following the initial census based on mailout/mailback and subsequent person follow-up of nonresponding households, the Bureau independently measured people living in 300,000 households on April 1, 2000. This very large survey was completed on schedule with a response rate of better than 99 percent.
The census plan calls for releasing redistricting numbers and all subsequent census data products in two versions: unadjusted and adjusted using the results of the Accuracy and Coverage Evaluation. One of these products will be denominated as the redistricting data--that is, the census results that the Bureau believes to be the most accurate. Census information will be available in a variety of formats and media, including the Internet, CD-ROMs, DVDs, and printed reports.
Concluding Note
What is surely most important about the 2000 decennial census are the data products--what we learn about ourselves as a nation and how that information is used to help us govern ourselves and to improve our economy and society.
But in getting to that point, the Census Bureau has had to navigate through complex political and operational waters. Census operations will be subjected to extensive evaluation studies, which will be reported over the next two years. These studies will reveal problems and mishaps as well as operational successes and achievements. It is premature to render final judgment, but it is likely that overall Census 2000 will prove to have been operationally robust--a "good census." If so, the shadow of the so-called 1990 failure will have been erased.
Whether the other shadow can be erased, the accusation that the Census Bureau would design a census to achieve a partisan goal, is also uncertain at this point. It is critically important that the scientific community join the Census Bureau in working to this end. Numbers that are believed to be politically rather than scientifically generated are numbers without public credibility. The nation's statistical system is far from having reached this point--and is, in fact, one of the most respected statistical systems in the world. But something eroded when that first charge was made in 1991 and then was amplified politically in the partisan battles over the next decade. We conclude this interim report with fingers crossed that Census 2000 will be accepted for the nonpartisan, scientific effort that it has in fact been.
Notes
Secretary of Commerce Robert Mosbacher, in 1991, ruled against the recommendation of the Census Bureau that the results of the 1990 count be adjusted to correct for the undercount measured by dual system estimation. His reasoning included the following passage: "...the choice of the adjustment method selected by the Bureau officials can make a difference in apportionment, and the political outcome of that choice can be known in advance. I am confident that political considerations played no role in the Census Bureau's choice of an adjustment model for the 1990 census. I am deeply concerned, however, that adjustment would open the door to political tampering with the census in the future."
The constitutional clause mandating the decennial census also anticipated that states would be taxed on the basis of size, a practice that fell into disuse early in the 19th century. James Madison, ever alert to the need to balance differing political interests, noted in Federalist No. 54 the benefit to an accurate census of attaching both representation and taxes to its count: "It is of great importance that the States should feel as little bias as possible to swell or reduce the amount of their numbers... By extending the rule to both [representation and taxation], ... the States will have opposite interests which will control and balance each other and produce the requisite impartiality." In this and in so many other areas, Madison was prescient. Now that the decennial census only conveys benefits and no penalties, there is a strong urge across thousands of jurisdictions to "swell" but never "reduce the amount of their numbers."