Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 71
The 2000 Census: Counting Under Adversity CHAPTER 3 The Road to 2000 CONFLICT BETWEEN THE CENSUS BUREAU’S PLANNING process and concerns of key stakeholders, which spanned much of the decade of the 1990s, put the design and successful execution of the 2000 census in jeopardy. In February 1997, the U.S. General Accounting Office added the 2000 census to its list of 25 high-risk activities of the federal government (U.S. General Accounting Office, 1997). The general shape of the 2000 census remained in flux until a 1999 U.S. Supreme Court ruling forced final changes in plan. It is instructive to review the evolution of the 2000 census plan in order to realize the unprecedented difficulties that confronted the Census Bureau staff in 2000.1 In this chapter, we review the basic operations and major problems of the 1990 census (3-A). We then outline the major research programs conducted during the 1990s, including work on questionnaire design and address list development (3-B), and complete our history by describing the extended struggle to finalize the design of the 2000 census (3-C). We close the chapter with a fundamental finding and two recommendations that flow directly from the evolution of the 2000 census plan (3-D). 1 Key sources for this history include: Anderson and Fienberg (1999, 2001); McMillen (1998); National Research Council (1995b, 1997, 1999b); Bureau of the Census (1997); U.S. Census Bureau (1998, 1999b); U.S. General Accounting Office (1992, 2002c).
OCR for page 72
The 2000 Census: Counting Under Adversity 3–A THE 1990 EXPERIENCE The natural point to begin the story of one decennial census is to look back at the census that preceded it. Thus, we begin by summarizing the major census operations in 1990 and describing the problems that census experienced. These problems drove the research and planning agenda for 2000. 3–A.1 Major Operations The 1990 census, broadly similar in structure to the 1970 and 1980 censuses, involved 10 major operations (see also Appendix C). Set up an organizational structure. The census structure for 1990 (see Bureau of the Census, 1995b:Ch.1,8) included: staff at Census Bureau headquarters in Suitland, Maryland (over a dozen units handled budget, overall management, data processing, and other functions, five of which were directly under the associate director for decennial census, a new position in 1990); a back-up computer center in Charlotte, North Carolina; the Bureau’s Data Preparation Division (later renamed the National Processing Center) in Jeffersonville, Indiana, which handled such operations as coding of place of work and served as a processing office; 13 regional census centers and the Puerto Rico Area Office, which managed the local field offices; seven processing offices (including the Jeffersonville facility), which handled data capture and also, for large central cities, computerized data editing and telephone follow-up for missing people and content; and 449 district offices (plus 9 offices in Puerto Rico), which handled address listing, outreach, nonresponse follow-up, and, in most areas, clerical review and follow-up for missing people and content. Develop an address list. Construction of the 1990 Address Control File (ACF) began in 1988 using only a few sources. In urban
OCR for page 73
The 2000 Census: Counting Under Adversity areas two lists purchased from vendors were field checked and supplemented by a field listing operation. In rural areas a field-generated list was updated by enumerators as they delivered questionnaires. A separate address list was developed for dormitories, nursing homes, and other special places. The Census Bureau’s innovative TIGER system was used to assign census geographic area codes to ACF addresses in mailout/mailback areas and to produce maps.2 Design short-form and long-form questionnaires and mailing materials. In 1990, in response to an administration effort to reduce the size and scope of the short and long forms (see Choldin, 1994), several housing items were moved from the short to the long form. The questionnaire format met the requirements of the data capture technology, but the questionnaires were unattractive and hard to read. The only additional mailing was a reminder postcard. Mail or hand deliver appropriate questionnaires to addresses on the address list. In mailout/mailback areas—84 percent of total addresses in 1990—Postal Service carriers delivered the questionnaires. In update/leave/mailback areas—11 percent of addresses—Census Bureau enumerators dropped off questionnaires and updated the ACF at the same time. (Update/leave was a new procedure in 1990 designed for areas that largely lacked city-style addresses—street name and house number—for postal delivery.) Census staff enumerated the remaining 5 percent of the household population in person. Separate enumeration procedures were used for such special populations as homeless people, residents of group quarters, and transients. Carry out advertising and outreach campaigns. This was intended to boost mail response and follow-up cooperation. The 1990 advertising campaign was conducted on a pro bono basis by a firm selected by the Advertising Council. Send enumerators to follow up addresses that failed to report. The nonresponse follow-up (NRFU) operation for 1990 began at the end of April and continued through early August. Subsequent follow-up operations rechecked the classification of virtually all 2 TIGER stands for Topologically Integrated Geographic Encoding and Referencing; see Glossary and Section 4-E.5.
OCR for page 74
The 2000 Census: Counting Under Adversity vacant and nonresidential addresses, relisted and reenumerated households in selected areas believed to be not well covered, and checked the enumeration status of people on parole or probation. Telephone calls and personal visits were also used to obtain responses for missing content items. These follow-up operations were central to the 1990 design, which emphasized the need to reduce the net undercount of the population and, especially, the difference between net undercount rates for minority and majority groups, by conducting as thorough a field enumeration as possible. In part because the Census Bureau took a position in court cases against statistical adjustment of the 1980 census for net undercount, the 1990 design maximized the use of telephone and field follow-up and minimized the use of computer-based statistical techniques to impute housing units and people for addresses of uncertain status. The Bureau was concerned that such imputation might be viewed as adjustment (see Chapter 6). The 1990 design also incorporated extensive follow-up for missing data content. Process the basic information through the steps of data capture, identification and removal of duplicates, editing and imputation, tabulation, and data release. Most questionnaires were routed to local offices for clerical review to identify cases for telephone or in-person follow-up for missing people or items. The completed questionnaires were sent to a processing office and the data captured by an in-house microfilm-to-computer optical scanning operation (write-in responses were keyed by clerks from the paper questionnaires or microfilm images). Computerized routines were used at headquarters for imputation of remaining missing data and preparing data files for dissemination. Process the long-form-sample information through the steps of data capture, weighting, coding, editing and imputation, tabulation, and data release. Data processing for the long-form sample was similar to that for the basic items on all records, except that item imputation routines were more complex to handle the greater number of long-form-sample items, and weighting was carried out so that estimates from the long-form sample would represent the entire population.
OCR for page 75
The 2000 Census: Counting Under Adversity Conduct coverage evaluation operations. In 1990, demographic analysis estimates of the population were constructed for age, race, and sex groups at the national level, and a Post-Enumeration Survey (PES) Program was used to develop estimates for additional groups and subnational areas for comparison with the enumerated census counts. The 1990 PES estimated a net undercount of the population of 1.6 percent, which was consistent with the demographic analysis estimate of a 1.9 percent net undercount (revised in October 2001 to 1.7 percent). The Department of Commerce decided not to use the PES results to adjust the 1990 census counts. Conduct experiments and evaluations. The 1990 census included a few experiments conducted during the census and evaluations of many aspects of procedures and results. 3–A.2 Major Problems The 1990 census had important technological achievements, including a computerized address file, the development of the TIGER system for producing maps and coding addresses to census geography, computerized tracking and control of questionnaires, and concurrent processing in which the information was captured and processed on a flow basis (see Bryant, 2000). However, three major problems of the 1990 census captured public, congressional, and stakeholder attention and motivated planning for the 2000 census. They were: (1) unexpectedly low mail response rates; (2) increased costs fueled by the need to hire additional nonresponse follow-up enumerators and stretch out the time for follow-up; and (3) evidence of somewhat worse coverage of the population, particularly of minority groups and children, in comparison with 1980. Poorer Mail Response Critical to the efficiency, cost, and quality of a mailout/mailback census is a high level of cooperation from the public in mailing back forms, since that reduces the workload for nonresponse follow-up and provides more complete responses. The Census Bureau uses the mail response rate to estimate the nonresponse follow-up workload. It is the number of mailed-back questionnaires divided
OCR for page 76
The 2000 Census: Counting Under Adversity by the total number of addresses to which Postal Service carriers and census enumerators delivered questionnaires for mailback.3 In planning the 1990 follow-up effort, the Census Bureau projected a mail response rate of 70 percent (down from 75 percent in 1980), but the actual rate at the time nonresponse follow-up began was 63 percent (the rate subsequently rose to 65 percent). A difficult-to read form and an inadequate publicity campaign were among the factors believed responsible. To cope with the unexpectedly low response, the Bureau sought additional appropriations, scrambled to hire sufficient workers for follow-up activities, and took steps to increase productivity (e.g., raising pay rates in selected areas). The nonresponse follow-up operation had been planned to take 6 weeks from when it began in late April; however, the workload took over twice as long to complete, not finishing until the end of July (Bureau of the Census, 1993:6-34,6-36). Subsequent field work for some planned and ad hoc operations also experienced delays. From reviewing housing unit counts by block in the Postcensus Local Review Program, large cities and other jurisdictions claimed that the Bureau had missed significant numbers of housing units. From July through November, the Bureau revisited about 20 million housing units (20 percent of the housing stock) in selected blocks in high-growth areas and for which local governments reported higher counts than the census. In 31 local offices, some or all households were reenumerated because of allegations that enumerators had fabricated data. A “Were You Counted” campaign (similar to the “Be Counted” effort in 2000) extended from June through November. A last-minute special program was implemented to improve the coverage of people on parole or probation, but the operation had low response, was not completed until early December, and, according to subsequent analysis, included a high proportion of erroneous enumerations. Some observers (e.g. Ericksen et al., 1991) concluded that the 1990 census was poorly executed overall. Higher Costs In inflation-adjusted fiscal 2000 dollars, the cost of the census rose from $24 per household in 1980 to $32 per household in 1990. (The 3 See Glossary and Box 4.1 for the distinction from the mail return rate.
OCR for page 77
The 2000 Census: Counting Under Adversity census had cost only $13 per household in 1970; see U.S. General Accounting Office, 2001b:6, and Table 3.1 below.) Part of the additional cost in 1990 was due to the unanticipated additional workload for nonresponse follow-up and other field operations. Worse Coverage The additional money and effort for the 1990 census did not buy improved coverage: the net undercount, measured by demographic analysis, increased from 1.2 percent of the population (2.8 million people) in 1980 to 1.9 percent of the population (4.7 million people) in 1990. Net undercount rates for blacks increased from 4.5 percent in 1980 to 5.7 percent in 1990; net undercount rates for nonblacks increased from 0.8 percent in 1980 to 1.3 percent in 1990 (National Research Council, 1995b:Table 2-1).4 Because net undercount rates increased more for blacks than others, the difference in coverage rates for blacks and nonblacks widened from 3.7 percentage points in 1980 to 4.4 percentage points in 1990, reversing the historical trend toward declining net undercount and the narrowing of differences among important groups. The October 2001 revised demographic analysis estimates for 1990 do not change this story—the revisions lowered the total net undercount rate from 1.9 percent to 1.7 percent, the black net undercount rate from 5.7 to 5.5 percent, and the nonblack net undercount rate from 1.3 to 1.1 percent, leaving the difference between the black and nonblack net undercount rates unchanged at 4.4 percentage points (Robinson, 2001b). 3–B RESEARCH AND DEVELOPMENT TOWARD 2000 The perceived failures of the 1990 census—low public response, high costs, and worse coverage—drove planning for the 2000 census, which began in late 1991 accompanied by oversight from an unprecedented array of outside groups (see Section 3-C.1). The Bureau devoted significant attention and resources to research in four major areas: (1) how to make the census address list more complete 4 Net undercount estimates for 1990 from demographic analysis differ somewhat from those from the PES; see Chapter 5. The demographic analysis estimates are used for 1990 for comparison with 1980 for which no accepted PES results are available (see National Research Council, 1985:Ch.4).
OCR for page 78
The 2000 Census: Counting Under Adversity by taking advantage of prior work by the Bureau and other organizations; (2) how to make the questionnaire more attractive and the mailing materials more compelling in order to bolster response; (3) how to streamline nonresponse follow-up to reduce costs and time and possibly reduce coverage errors; and (4) how to implement a coverage evaluation survey of a size and on a schedule to permit the results to be used to adjust the census counts for reapportionment, redistricting, and other uses of the data. Another line of research—on the possible uses of administrative records (e.g., tax records, Social Security records) to assist in the census enumeration—proceeded to the point of assembling administrative records databases for several sites and testing their completeness in a 1995 census test. However, administrative records research was dropped except for an experiment as part of the 2000 census (see National Research Council, 2000a). 3–B.1 Address List Research: Development of the Master Address File Almost before the 1990 census data collection was completed, the director of the Census Bureau endorsed the concept of maintaining a Master Address File (MAF) on a continuous basis from one census to the next. The expectation was that a continuous updating process would, in the end, be cheaper and would produce a more accurate and complete address list than the process used since 1970 of building a new list from scratch each decade a few years prior to the census. The importance of an accurate MAF was underscored by research estimating that two-thirds of the people missed in the 1990 census were in housing units or structures that were missed, presumably because of problems in the ACF; the other one-third of missed people were members of enumerated households (Childers, 1993). In urban areas two sources were expected to provide the bulk of the updates to the 1990 census address list: (1) the U.S. Postal Service Delivery Sequence File (DSF), which is the master list of addresses for mail delivery maintained by the Postal Service, organized by carrier route; and (2) input from local governments from reviewing the address list for their area. Field work in selected areas would be used to spot check the completeness of the list. In rural areas
OCR for page 79
The 2000 Census: Counting Under Adversity lacking city-style addresses a field listing operation was planned, supplemented by local government review. This MAF development concept quickly won wide acceptance (see, e.g., Commerce Secretary’s 2000 Census Advisory Committee, 1995; National Research Council, 1994, 1995b). Because of confidentiality constraints on sharing address lists, legislation was needed to implement the concept. The 1994 Census Address List Improvement Act (P.L. 103-430) authorized the Postal Service to share the DSF with the Census Bureau. It also set up a process whereby the Census Bureau was required to allow interested local governments to review the addresses for their area, swearing in local officials as census agents for this purpose. (Local review efforts in 1990 and 1980 limited participating localities to review of preliminary housing unit counts by block for their areas, not individual addresses.) At 6-month intervals beginning in 1995 (later than originally hoped), the Bureau received the latest DSF, which included newly constructed units, to match to and update the MAF. However, internal Census Bureau evaluation suggested that the DSF files were not updated at the same rate across all areas of the country, missed many addresses for new construction, and often did not identify individual addresses in multiunit structures. Moreover, planning to involve local governments in address review fell behind schedule. The Bureau developed a Program for Address List Supplementation (PALS), which invited local governments to send address files to the Bureau for matching to and updating the MAF, but participation was low and files arrived in such variable formats (including paper) that they were not usable. Hence, in September 1997 the Bureau announced a revised and more costly plan for producing a 2000 MAF of acceptable quality. Instead of conducting field checks of the MAF in selected areas, the new plan called for blanket field canvassing operations in 1998 and 1999 similar to those carried out for the 1990 census. Local governments in urban areas would be given the opportunity to review MAF addresses for their jurisdiction in the 1998 Local Update of Census Addresses (LUCA 98) Program; governments in rural areas would be given housing unit counts by block to review in the 1999 Local Update of Census Addresses (LUCA 99) Program.
OCR for page 80
The 2000 Census: Counting Under Adversity 3–B.2 Questionnaire and Mailing Package Research Questionnaire research had high priority in the early 1990s because of the perception that the format and length of the 1990 questionnaires, particularly the long form, had adversely affected response. Mail response rates fell slightly between 1970 and 1980 (78 to 75 percent) and dropped sharply between 1980 and 1990 (75 to 65 percent). To control the number of questions, the Census Bureau decided that items could not be asked of everyone unless block-level counts for the items were mandated in legislation or federal regulations and should not be asked of those in the long-form sample unless legal or regulatory authority could be cited for those data for census tracts or larger geographic areas.5 Consequently, most housing items that were complete-count items in 1990 were relegated to the 2000 long-form sample or dropped; marital status was also moved to the long form. Ultimately, the 2000 short form had only five population items and one housing item, compared with six population items and seven housing items in 1990 (see Appendix B). The long form had both additions and deletions. Some 1990 long-form items were dropped entirely for 2000, but Congress mandated a new set of long-form questions on responsibility for care of grandchildren in the home. On balance, the 1990 and 2000 long forms included about the same number of questions, but the 2000 long form (and short form) provided space for characteristics of only six instead of seven household members as in 1990. Additional household members’ names could be listed but not their characteristics. Mail returns that indicated more than six persons in the household were to be contacted by telephone to obtain basic characteristics for the additional members. (For enumerations obtained in the field, the questionnaire included space to record characteristics of five household members; enumerators were to fill out continuation forms for additional household members.) The issue of how much room the questionnaires would provide 5 See http://www.census.gov/dmd/www/content.htm [12/1/03] for the legal or regulatory authority for each of the 2000 census questions, as compiled by the Census Bureau in March 1998. Input from stakeholder groups, such as the Association of Public Data Users, was instrumental in retaining some items on the 2000 long form that were not specifically mandated or required for federal programs.
OCR for page 81
The 2000 Census: Counting Under Adversity for individual household member characteristics was debated over the decade. Leaving space for detailed data from seven household members had merit but, when combined with efforts to enhance the readability and the attractiveness of the questionnaire, could make the document look long and burdensome and discourage response. Early testing used questionnaires that provided space for characteristics of only five household members, with the intent to use telephone follow-up to obtain characteristics for additional household members when necessary. However, providing space for only five members was believed likely to lead to undercount of members of large households, so a compromise was reached on providing space for six members. The desire to limit the length of the 2000 census forms also made it infeasible to implement some findings from a 1994 Living Situation Survey, which suggested cues that might improve coverage of such hard-to-count groups as young minority men who floated among several households (see Martin, 1999). With regard to the format of the questionnaire and mailing package, the Census Bureau conducted several experiments early in the 1990s to identify formats that would maximize response. A decision to use optical mark and character recognition technology for data capture made it easier to create more visually appealing questionnaires in 2000 than those used in previous censuses (see Appendix C.5). In a 1992 test, mail response to a user-friendly short form was 3.4 percentage points higher than response to the type of short form used in 1990; the difference in response rates for areas that were hard to enumerate in 1990 was even greater, 7.6 percentage points. A short form with fewer questions than the 1990 form further boosted mail response (Dillman et al., 1993). The results from that experiment and a second experiment suggested that the use of more mailings could substantially increase response. It appeared that sending an advance letter (not used in 1990) could increase response by 6 percentage points; that sending not only an advance letter, but also a reminder postcard (used in 1990) could increase response by a total of 13 percentage points; and that sending a second questionnaire to nonrespondents (not used in 1990) could increase response by another 10 percentage points (Dillman et al., 1994). Yet another test in 1993 demonstrated that stressing the mandatory nature of filling out the questionnaire on the envelope was effective in encouraging
OCR for page 86
The 2000 Census: Counting Under Adversity ings and readable questionnaires), and these innovations apparently helped stem the historical decline in mail response rates (see Section 4-B.1). The coverage evaluation research supported the basic DSE survey-based design and led to innovations in the A.C.E. that promised to achieve gains in accuracy and timeliness compared with the 1990 PES. Many of these gains were in fact achieved, although a failure to detect large numbers of duplicate enumerations compromised the A.C.E. population estimates (see Section 6-C.2). Research in the other two priority areas had less of an impact. The research on sampling for nonresponse follow-up had to be discarded because the Supreme Court decided that a sample-based census was unlawful. The MAF research did not proceed far enough in the early 1990s, with the result that additions to the program and changes in the schedule—late implementation of the full block canvass, rushed plans for local review, and questionnaire labeling before verification of most of the addresses submitted by local governments—had to be made late in the decade. 3–C DETERMINING THE 2000 DESIGN We now turn to the political and advisory processes that shaped the 2000 census design over the 1990s. In the first half of the decade, the balance of stakeholder opinion provided impetus to the Census Bureau to plan an unprecedented use of statistical methods to reduce costs and improve the completeness of population coverage in 2000. In the second half of the decade, a battle ensued between those who welcomed and those who feared statistical adjustment of the census counts. The result was to delay a final decision on the 2000 design and the provision of needed funding until very late in the decade. 3–C.1 1991 to 1996 The concern over the perceived failings of the 1990 census led the Census Bureau, Congress, and stakeholder groups to take an unprecedented interest in beginning to plan for 2000 before release of data products had even been completed for 1990. In late 1991 the secretary of commerce established a 2000 Census Advisory Committee, consisting of over 30 representatives from a wide range of associations representing business, labor, minority groups, data
OCR for page 87
The 2000 Census: Counting Under Adversity users, scientific professions, state and local governments, and others (see Commerce Secretary’s 2000 Census Advisory Committee, 1995, 1999). This committee met several times a year over the decade (and continues to meet on planning for the 2010 census). Also meeting regularly during the 1990s on 2000 census issues were the Bureau’s long-established Advisory Committee of Professional Associations and advisory committees for minority groups (Citro, 2000a). Oversight committees in Congress were active in holding hearings and taking such steps as passing the 1994 Address List Improvement Act to facilitate development of the MAF for 2000. Two Committee on National Statistics panels were established in 1992 to address 2000 census planning. One panel was convened at the behest of Congress to undertake a thorough review of data requirements and alternative designs for 2000 (National Research Council, 1995b); the other panel was convened at the Census Bureau’s request to consider detailed methodology (National Research Council, 1994). Subsequently, a third CNSTAT panel was organized in 1996 to comment periodically on the Bureau’s maturing plans for 2000; this panel issued its final report in 1999 (National Research Council, 1999b). All three panels supported the Bureau’s research on DSE and sampling for nonresponse follow-up as ways to contain costs and improve accuracy in the 2000 enumeration. The Bureau initially presented a range of alternative designs to these groups and others, such as the U.S. General Accounting Office and the Department of Commerce Inspector General’s Office. It then widely shared the results of research and testing as it homed in on a design. 3–C.2 Original Design for 2000 As early as 1993 the Bureau adopted the overall concept of a “one-number census” for 2000, in which a combination of counting, sampling, use of administrative records, and estimation would be used to produce the best possible population data for reapportionment, redistricting, and other uses of census data (Miskura, 1993). In contrast, the 1990 census used a dual strategy, in which two sets of population totals were produced—from the enumeration and adjusted for coverage errors using the PES—and an ex post facto decision was made about which set to use (see Chapter 5).
OCR for page 88
The 2000 Census: Counting Under Adversity In early 1996, following a major test in 1995, the Bureau settled on a specific design for 2000, which would have used sampling for nonresponse follow-up and Integrated Coverage Measurement (see Sections 3-B.3 and 3-B.4). Other innovative features of the 2000 design compared with 1990 included development of the Master Address File from multiple sources in order to reduce census omissions because of overlooked addresses (see Section 3-B.1); use of redesigned mailing materials and questionnaires to promote mail response (see Section 3-B.2); expansion of advertising and outreach and use of paid advertising; aggressive recruitment of enumerators; adoption of new data capture technology; and hiring of contractors for data capture and other operations instead of conducting all operations with census staff. Finally, the design deemphasized field work, not only as a way to improve completeness of coverage but also as a way to obtain data content, relying heavily on computer editing and imputation in place of additional follow-up for missing household data. Indeed, the 2000 design focused primarily on coverage and much less on data content in comparison with the 1990 design. 3–C.3 1996 to 1998: Two Dress Rehearsals Views in the Congress and the executive branch on the merits of the Bureau’s 2000 census plan—particularly, the proposed use of SNRFU and ICM—divided along partisan lines: the Clinton administration and House Democrats generally supported the use of sampling while House Republicans generally opposed its use (see McMillen, 1998). Complicating matters throughout 1996 and into 1997 was that the Bureau had not yet issued a detailed plan of operations for the census (see U.S. General Accounting Office, 2002c). In Title VIII of an emergency supplemental appropriations act, Congress ordered the Census Bureau to provide a detailed plan within 30 days of enactment; the act was signed into law as P.L. 105-18 on June 12, 1997. The Census Bureau plan was delivered July 12 (Bureau of the Census, 1997), but it was discovered to have an error in calculation of the likely savings from sampling for nonresponse follow-up, which hurt credibility with Congress. Conflict between the two sides resulted in a delay of funds with which to conduct the 1998 dress rehearsal. In fall 1997 compromise
OCR for page 89
The 2000 Census: Counting Under Adversity legislation stipulated that unadjusted data would be released from the 2000 census in addition to any adjusted data. It also provided for expedited judicial review of the legality of the use of sampling for the census and established a 2000 Census Monitoring Board to consist of four members appointed by House and Senate Republican leaders and four members appointed by President Clinton in consultation with House and Senate Democratic leaders. The Monitoring Board was proposed as a way to address the concerns of some in Congress that the Census Bureau might manipulate the census data for political gain. The congressional and presidential appointees each had their own budgets and staffs. They were charged to report periodically to Congress through September 2001 and then go out of existence. In none of its reports did either the congressional or presidential members of the Monitoring Board cite evidence or make arguments suggesting partisan intent in the choice of design features or implementation of the 2000 census (see U.S. Census Monitoring Board Congressional Members, 1999, 2001; U.S. Census Monitoring Board Presidential Members, 2001a,b). With the mechanism of the Monitoring Board in place, there was agreement that the dress rehearsal and other necessary census planning activities would be fully funded and that the dress rehearsal in one of the planned sites would be conducted without the use of either SNRFU or ICM. The House of Representatives also established a Subcommittee on the Census in the Committee on Government Reform and Oversight, which began census oversight activities in early 1998 and ceased operations at the end of 2001.6 The Census Bureau consequently adjusted its planning to allow for two possible census designs—one based on SNRFU and ICM, the other implementing follow-up on a 100 percent basis and using a postenumeration survey to evaluate coverage. The Bureau implemented each design in a dress rehearsal in spring 1998: the rehearsal in Sacramento, California, used SNRFU and ICM; the one in Columbia, South Carolina, and surrounding counties, used 100 6 At the end of 2001, authority for census issues was vested in a Subcommittee on Civil Service, Census, and Agency Reorganization. Subcommittee structures were revised when the chairmanship of the full House Committee on Government Reform changed hands in the 108th Congress. Census issues are now in the jurisdiction of the Subcommittee on Technology, Information Policy, Intergovernmental Relations, and the Census.
OCR for page 90
The 2000 Census: Counting Under Adversity Box 3.1 Department of Commerce v. U.S. House of Representatives: Sampling in the 2000 Census In 1997, amid debate over the role of sampling methods in the 2000 census, Congress passed an amendment to the Census Act, Title 13 of the U.S. Code, prohibiting “sampling or any other statistical procedure, including any statistical adjustment,” in deriving counts for congressional apportionment (H.R. Conf.Rep. 105-119). However, the appropriations bill containing this provision was vetoed by President Clinton. Subsequently, Congress revised the proposed language, dropping the prohibition on sampling but requiring the Census Bureau to prepare to test a census design “without using statistical methods” (111 Stat. 217). The same legislation also required expedited judicial review of cases filed by parties who argued that they would be harmed by implementing sampling methods in the census; among the parties explicitly authorized for this fast-track review were members of Congress or either house of Congress. Two subsequent legal challenges were heard by federal district courts: Glavin v. Clinton, filed in Virginia by four counties and residents of 13 states, and U.S. House of Representatives v. Department of Commerce, filed in the District of Columbia. In both cases, the district courts concluded that the Census Bureau’s proposed use of sampling for purposes of generating apportionment counts violated the Census Act; having concluded such, neither court addressed the question of whether the sampling methods were unconstitutional. The two cases were consolidated on appeal and heard by the U.S. Supreme Court on November 30, 1998. On January 25, 1999, Justice Sandra Day O’Connor’s opinion for a 5–4 Court majority concurred with the lower court findings and prohibited the use of sampling-based methods in tabulating population counts for reapportioning the U.S. House of Representatives (Department of Commerce v. United States House of Representatives, 525 U.S. 316). Chief Justice William Rehnquist and Justices Antonin Scalia, Anthony Kennedy, and Clarence Thomas joined the key portions of the O’Connor opinion, with Justice Stephen Breyer joining the section regarding the standing of the litigants to bring their suits; one subsection of the O’Connor opinion on the lack of discussion of relevant sampling issues during congressional debate was not adopted by a majority of the Court. Census Act language passed in 1957 provided that “[e]xcept for the determination of population for apportionment purposes, the [secretary of commerce] may, where he deems it appropriate, authorize the use of the statistical method known as ‘sampling’ in carrying out the provisions of this title” (this initial authorization allowed for the first long-form sample involving questions not asked of all respondents; see Section 2-D). The majority Court opinion concluded that subsequent revisions and interpretations of the language must be interpreted in the “context … provided by over 200 years during which federal statutes have prohibited the use of statistical sampling where apportionment is concerned. In light of this background, there is only one plausible reading of the [Census Act language, as revised in 1976]: It prohibits the use of sampling in calculating the population for purposes of apportionment.”
OCR for page 91
The 2000 Census: Counting Under Adversity The Court continued that, because the methodology was prohibited by existing statute, “we find it unnecessary to reach the constitutional question presented”—that is, whether sampling is unconstitutional rather than simply unlawful for purposes of deriving apportionment counts. The applicability of sampling-based methods for purposes other than apportionment was not at issue. Principal dissents were filed by Justices John Paul Stevens and Stephen Breyer. Stevens argued that Census Act language enabling the secretary of commerce to “take a decennial census of population … in such form and content as he may determine, including the use of sampling procedures and special surveys” constitutes an “unlimited authorization” that is not impaired by the “limited mandate” contained elsewhere in the act. Breyer’s dissent also concluded that the Census Act does not prohibit sampling for purposes of apportionment counts, but on different grounds. Specifically, Breyer argued that the Census Act “permits a distinction between sampling used as a substitute and sampling used as a supplement;” interpreting the Census Bureau’s proposed ICM plan as a complement to rather than substitution for a “traditional headcount,” Breyer found no prohibition in the statute. Justice Stevens extended his argument to the issue of constitutionality of sampling methods, arguing that “the words ‘actual Enumeration’ require post-1787 apportionments to be based on actual population counts, rather than mere speculation or bare estimate, but they do not purport to limit the authority of Congress to direct the ‘Manner’ in which such counts should be made.” percent nonresponse follow-up with a postenumeration survey. A third dress rehearsal in Menominee County, Wisconsin (which includes the Menominee Indian Reservation), was a hybrid using 100 percent nonresponse follow-up and ICM. 3–C.4 1999: Supreme Court Decision In January 1999 the U.S. Supreme Court ruled that existing law (in Title 13 of the U.S. Code) prohibited the use of sampling in producing population counts for congressional reapportionment. Further detail on the Court’s decision is given in Box 3.1. The ruling did not preclude the use of sampling to adjust census data for other purposes. Following the Supreme Court decision, the Census Bureau again revised its plans. The final design for the 2000 enumeration was announced by Director Kenneth Prewitt at a press conference in February 1999, little more than a year before Census Day on April 1, 2000
OCR for page 92
The 2000 Census: Counting Under Adversity (see Section 3-C.5). After intense debate, Congress approved the design and provided the full amount of funding—over $4.5 billion—requested by the Census Bureau for fiscal year 2000. Total appropriations for fiscal years 1991–2003 provided a budget of over $7 billion for the 2000 census. 3–C.5 The 2000 Census Design The final design for 2000 dropped the use of SNRFU and ICM and adopted 100 percent follow-up of nonresponding households. It also expanded the number of field offices to manage the enumeration (from the originally proposed number of 476 to 520 offices, excluding Puerto Rico) and expanded the advertising program. For these reasons, the fiscal year 2000 budget request for census operations requested an increase of 61 percent over the amount originally requested (U.S. General Accounting Office, 1999a:4). The final design, however, retained other features originally planned for 2000, including the emphasis on computer imputation to supply missing data for households and persons in place of repeated follow-up attempts and the use of multiple sources to develop the MAF. The MAF development procedures were modified in the late 1990s to respond to problems in the original plans, and a special operation to reduce duplicate enumerations resulting from duplicate MAF addresses was added to the census in summer 2000. The final 2000 design included the Accuracy and Coverage Evaluation Program, which fielded an independent household survey smaller than that planned for ICM (300,000 compared with 700,000 households), but larger than the 1990 PES (165,000 households). The plan was to use the A.C.E. results, if warranted, to adjust the census counts for redistricting and all other purposes except reapportionment of the Congress. The first adjusted results under this scheme would be released by April 1, 2001. As described in Chapter 5, this plan was ultimately not carried out. 3–D FINDINGS AND RECOMMENDATIONS The lack of agreement among key stakeholders about an appropriate design adversely affected the planning and decision-making process for the 2000 census and added to its costs. The level of
OCR for page 93
The 2000 Census: Counting Under Adversity rancor and mistrust between the Census Bureau and Congress, and between parties within Congress on census issues, reached distressing heights. An overstretched Census Bureau staff spent much energy late in the decade on developing plans for alternative designs—with and without extensive use of statistical sampling and estimation—and consequently could not make firm plans for implementation. The Bureau staff also faced unprecedented demands for information and testimony on an almost daily basis in response to an extremely large number of congressional, administrative, and nongovernmental oversight groups. The 1998 dress rehearsal was a large-scale comparative test, not a rehearsal of an agreed-on plan to work out the operational kinks. Uncertainty about funding impeded hiring of needed staff and resolution of specific design elements until less than a year before the census was to begin (see Waite et al., 2001). The last-minute decision on the design and provision of funding put the census in substantial jeopardy. Specific procedures for several operations were finalized very late. Likewise, many data management and processing systems were specified and developed at the last minute and implemented almost as soon as they were completed, without benefit of advance testing. The cost of the census escalated greatly compared with 1990—from $32 per household enumerated to $56 per household enumerated (in fiscal 2000 dollars). The cost increase was due not only to the decision to conduct a basically traditional census with complete field follow-up for household coverage, but also to the lateness of the design decision, which meant that careful planning for cost efficiency was not possible. To get the job done, the Census Bureau poured on resources; there was no opportunity for careful consideration of how to constrain costs at the same time. (See Table 3.1 for a summary of the total costs of the past four decennial censuses.) To the Bureau’s credit, the census was conducted on time and within the specified budget ($400 million of the $7 billion was not spent). Most operations went smoothly (see Chapter 4). However, some of the problems in the census—such as difficulties in enumerating group quarters (see Section 4-F)—very likely can be traced, at least in part, to the haste with which the Bureau had to develop specific operational plans to implement the final agreed-on design.
OCR for page 94
The 2000 Census: Counting Under Adversity Table 3.1 Decennial Census Costs, Total and Per Housing Unit, 1970–2000 (in constant fiscal year 2000 dollars) Census Period Costs (billions of dollars) Cost per Housing Unit (dollars) 1970 1964–1973 $0.9 $13 1980 1974–1983 2.2 24 1990 1984–1995 3.3 32 2000 1991–2003 6.6 56 SOURCE: U.S. General Accounting Office (2001b:Table 1). In structuring this report, we defer most of our findings and recommendations to specific chapters devoted to assessment, but this chapter’s narrative of the convoluted development of the 2000 census plan is the natural point to highlight a most basic finding: Finding 3.1: The lack of agreement until 1999 on the basic census design among the Census Bureau, the administration, and Congress hampered planning for the 2000 census, increased the costs of the census, and increased the risk that the census could have been seriously flawed in one or more respects. Looking ahead to the next census cycle, it is in the interest of the Census Bureau, Congress, and the public that the design and planning process for the 2010 census proceed as smoothly as possible. Of course, there should be continued input and critical review of the Bureau’s planning and research from a wide range of stakeholders. Yet there should be a commitment by all parties to reach a decision on the basic design by 2006 so that the dress rehearsal in 2008 can be as effectively planned as possible. The need to reach early closure on the design of the 2010 census is heightened by the fundamental departure of the Census Bureau’s emerging 2010 census plan from the 2000 census model. In particular, the Census Bureau has proposed replacing the census long-form-sample with the American Community Survey (ACS). The ACS is anticipated to include 250,000 housing units per month—sufficient sample size to provide detailed population profiles for counties,
OCR for page 95
The 2000 Census: Counting Under Adversity cities, and other local areas. Yearly estimates would be produced for areas of at least 65,000 population; estimates for areas with smaller populations would be based on aggregating the data over 3- or 5-year periods. Our companion Panel on Research on Future Census Methods reviewed the current plans for the ACS. The panel concluded that significant work must still be done by the Census Bureau in evaluating the relative quality of ACS and census long-form-sample estimates and in informing data users and stakeholders of the features and the problems of working with moving average-based estimates. However, the panel concluded (National Research Council, 2003a:7): We do not see any looming flaw so large in magnitude that full ACS implementation should be set aside. We therefore encourage full congressional funding of the ACS. It is important, though, that Congress recognize that funding of the ACS should be viewed as a long-term commitment. Cuts in funding in subsequent years (and with them reductions in sample size) will impair the overall quality of the survey, with first and most pronounced impact on the ability to produce estimates for small geographic areas and population groups. In concordance with the Panel on Research on Future Census Methods, and consistent with the history of the 2000 census plan, our panel agrees that the replacement of the census long-form sample is a sufficiently sweeping design choice that agreement must be secured early to facilitate a successful 2010 census. Recommendation 3.1: The Census Bureau, the administration, and Congress should agree on the basic census design for 2010 no later than 2006 in order to permit an appropriate, well-planned dress rehearsal in 2008. Recommendation 3.2: The Census Bureau, the administration, and Congress should agree on the overall scheme for the 2010 census and the new American Community Survey (ACS) by 2006 and preferably earlier. Further delay will undercut the ability of the ACS to provide, by 2010, small-area data of the type traditionally collected on the census long form and will jeopardize 2010 planning, which currently assumes a short-form-only census.
OCR for page 96
The 2000 Census: Counting Under Adversity This page intentionally left blank.
Representative terms from entire chapter: