THE FUNDAMENTAL GOAL of the U.S. decennial census is to count each person living in the country once, only once, and in the correct place. Since its inception, the census has followed a variant of a de jure standard for defining residence, seeking to count people at a single “usual residence.” The census does contain some elements for which the alternative de facto standard—counting people at their current residence or where they are found at census time—is used, including operations to count the homeless. However, what was true for the first U.S. census in 1790 remains so for the 2010 and future censuses: residence can be extremely difficult to define and measure. Though most census respondents can readily identify a single usual residence, some people may have ties to two or more residences, and others may lack ties to any fixed residence. The basic concept of “residence” has evolved over time and can vary greatly across segments of the population, as can related concepts like “house,” “home,” and “family.”
The U.S. Census Bureau used a set of 31 formal residence rules for the 2000 census. As in previous censuses, respondents generally saw only the limited extract of these rules that formed the instructions to the first question on the census questionnaire. Rather than subject respondents to the complete and complex list of rules, the Bureau tried to find a mix of instructions and cues to lead respondents to conceptualize residence in the same way as the Bureau.
As it began reworking the residence rules for the 2010 census, the Census Bureau requested that the National Academies’ Committee on National Statistics (CNSTAT) convene this Panel on Residence Rules in the Decennial Census, charged to:
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.
Though the concept inherits from a long tradition of practice dating to the 1790 census act, active census law and regulation do not define the residence standard for the decennial census (de jure or de facto), nor do they define what constitutes “usual residence.” Accordingly, census rules and procedures for defining residence are prime candidates for periodic review, such as we provide in this report.
A PRINCIPLE-BASED APPROACH TO RESIDENCE
Census residence rules must satisfy several functions at once—among them, a reference for enumerators and interested respondents, a guide to the construction of census questions and instructions, and a template for the design of census operations. We find that, as developed and used in the 2000 census, the residence rules for the decennial census were too complicated and difficult to communicate. The set of 31 formal residence rules was not organized for ease in comprehension, and instead seemed to be a loose amalgamation of previously encountered problematic residence situations. The sheer number and redundancy of the rules detract from their effectiveness in training temporary census enumerators.
A basic flaw of the residence rules for the 2000 census is their lack of a conceptual base: they were essentially a set of exceptions to a concept of usual residence, and not an explication of one. Inferences as to the underlying logic of the residence rules required careful scrutiny. As a result, the residence rules for the 2010 and future censuses should be substantially rewritten (relative to those used in 2000), and the Census Bureau should make a concerted effort in 2010 to improve the communication of residence rules. Core concepts should be expressed as a small number of concise residence principles. These residence principles should then be used to develop other products, such as any instructions or cues to respondents on the census questionnaire, training materials for enumerators, census processing and editing routines, and a “frequently asked questions” list for enumerator and respondent reference and posting on the Internet.
As a candidate set, the panel recommends the following suggested statement of residence principles: The fundamental purpose of the census is to count all persons whose usual residence is in the United States and its territories on Census Day.
All persons living in the United States, including non-U.S. citizens, should be counted at their usual residence. Usual residence is the place where they live or sleep more than any other place.
Determination of usual residence should be made at the level of the individual person, and not by virtue of family relationship or type of residence.
If a person has strong ties to more than one residence, the Census Bureau should collect that information on the census form and subsequently attempt to resolve what constitutes the “usual residence.”
If a usual place of residence cannot be determined, persons should be counted where they are on Census Day.
As indicated above, the core set of residence principles should be used to develop specific products, such as an operations guide, the instructions and questions on the census form, and enumerator training materials. An operations guide should be something akin to—but clearer and better organized than—the formal rules of the 2000 census, describing how the residence principles apply to living situations in which residence is complex or ambiguous to define. Such a mapping of the principles to specific living situations would be a valuable resource, not only for Census Bureau staff who provide questionnaire assistance to respondents, but also for interested respondents. This guide could usefully be structured as a “frequently asked questions” list.
A focus on residence principles suggests a specific implication for census operations. Under the normal census calendar, the mass mailout of census questionnaires begins in mid-March; this early mailout is essential, given the operational complexity of the census and statutorily fixed deadlines for the release of selected census results. However, the problem lies in encouraging rapid return of census questionnaires, possibly before the Census Day reference date. The 1990 census flatly instructed respondents to reply by April 1 while simultaneously asking them to report their household composition as of that date; in 2000, language on the questionnaire in two pre-Census Day mailings asked for prompt response. However small the effect may be, it is a logical contradiction to actively encourage people to use a reference date but to report their situation before that date arrives; the Census Bureau should refrain from overtly directing that respondents commit this basic violation of residence principles. To be consistent with the principle of the basic residence question on the census form—where did you live on April 1?—the Census Bureau should encourage prompt response but make clear that the form should be completed and returned on Census Day or as soon thereafter as possible.
CHANGING THE STRATEGY FOR COLLECTING RESIDENCE INFORMATION
Survey research, particularly work on cognitive response to self-administered survey instruments, shows clearly that their form significantly affects the responses. Responses to self-administered census forms depend
upon the visual layout and design of questionnaires as well as the actual wording of questions and residence cues. Evidence suggests that people often ignore instructions on questionnaires. In addition, they may disregard instructions with which they disagree, even if they do read them. They may find the instructions confusing or contradictory with their own views, or may feel that they do not need direction. The Census Bureau should continue and strengthen its research on the combined effects of visual layout features and specific wording situations in the development and testing of questionnaires and their effectiveness. That said, the Bureau should vigorously pursue approaches to collecting residence data that do not depend critically on the wording of a limited set of instructions.
Questions, Not Instructions
The 2000 and other recent censuses put the onus of interpreting and applying residence concepts on respondents, using limited instructional text and lists of specific groups of people to include or exclude to try to lead respondents to follow the Census Bureau’s unwritten concept of residence. The 2000 census form was more visually attractive and easier to follow than its instruction-heavy 1990 counterpart, but it still forced respondents to adhere to instructions that they may not have bothered to read or could not easily understand, absorb a complex concept on the basis of limited information, and accept interpretations of residence they may not share.
Instead, we recommend an approach based on asking guided questions—and multiple questions, as necessary—to elicit residence data. This approach would shift the burden of deciding what constitutes “usual residence” from respondents to the Census Bureau: the census form would allow collection of sufficient information from respondents to allow the Bureau to determine residence during processing and editing for situations that are not straightforward. In the 2010 census, the Census Bureau should conduct a major experiment to test a form that asks a sufficient number of residence questions to determine the residence situation of each person, rather than requiring respondents to follow complicated residence instructions in formulating their answers. The results of this test, and associated research, should guide decision on full implementation of the approach in 2020.
A question-based form might take a “worksheet” approach, spreading the burden of answering over a small number of questions by asking for counts of specific types of people; for example, people who are staying temporarily on Census Day with no other place where they usually reside. We believe that adoption of a question-based rather than instruction-based form could be particularly advantageous given the prospect of multiple response modes to the census—the standard mailed census forms, nonresponse follow-up interviews using handheld computers, and possibly responses by telephone or the In-
ternet. A question-based approach may be more robust against differences in answers due to the mode by which the questionnaire is administered. Regardless of the final structure of residence questions chosen for 2010, research must be done on response effects created by mode of administration—mail, phone, Internet, and interview with handheld computers.
The Short Form Is Too Short: “Any Residence Elsewhere” and Other Questions
In 2000, respecting the demand to keep burden on respondents as low as possible as well as following direction from congressional oversight authority, the Bureau made sure that every question on the census short and long forms was matched to specific legal and regulatory uses. But paring the questionnaire too far can prevent the census from achieving its core mission of gathering accurate resident counts. Collection of enough information to determine “usual residence” requires the addition of some additional questions.
The principal addition that is needed is a question that asks whether each person has any other residence. Foreign censuses have found ways to collect auxiliary address information with an economy of space, and the Bureau’s own valuable work in matching the complete set of 2000 census results against itself (using probability models based on name and date of birth) suggests that the computational power needed to process and retain auxiliary address information is at hand. Consequently, information on “any residence elsewhere” (ARE) should be collected from census respondents. This information should include the specific street address of the other residence location. A followup question should ask whether the respondent considers this ARE location to be their usual residence, the place where they live or sleep more than any other place.
Though we believe that ARE data collection is something that could be implemented for the 2010 census, it may be prudent to include it as a major experiment instead. A major test of census residence concepts, conducted in conjunction with the 2010 census, should be the basis for postcensal development leading to the 2020 census. This test should include both a question-based approach to collecting resident count information and a provision for ARE reporting by all census respondents, including those living in group quarters (nonhousehold) situations. The information should be gathered and processed, field verified on at least a sample basis, and reported on in census evaluations, in order to direct research over the next decade and fuller implementation in 2020.
Finally, no recent census has allowed respondents the ability to directly indicate that they believe that address information on their census questionnaire is inaccurate. Respondents have been unable to indicate, for example, that they have received the form at a seasonal home or that the Postal Service
delivered the form to the wrong unit or apartment. The census questionnaire should allow respondents to correct the address printed on the form if it is wrong (e.g., address is listed incorrectly or questionnaire is delivered to wrong unit or apartment number). In addition, respondent-corrected address information should be one source of information to update the Master Address File.
Related Census Operations: Master Address File
The importance of the Master Address File (MAF)—the address list used by the Census Bureau to mail out questionnaires and assign nonresponse follow-up work—to the census process is difficult to overstate. An accurate MAF is crucial to the quality of the census, as well as the Bureau’s other major survey programs. Together with the Topologically Integrated Geographic Encoding and Referencing (TIGER) system database, the MAF provides the key linkage between personal census questionnaire responses and specific geographic units. Inaccuracy in MAF and TIGER detracts from the quality of the decennial census, producing errors of inclusion and omission.
We endorse the recommendations of previous CNSTAT panels that the Census Bureau’s efforts to continually update the MAF are vitally important and need careful planning. The Bureau must find ways to solicit and use input from local and tribal authorities in updating and correcting the MAF on a regular basis, with particular attention to obtaining information on unusual housing stock (such as multiple housing units inside family homes, leased hotel or motel quarters).
Several of the most prominent situations where usual residence is not easily defined involve people living in group quarters or nonhousehold situations. These include college students, prisoners, and patients and residents in health care facilities. Consistent with the findings of predecessor CNSTAT panels, we find that, as implemented in the 2000 and recent censuses, group quarters enumeration is unacceptably bad. Failure to reconcile the group quarters roster with the MAF contributed to a host of census errors. Group quarters frames were constructed without sufficient standardization and awareness of diversity in housing unit and group quarters stock, and data from the 2000 census long-form sample were particularly marred by extremely high levels of item nonresponse. The latter failure—the quality of long-form-sample data—will be obviated in the 2010 census by the advent of the American Community Survey (ACS) and, consequently, the use of only a short form. However, the challenge of collecting even the basic census items from group quarters’ populations remains.
It is particularly important that the Bureau’s planned integration of the MAF with its roster of group quarters/nonhousehold locations—which were completely separate in 2000—be executed effectively. Further, it is essential that the Bureau establish programs for continuous update and refinement of group quarters address listings. In 2000, the segment of the Local Update of Census Addresses (LUCA) program dedicated to the review of “special places” (larger entities like colleges and military bases that contain group quarters) was particularly rushed and not given ample consideration. We strongly urge that participants in 2010 census local geographic partnership programs should be allowed to review address listings for group quarters in their jurisdictions, not just the household population listings. In addition, the Census Bureau should consider an improved special place LUCA program under which colleges and universities, medical facilities, and other group quarters locations may review the Bureau’s address listings for their facilities.
The Census Bureau has begun serious work to redefine group quarters, to provide more meaningful categories, and to be more consistent with terms used by practitioners. Although some interim definitions were apparently finalized too late to be tested effectively in 2004, these revised definitions should be tried out in the 2006 census test and the 2008 dress rehearsal. We encourage the redefinition efforts but also suggest a broader focus: we conclude that there is sufficient diversity in what the Census Bureau has treated as the “group quarters” population that the term “group quarters” no longer makes conceptual sense. Its compartmentalization as a separate list and a separate operation—trying to force this entire segment of the population to respond to the census using a single form—is fundamentally flawed. Other dimensions may be more meaningful and easier to implement, such as a distinction between institutionalized and noninstitutionalized populations or a distinction based on length of stay (short-term versus long-term facilities).
As recommended above, ARE information should be collected for all group quarters/nonhousehold residents, just as we advocate its collection in the main household census form. In 2000, the Individual Census Reports administered to all group quarters residents asked whether the person had a “usual home elsewhere” (UHE) and asked for address information. However, the formal residence rules only made residents of a few group quarters types eligible to be counted at their UHE location (if that address was found to be valid). The Census Bureau’s failure to analyze the rich UHE data collected from other group quarters is highly regrettable; ARE information should be collected from all group quarters residents in 2010, and those data should be analyzed and evaluated extensively.
We favor a system in which group quarters/nonhousehold residents are approached and enumerated in the same manner as the general household population. Direct enumeration (questionnaires distributed to and filled out by respondents, or administered by enumerator interview) is preferable to other means of data collection. However, the 2000 census experience underscored a key practical reality: even with its vigorous and generally successful partnership operations with community organizations, only about one-half of census records for the group quarters/nonhousehold population were obtained from direct enumeration. Instead, many were culled from facility and administrative records. Direct questionnaire distribution and enumerator access to all parts of the group quarters/nonhousehold universe cannot be assumed; many group quarters administrative staff will be either unwilling or unable to permit such direct access. In these cases, the Bureau needs to consider optimal use of enumerator time and facility resources. The Census Bureau should produce a small number of alternative census forms that collect a common core of information for different types of residence settings, such as those that are known to have long lengths of stay rather than short-term stays. The Census Bureau should also develop a spreadsheet-type ledger form that reflects the reality that some “responses” will have to be obtained from facility administrative records or a central “gatekeeper.”
People in Prisons
A particular issue involving the group quarters/nonhousehold population that has drawn considerable attention in the buildup to the 2010 census is whether prisoners should be counted at the prison location or at some other place. A provision in the Census Bureau’s 2006 appropriations required it to provide Congress with a report on the feasibility of counting prisoners at a “permanent home of record.”
Major growth in the prison population, accompanied by expansion in the number of correctional facilities maintained by the federal government and the states, has prompted challenges to the Census Bureau’s “usual residence” standard regarding the counting of the incarcerated population. The prison population includes disproportionate numbers of racial minorities and persons from large urban areas; that this population is counted in the largely rural areas where prisons tend to be located, and that they are included in redistricting calculations despite being barred from voting in most cases, raises legitimate concerns of equity and fairness in the census.
Under the panel’s recommended principles for determining residence, federal and state prisoners would be counted at the prison location because that location is the place where the prisoner lives and sleeps more than any other place; this is consistent with current Census Bureau practice. However, two corollaries are in order:
Our principles hold that determination of usual residence should be made at the level of the individual; this would mean that persons in prison need not have their residency fixed solely by virtue of their location in a structure identified as a prison. Using the panel’s recommended question-based approach and revised nonhousehold enumeration operations, the census could obtain individual-level information on time spent in prison and expected date of release. The Bureau would then have the information needed to implement an individual-level assignment of residence under whatever set of rules is in place prior to the actual enumeration.
In 2000, less than 20 percent of the population in correctional facilities was enumerated through self-response to a questionnaire or through enumerator interview. Even with enhanced efforts to facilitate direct interviews in prisons, it is only realistic to assume that a major share of the prison population in 2010 will have to be counted using administrative and prison records. Hence, any prospect for counting prisoners at locations other than the prison depends vitally on the completeness, consistency, and accessibility of records maintained by individual prisons or by state and federal departments of corrections. The quality of these data resources is not well known. Absent this knowledge, it is difficult even to identify the alternative to counting at the prison: Should prisoners be counted at their last preincarceration address (if one can be specified), or do records only permit prisoners to be allotted to the county or city (much less the tract or block) from which they were sentenced? The Census Bureau should participate in a comprehensive review of the consistency of content and availability of prison records. The accuracy of prisoner-reported prior addresses is uncertain, and should be assessed as a census experiment. Though an “enduring ties” argument is frequently invoked to argue for changes to the Bureau’s prisoner counting policy, the strength of those ties merits empirical assessment. For example: Has the property (to which a prisoner is connected) changed ownership? Do respondents at the address have any contact with or relation to the prisoner?
The evidence of political inequities in redistricting that can arise due to the counting of prisoners at the prison location is compelling. Short of counting prisoners at some location other than the prison—for which there is currently insufficient information as well as the lack of any principled way to do so—a partial remedy might be to provide tract- or block-level counts of prisoner populations as part of the Bureau’s data products for redistricting. State redistricting bodies would then have the capacity to decide whether to include or exclude prisoners from proposed districts. The states’ interest in having such a separate pris-
oner count should be assessed by the Bureau as part of its work with state officials to determine the layout of the standard redistricting data file in 2010.
In the future, the Bureau may decide on an exception being made or a principle being added to count prisoners at a location other than the prison. However, the information necessary for such a decision does not now exist. A research and testing program, including experimentation as part of the 2010 census, should be initiated by the Census Bureau to evaluate the feasibility and cost of assigning incarcerated and institutionalized individuals, who have another address, to the other location. Such a program would include collecting and analyzing individual-level residence information (either self-reported ARE locations or the location coded in corrections department files).
EXPERIMENTATION AND TESTING FOR THE FUTURE
Our review of residence-related research at the Census Bureau, including the design of mid-decade census tests in 2005 and 2006, suggests critical deficiencies in the Bureau’s overall research agenda. One such deficiency is that the Census Bureau often relies on small numbers (20 or less) of cognitive interviews or very large field tests (tens or hundreds of thousands of households, in omnibus census operational tests) to reach conclusions about the effectiveness of changes in census enumeration procedures. As a consequence many important questions about the effectiveness of residence rules do not get addressed effectively.
We strongly suggest that the Bureau make fuller use of moderate-scale surveys and tests, combining field tests of a few hundred households with cognitive interviewing. The Census Bureau should undertake analytical research on specific problems in order to better evaluate the effectiveness of residence and other questions on the census forms. These studies should be designed to focus on particular populations of interest. Candidates for such research include:
why babies are often omitted from the census form (targeted at households with newborns);
whether census respondents find a pure de facto residence rule easier to follow and interpret than a de jure rule (generally, and with specific reference to large households);
whether additional residence and location probes on questionnaires—increasing the length of the survey—impairs response or other operational activities (e.g., page scanning);
the difficulty and advantages of including a reference date or time frame;
multilingual and linguistically isolated households; and
whether the Census Bureau standard of “live or sleep most of the time” is consistent with respondent notions of “usual residence.”
On a similar note, data similar to those collected by the 1993 Living Situation Survey should be conducted on a regular basis. A convenient form for a more regular study could be inclusion of a supplement to the ACS or a stand-alone survey.
An advantage of the use of moderate-scale experiments is that they can focus on specific problems; the major omnibus tests and experiments performed by the Bureau often combine multiple topics and treatments so that effects of specific changes are difficult or impossible to determine. When designing experimental tests, the Census Bureau should always include a control form—either the questionnaire items used in the preceding census or the exact items used in immediately previous census tests—so that individual modifications can be more effectively assessed.
Another deficiency in the Bureau’s research program involves unanalyzed data from the 2000 (and previous) censuses. Through its own resources as well as contacts with outside researchers, the Census Bureau has data on diverse residence situations that could be used to inform residence-related decisions. Although the research done to date does provide some information on the nature of omissions and duplicates in the 2000 census, the analyses are not sufficient to fully sort out important effects, and the data that have been collected need further analysis. One specific example of this deficiency concerns census reporting in large households. The decennial census form only allows for detailed data collection on six persons (names can be listed for an additional six people); this may contribute to an undercount of children and babies if census respondents tend to list their household members in descending order of age. Possible trends in age reporting should be able to be detected using census operational data: extant data from the 2000 census on large households of seven or more members should be reanalyzed for better understanding of the nature of the households and to inform better practices to collect data for large households. Generally, the Census Bureau should conduct and facilitate further research using its detailed census and survey results; as needed, the Bureau should consider ways to facilitate this work through contracts with outside researchers.
The ACS that will replace the long-form sample in the 2010 census should be a major focus of residence-related research by the Census Bureau because its residence standard differs from the decennial census. Specifically, the ACS uses a “two-month rule” or “current residence” concept that is akin to a de facto standard, while the decennial census “usual residence” standard is a de jure type. The nature and extent of interpretation problems that may arise due to these discrepant standards is vital to evaluating both. The Census Bu-
reau should plan to ask a question on the usual residence of each household member in the ACS questionnaire, in order to evaluate the extent of incongruity of residence standards between the long-form replacement survey and the decennial census. The usual residence question should first be tested using the survey’s experimental “methods panel”; the resulting data should be fully evaluated and analyzed to refine final versions of the question.
The Census Bureau is considering unduplication methodologies for the 2010 census, building from innovations in the 2000 census and its success in probabilistically matching census records based on name and date of birth. In particular, the Bureau is exploring real-time unduplication during census processing and is developing an expanded coverage follow-up operation to provide data to help identify potential duplicates. Focus on duplication is important; however, Census Bureau research on living situations that do not easily fit census residence rules should strive to gather data on the sources of omissions in the census, as well as sources of duplication. In addition, a comprehensive assessment of the components of gross coverage error (both undercount and overcount) should be added as a regular part of the census evaluation program.
The mechanics of censustaking have changed greatly since marshals were first sent out on horseback in 1790; as times have changed, the “usual residence” concept has endured even though its exact interpretation has shifted. The most recent paradigm shift in defining residence in the census came with the adoption of mail-based enumeration for most of the census population in 1970; that shift included drawing a linkage between census residence and a specific mailing address. Looking to the future, over the long term, the Census Bureau research program needs to consider broader shifts that lie ahead—the impact of the Internet and e-mail and the diminished importance of traditional mailing addresses (and paper mail) in people’s lives, more transitory living arrangements, the changing need for census data as private and public databases grow in completeness.
There is a serious need for additional quantitative information on the magnitude of emerging social trends for groups, as well as a need for further qualitative assessment and better definitions of concepts. Important hypotheses can emerge from qualitative techniques such as ethnographic research, but these need to be tested quantitatively. People’s attachment to households and group quarters has changed significantly over several decades and is likely to continue to change in ways that cannot now be predicted with confidence. The Census Bureau should establish a standing research office whose task it is to continually monitor changes in factors influencing people’s attachments to locations where they are counted, and the connectedness of changes among them, using such information to generate appropriate research and recommendations for changes in how people can be more accurately enumerated in the decennial census.