THE 23RD U.S. DECENNIAL CENSUS in 2010 promises to be markedly different from its predecessors in several important respects, if current plans hold. Data from the American Community Survey (ACS), a continuous monthly survey of households, are intended to replace data traditionally gathered by the census “long-form” sample. Through this change, the U.S. Census Bureau hopes not only to provide more immediate and accurate data on the detailed economic and demographic topics formerly covered on the census long form, but also to streamline and focus the decennial census on the basic “short-form” questions that provide the population counts used to reapportion legislatures and allocate federal funds. It is also hoped that technology—in particular, the use of handheld computers by temporary enumerators for nonresponse follow-up—will improve the quality of the census data.
In spirit and effect, though, the modern decennial census draws fundamentally from the legislation enacted by Congress to conduct the very first decennial census of the United States in 1790. That legislation dictated that residents should be counted at their “usual place of abode” (1 Stat. 101, §5). Every subsequent decennial census has adopted the same basic goal of counting every resident of the United States once, and only once, at what is determined to be his or her usual place of residence.
The verbs “enumerate” and “count” are used interchangeably in everyday discourse, but they are fundamentally different in the context of population censuses and surveys. In a census, enumeration is the actual collection of data from a person: a person is “enumerated” at the location where he or she is found, whether through a mailed questionnaire, personal contact, or other
means. This location may or may not be the same as the location where a person is “counted”—recorded or tabulated in the census returns. Indeed, the difference between “enumerate” and “count,” between collection and recording, defines two distinct concepts of residence that have historically shaped censuses and surveys. In purest form, a de facto approach to defining residence directly equates the two locations: people are counted exactly where they are found or enumerated at the time of the census. Alternatively, a pure de jure approach counts people at a place of legal residence, which may not be the same as the place of enumeration.
Both the de jure and de facto approaches are ideal models that are difficult or impossible to execute in their purest forms. For instance, unless a nation maintains a central population register or national identification program (and all residents comply with the program), different views of what constitutes legal residence complicate a purely de jure count. Likewise, a pure de facto snapshot of the population of a country would require so massive a deployment of resources—covering every house, hotel, and vehicle at the designated time—as to be technically infeasible. Instead, hybrid approaches that may be close to (but not strictly) one of the ideal models are commonplace. Accordingly, when a census or survey is identified as de jure or de facto, it typically has a residence standard that is closer to one of these models than the other, and the label is used as a convenient shorthand. The U.S. census—based on the concept of “usual residence”—is commonly referred to as a de jure census even though it is not purely so. It contains elements of a de facto system, such as programs for counting people in temporary or transient residences. It is also not a pure de jure system because the definition of “usual” residence is not written into legal code or regulation; rather, it is open to varying interpretations.
“Usual residence” is both a very simple and a bewilderingly complex concept, due to the inherent ambiguity of defining “usual” for some living situations. Although, for most respondents, identifying a usual residence is a straightforward exercise, for many others it is not. Moreover, neither is it always clear that respondents’ notions of their usual residences are the same as the Census Bureau’s concept. Even the marshals on horseback dispatched to conduct the 1790 census had to confront the potential residence ambiguities posed by people splitting time at more than one residence or whose travels put them away from home for lengthy periods. Modern censuses have to find ways to count a population that is increasingly mobile and diverse, with complex living arrangements: for example, children in the joint custody of divorced parents, “snowbirds” and other second-home owners who spend large parts of the year in different places, and long-term recreational vehicle users and others who may be true nomads who do not have any place where they stay “most of the time.” Complex living situations also challenge core definitions and approaches to taking a census: how do terms like “household” apply
in migrant worker communities and densely populated immigrant enclaves of major cities? Should prisoners be counted in exactly the same manner, or asked the same questions, as residents of college dormitories or nursing homes?
As a guide to how different living situations should be reconciled with the “usual residence” standard, the Census Bureau maintains a set of residence rules for the decennial census. By 2000, the Census Bureau’s listing included 31 formal residence rules. The actual compilation of residence rules is rarely if ever viewed by the general public. Instead, census respondents typically see only the instructions and questions on the census form, designed to distill the basic residence concepts and lead respondents to provide answers consistent with the “usual residence” standard. Though the full set of rules is rarely seen by census respondents, a clear concept of the meaning of residence, coupled with an effective mapping of that concept to the actual conduct of the census, is crucial to the accuracy of a census.
THE PANEL AND ITS CHARGE
In 2004 the U.S. Census Bureau requested that the Committee on National Statistics (CNSTAT) of the National Research Council convene a Panel on Residence Rules in the Decennial Census, with the following charge:
This study will examine census residence rule issues and make recommendations for research and testing to develop the most important residence rules for the 2010 census. Recommendations will address potential ways to modify census residence rules to facilitate more accurate counting of the population or identify the reasons why the rules should stay the same.
The panel would consider residence rules in terms of how they contribute to or inhibit an accurate count of the population. Its deliberations may include the appropriate geographic location for enumerating each person but would not include the issue of who should be enumerated in the census—for example, whether civilian citizens who live abroad or undocumented immigrants should be included.1
The latter issue, as to whether illegal immigrants—or, for that matter, any non-U.S. citizen—should be included in the census count, remains a contentious one. The issue was the subject of major legal challenges in the 1980 and 1990 censuses (Federation for American Immigration Reform v. Klutznick and Ridge v. Verity, respectively). A segment of the 1986 Council of Professional Associations on Federal Statistics workshop on residence rules (described in Section 1–B) provoked a lengthy discussion on this question. Most recently, H.J.Res. 53 introduced in the 109th Congress proposes that the Constitution be amended so that census totals used to apportion the House of Representatives “shall be determined by counting the number of persons in each State who are citizens of the United States.” As of August 2006, the bill had not been acted upon by the House Judiciary Committee; the House Subcommittee on Federalism and the Census (of the Government Reform committee) held hearings on the bill in December 2005. (See Massey and Capoferro  for a recent overview of the limitations of current data sets in examining the size and trends in undocumented migration.)
The proposal for the panel study further identifies basic questions of interest:
What kinds of population groups are most likely to be double counted or omitted due to confusion about or inappropriate application of residence rules?
What relevant socioeconomic trends are occurring that could explain why one or more residence rules may no longer be relevant or may be difficult to enforce in the census context?
What kinds of solutions to residence rule problems should be explored?
If the panel proposes possible changes to the rules, what kind of test design is necessary to determine if the proposed changes have the anticipated impact?
Our panel is one of three CNSTAT panels simultaneously studying different aspects of the decennial census; the others are the Panel on the Functionality and Usability of Data from the American Community Survey and the Panel on Coverage Evaluation and Correlation Bias in the 2010 Census. Though the three panels differ in their core topics, there are areas of substantive overlap between them—for instance, the difference in the underlying residence concept between the ACS and the decennial census and the design of residence probe questions for census follow-up and coverage measurement operations. Accordingly, members of our panel have participated in activities of our sister panels when applicable, and their members have joined some of our discussions.
As our panel examines the residence rules and concepts for the 2010 and future censuses, we obviously build on decades of work and experience by the U.S. Census Bureau. Significantly, we also draw from the experience of previous conferences and studies related to census residence.
In December 1986 the Council of Professional Associations on Federal Statistics (COPAFS) convened a Workshop on Residence Rules, sponsored by the Census Bureau. That conference featured several overview papers on the residence concept, as well as detailed presentations on specific longstanding residence concerns such as Americans living overseas, prisoners, and migrant workers. The proceedings of the workshop are printed in Council of Professional Associations on Federal Statistics (1987), which includes a summary of recommendations (CEC Associates, 1987). The panel has benefited greatly from the comprehensive coverage of the 1986 conference; throughout this report, we refer to this event as “the COPAFS residence rules workshop.”
The 1986 workshop provided particularly strong discussion of the counting of military personnel and of American civilians living abroad and a vigorous discussion of the inclusion of illegal immigrants in the census. Given the boundaries of our charge, we do not examine these topics with the same level of detail, but direct interested readers to the conference proceedings for fuller discussion.2
Other previous CNSTAT study panels have also discussed issues of census residence as part of their work. The Panel to Evaluate Alternative Census Methods (National Research Council, 1994) included a subgroup on response and coverage issues, which commented on residence rules and methods for enumerating hard-to-count populations. In its interim report, that panel argued strongly for continued ethnographic studies and improved methods for collecting information from traditionally hard-to-enumerate populations, of the sort which typically present difficulties for the “usual residence” concept (National Research Council, 1993). Our panel’s work is also informed by the work of two immediate predecessor panels dedicated to the decennial census, the Panel to Review the 2000 Census (National Research Council, 2004c), which analyzed duplication in the 2000 census in great detail, and the Panel on Research on Future Census Methods (National Research Council, 2004b), which analyzed the developing plans for the 2010 census.
The existing record may lead some to wonder why this study is being conducted, at this time. The basic arguments in response to this question are twofold. First, the topic of residence rules in the decennial census sounds, deceptively, like a very narrow focus. In fact, the topic is quite massive. Developing residence rules and effectively conveying residence concepts to census respondents requires careful attention to changes in the demography of the United States (no small task, in itself) along with the latest experience and practice in the psychology of survey data collection. Given the breadth of subject areas, residence rules in the census is a ripe topic for periodic reexamination and evaluation based on the latest research. Second, and more directly, the findings of coverage evaluations of the 2000 census—in particular, high levels of duplication and an estimated net overcount (as discussed below)—are a major impetus for this work. The 2000 census experience heightens the need for attention to both undercoverage and duplication in the census, two components of gross census error.3 Problems with residence rules can contribute to both undercoverage and duplication; as part of a fuller examination of census error, then, it is important to examine whether residence concepts are being conveyed most effectively.
Several of the papers were later printed in the March/April 1988 issue of Society. Some of the discussion related to the evolution of rules for counting Americans living overseas is summarized in Appendix C.
“Error” is used in the statistical sense in this discussion, meaning a difference between an estimate and the (unknown) true value of a quantity.
PLANS FOR THE 2010 CENSUS
As we begin discussion of the role of residence rules in the 2010 census, it is useful first to briefly consider the proposed general shape for the 2010 census and the testing milestones that will lead to the decennial count. (Additional detail regarding the plan, and analysis of it, is provided in National Research Council [2004b].)
We have already mentioned one major innovation planned for the 2010 census: the replacement of the decennial census long form with the ongoing ACS. Though it is not a direct part of this panel’s focus, we discuss the ACS in more detail in Section 8–C. With the arrival of the ACS, the Census Bureau is in the position of having two flagship products—the decennial census and the ACS—follow two complex residence standards that are very different conceptually.
A second major component of the current planning for 2010 is a major upgrade of the Census Bureau’s geographic resources, the Master Address File (MAF) and the Topologically Integrated Geographic Encoding and Referencing (TIGER) system database. The Census Bureau uses the MAF to mail questionnaires and to follow up with nonrespondents. For 2010, the Bureau is planning to integrate the MAF so that the roster of group quarters locations—places such as college dormitories, prisons, and nursing homes—is no longer maintained separately from the roster of housing units. TIGER is the Census Bureau’s digital map for the entire nation and is thus critical for geocoding addresses in order to tabulate census returns at the correct locations. The major share of activities in the Census Bureau’s MAF/TIGER Enhancements Program is aimed at realigning TIGER features using local geographic information system sources and other sources, such as satellite and aerial photography.
The Census Bureau has planned for a mix of mail-only and field census tests (the latter involving deployment of enumerators to follow up with non-responding households) prior to 2010. The first of these operational trials, the 2003 National Census Test, was mail-only and focused on revised layout of the questions on race and ethnicity. The test was also intended to assess the use of alternative response modes in answering the questionnaire—mail, telephone/interactive voice response, and Internet. In 2004, the Census Bureau conducted a census test in four areas: a part of the borough of Queens in New York City and Colquitt, Thomas, and Tift counties in Georgia. This mail and field test was principally intended to examine the feasibility of handheld computing devices for follow-up data collection. The test also included a dry run with new group quarters definitions and revised race and ethnicity questions (the latter based on input from the 2003 test).
In 2005, the Bureau conducted another mail-only National Census Test. As we will discuss in detail in Section 6–D, this test examined a number of re-
vised residence questions and coverage “probe” questions. In October 2005, the Census Bureau posted notice of a further mailout experiment, which was to occur in early 2006 (Federal Register, October 5, 2005, pp. 58180–58182). Plans call for 24,000 housing units to be mailed an experimental form that restructures the basic residence question on the census form, using a design from the 2005 test; the experimental forms add items that prompt respondents to make sure their forms are complete. The experimental form will also prominently display a “due date” to try to determine the effects on the rate and speed of response.
The last test in the sequence is planned to be a mail-and-field test in 2006, on the Cheyenne River Reservation and Off-Reservation Trust Land in South Dakota, and in part of Travis County (Austin), Texas. This test will feature additional testing of handheld computers; for our purposes, the test is important because it is supposed to be an operational test of final residence rules concepts.
The culmination of the testing cycle is a dress rehearsal in 2008; in January 2006, the Census Bureau announced that this rehearsal will take place in San Joaquin County, California, and a nine-county region around Fayetteville, North Carolina. Like the other mail-and-field tests, it will be conducted in a small number of selected test sites. The Census Bureau’s intent is to resolve major experimental components in the earlier tests so that the 2008 dress rehearsal is a true rehearsal. By comparison, the 1998 dress rehearsal for the 2000 census was itself a major test of competing census models, pitting “traditional” census techniques against variants that would include probability sampling of nonresponding households rather than comprehensive follow-up.
The Census Bureau’s test schedules—and the difficulty of making changes in census procedure in a short time frame—limit the ability of this panel to effect major changes in census operations for the next census. The decennial census is a highly complex and resource-intensive operation, requiring substantial lead time not only to test new proposed changes, but also to procure the equipment and materials needed to conduct the count. However, the timing of this panel has allowed it to meaningfully contribute to the 2005 and 2006 census tests through discussions at our public meetings, and our report lays out a broader agenda of testing and experimentation, some of which can and should be conducted in 2010 and some of which is intended to provide a knowledge base to inform residence rule considerations for the 2020 census and beyond.
OVERVIEW OF THE REPORT
idence rules and residence-related concepts in the decennial census context. We discuss the form of the residence rules used in the 2000 census and the basic difficulties faced by both census respondents and the Census Bureau itself in defining residence.
Part II describes numerous living situations for which specifying a “usual residence” is not straightforward. Chapter 3 begins by describing major segments of what the Census Bureau has traditionally termed the “group quarters” population. These groups—students in dormitories, health care patients, and prisoners, along with parts of the military population—are major potential sources of census error. Chapter 4 focuses on general living situations and social and demographic trends that make specification of a single residence extremely difficult. The chapter also includes discussion of some population groups that may be missed, or not well handled, by current census procedures. Chapter 5 draws some basic conclusions and directions for specific research from the overview in the preceding two chapters.
The panel’s core findings and recommendations are detailed in a sequence of chapters contained in Part III. Chapter 6 argues for a reconceptualization of residence rules as they have been used in the past, suggesting instead the development of a core set of residence principles. We discuss the nature of the census questionnaire itself in light of this approach and recommend that the Bureau move toward a question-based (rather than instruction-based) method of collecting residence information. Chapter 7 recommends improvements in group quarters and nonhousehold enumeration and Chapter 8 suggests guidance on other residence-related census operations. In addition, Chapter 8 discusses the Census Bureau’s research and testing program, directed in part at specific concepts that should be tested as part of the 2010 census, but also focused on a broader research agenda to improve the collection of basic census data in the future.
The Census Bureau’s residence rules for the 2000 census are reproduced in Appendix A. The residence concepts and questionnaire structures used in selected foreign censuses are outlined in Appendix B. Finally, though our charge precludes consideration of whether American citizens living overseas should or should not be included in the census, that population—including military and federal government personnel stationed overseas—has been at the core of several residence rules revisions; Appendix C explores that history.