Mirroring America: Living Situations and the Census
IN A PAPER originally prepared for the 1986 Council of Professional Associations on Federal Statistics residence rules workshop, Anderson (1988:50) observed that
for some groups with ambiguous residential statuses the [Census Bureau] has accumulated a body of information on where they are and has developed mechanisms to count them. I would include students, military personnel, and persons in institutions in this category. For others, less is known. Here I would include undocumented immigrants, the homeless, and migrant workers. I would also include those groups that display newly emerging residential and family patterns: children whose parents share custody; families with earners in different cities or regions and thus two residences; individuals with seasonal residences.
Today, the state of affairs is not much changed. The Census Bureau’s ethnographic research program in the 1990 and 2000 censuses yielded small glimpses at the dynamics of some ambiguous residence groups, such as new immigrants and migrant farm workers; see Box 5-1. However, other key groups of interest remain mysteries: the full nature and extent of commuter marriage arrangements is unknown, and information on the snowbird and sunbird populations is limited to the local-area studies conducted by some particularly affected areas. Empirical information on ambiguous residence groups and their response to different question styles and enumeration approaches is rare; the stand-alone 1993 Living Situation Survey (see Box 5-2) and the Alternative Questionnaire Experiment conducted as part of the 2000
census (see Box 5-3) provided some information on general response to residence questions and cues, but not with so large a sample that inference can be made about key groups like households with college or boarding school students or children in joint custody arrangements.
LESSONS FROM A REVIEW OF LIVING SITUATIONS
Our review of living situations in Chapters 3 and 4—situations in which ties to a single, “usual” residence are ambiguous—is meant to be comprehensive but not exhaustive. Moreover, even for the groups we profile in these chapters, not every detail or conceptual wrinkle that apply are described in these pages. However, we believe that the discussion in these two chapters does give rise to several fundamental baseline messages.
The first is a reaffirmation of a basic notion that is all too easy to forget. A recitation of the various hard-to-count population subgroups is a reminder that the American populace has never been easy to count. Conducting the decennial census is a tremendously difficult and complex task. It is useful to bear in mind that any process that tries to count approximately 300 million individuals, deploying hundreds of thousands of temporary staff and processing millions of questionnaires of various sorts, in roughly 6–8 months will necessarily have some shortcomings.
Second, the array of challenging living situations should dispel the notion that there are any quick or easy fixes. No set of residence rules or set of questions, however carefully developed and articulated, can completely eliminate error in the reporting of residence. Individual living situations, as well as respondents’ ability, willingness, and incentive to answer questions, are too varied to be addressed by any single procedure or “one size fits all” approach to data collection. What can be done is to solidify core concepts and adjust enumeration procedures to try for greatest improvement; this is the motivation for our call for a delineation of core residence principles in Chapter 6.
Third, though individuals’ living situations can vary greatly, our review does suggest some commonalities, in general structure and motives, across various hard-to-count groups. Single residence rules cannot be considered in isolation from each other; changes in the way one group is counted may have implications for the enumeration of others. For instance, the intent to return to a particular location, even when one is staying at a different place for some period of time, plays into the definition of “usual” residence for some people. The fact that prisoners may eventually return to a “home” community is suggested as an argument for counting prisoners there rather than at the prison. But in assessing that change, it is important to consider how much weight to put on intent to return: for prisoners, even though property may have changed hands or family members at “home” may have severed ties with
Ethnographic Research in the Census
Ethnography is a social science research methodology that provides intensive study of particular communities or groups through observational analysis and extensive interviews. The Census Bureau has used ethnography to study coverage in the census since the 1970s, with major efforts associated with the 1990 and 2000 censuses.
The 1990 Ethnographic Evaluation of the Behavioral Causes of Census Undercount studied 29 sites and groups (de la Puente, 1993). By comparison, the 2000 census program of formal experimentation included six ethnographic research projects, focused on: attitudes toward protecting privacy (Gerber, 2003); participation by “Generation X,” persons born between 1968 and 1979 (Crowley, 2003); mobile populations (Hunter et al., 2003); complex households—in particular, on-reservation Navajo Indians, Inupiaq Eskimos, Korean immigrants, Latino immigrants, rural non-Hispanic whites, and blacks in southeastern Virginia (Schwede, 2003); social network tracing of highly mobile people (Brownrigg, 2003); and colonias—poor rural communities near the U.S.-Mexican border (see Box 4-2) (de la Puente and Stemper, 2003). The results of each of the six projects were summarized in one of the Bureau’s evaluation “topic report” series (de la Puente, 2004), and several of the papers are published in revised or extended form in Schwede et al. (2005).
Among the 2000 census ethnographic studies, Schwede (2003) found that examination of “complex households” in six socioeconomic groups highlighted key differences between the Census Bureau’s definition of “household” and the definition envisioned by respondents. For instance, the Census Bureau’s conception of a household as the set of all persons living in one housing unit runs contrary to the experience of Navajo Indians and Inupiaq Eskimos, for whom family ties are more central to the notion of a household than physical location. Members of the family who live at a further distance, in sheep camps, in another housing unit on the reservation, or off the reservation may be considered part of the household if they are contributing to the family’s income or subsistence; conversely, people living in the same housing unit—even if they are related—may not consider themselves part of the same household if they are not pooling incomes or sharing food.
Ethnographic research also suggested a rich set of high-mobility patterns, for economic and other reasons, that complicate the definition of usual residence (Schwede, 2003; Lobo, 2001; Fleisher, 2001):
Living Situation Survey
The Living Situation Survey (LSS) was a special survey conducted in May–October 1993 by Research Triangle Institute, sponsored and designed by the Census Bureau. The basic objective of the survey was “to test experimental roster probes designed to improve coverage of tenuously attached people. The LSS also collected a great deal of detailed data about movements in and out of sample households” to provide basic information on fundamental residence issues, such as the assumption that each person in the United States has a single, uniquely defined usual residence (Martin, 2004:1).
LSS data were collected from a national multistage probability sample of housing units, designed so that it oversampled minority and renter areas. Interviewers performed personal visits, contacting a household respondent for each sampled housing unit and asking that respondent to list all the persons associated with that unit within the previous 2–3 months. The interviewer then guided the respondent through a series of 13 roster questions; though some emulated the decennial census usual residence approach, the LSS questionnaires did not provide residence rule cues such as lists of specific include/exclude suggestions. After the battery of rostering questions, the respondent was then asked questions about each individual they had identified in rostering. Upon completion of the interview with the designated household respondent, the multistage nature of the sampling design came into play, as selected individuals identified in the rosters were asked to answer questions about themselves. The survey had a 79 percent response rate and was completed in 999 housing units in 1993; those housing unit interviews yielded 3,537 identified persons, 2,825 of which had “more than a casual connection” to the sampled housing units (Sweet and Alberti, 1994:319).
The LSS asked the household respondent to identify whether each person they identified was a usual resident at that address, based on the census definition (where they live and sleep most of the time). The detailed queries also permitted a determination as to whether each person was a usual resident based on a time-based criterion, based on whether they had moved in or out of the residence during the 2- to 3-month reference period used in the survey. Sweet and Alberti (1994:320) report that the two standards agreed most of the time but not exclusively; 94.89 percent of the respondent-labeled and time-based assessments of usual residence agreed with each other; 3.71 percent were dubbed usual residents but did not live in the unit for more than half of the reference window, and 1.41 percent had lived at the unit more than half the time but were not correctly labeled by the respondent. Inconsistency was most prevalent for 18–29-year-olds (Sweet and Alberti, 1994:320).
An important finding of the LSS concerned the reporting of black and Hispanic males aged 18–29. Nearly a quarter of these minority males found by the survey were misclassified in terms of their status as usual residents: 17.5 percent were dubbed usual residents by the survey respondent but had not actually spent most of the past 2–3 months at the residence, and 5.6 percent had spent most of the past several months at the household but were left off the list of usual residents by the respondent (Sweet and Alberti, 1994:321). Related issues are discussed in more detail in Section 4–A.5.
Alternative Questionnaire Experiments
The past three decennial censuses have all included some form of Alternative Questionnaire Experiment (AQE), designed to test responses to different versions of either the census short-form or long-form sample questionnaire. The 1980 AQE tested two “matrix”-type forms more conducive to the optical sensing technology then used to extract electronic data from microfilmed questionnaire images against a more respondent-friendly design; the 1990 AQE tested five different variations of the census long-form questionnaire (DeMaio and Bates, 1992; Martin et al., 2003).
The 2000 incarnation of the AQE was broader in scope than its predecessors, incorporating three separate subexperiments. One of these focused on the census long form and the branching and “skip” instructions that are meant to guide respondents through the questionnaire (Redline et al., 2002). Another tested different versions of the race and Hispanic origin questions and changes in reporting induced by those questionnaire features (Martin, 2002). It is the third component of the 2000 AQE—“An Experiment to Improve Coverage Through Revised Roster Instructions” (Gerber et al., 2002)—that is germane to this panel’s study; hence, in this report, we will often use “2000 AQE” to refer specifically to the portion of the experiment that studied roster and residence concepts.
The full 2000 AQE consisted of eight panels of 5,000 households (10,000 were included in one of the panels); five of these focused on changes to the cues used to help respondents navigate through the census long form, and one presented the 2000 census short-form questions but used the design conventions of the 1990 census. The two panels of key interest from the perspective of census residence rules were a modified short form with revised residence and roster questions and a control group consisting of the standard 2000 census short form. Both of these panels were mailed to approximately 5,200 selected households, and both experienced return rates on the order or 73 percent. The sample of households was stratified by “high coverage areas” (low proportions of minorities and renters in the 1990 census) and “low coverage areas” (high proportions), and were drawn from the Master Address File for mailout/mailback areas of the country (Martin et al., 2003).
Respondent households that returned the revised-roster questionnaire and the 2000-census control questionnaire by mail and included phone numbers were sampled so that the respondents could be reinterviewed (4 months later) and the results used to study the effects of design changes. In total, 4,218 households were selected for reinterview; 2,958 interviews were completed.
The specific changes to the residence question and its related instructions that were tested by the 2000 AQE are described in Box 6-2.
the prisoner; for persons displaced by a natural or manmade disaster, when the structure they occupied at “home” may no longer exist and may not for the foreseeable future; or for long-term hospital patients, for whom there may be the intent to return but not the physical capability.
For another instance, if one considers changing rules or instructions to strengthen some peoples’ possibly tenuous ties to a household and remind the person filling in the form that they should be included, how would the changes affect other groups? If such a cue is targeted at partners in cohabiting couples, should similar cues be used to signal the inclusion of foster children (no matter how short the placement) or extended kin and family members staying in American Indian households? As a final example, a consistent approach would seem to be desirable for handling housing units that are inherently mobile and not tied to a particular land location, whether they have small “crews” (long-haul trucks, trains, or recreational vehicles) or large ones (military and commercial sea vessels).
The sheer number of living situations for which residence can be difficult to define fosters our fourth message. Trying to craft a rule for every eventuality and circumstance is a losing proposition. The 31 formal residence rules of the 2000 census were an attempt to be more comprehensive and inclusive than the 1990 census (see Box 5-4), and in the history of the American census may be bested only by the 1950 decennial for length and intricacy (see Box 5-5). The 1950 enumerator instructions—like the 31 rules of 2000—were ambitious, and reading through them does give a sense of the magnitude of census operations. Moreover, as we argue in Chapter 6, there remains a need for some detailed explanation of how specific residence scenarios should be handled in the census (e.g., for reference by field enumerators or by census staff manning telephone questionnaire assistance centers). But maintaining a long list of residence rules, built by aggregating experience with difficult problems in previous censuses, results in a flawed product, the internal logic of which is difficult to immediately discern.
Our final message from the review in the past two chapters is simple. Not enough is known about hard-to-count populations. Key questions concern the size of the populations in question and the social or demographic trends that are making them more (or less) problematic for the census. As Anderson (1988) noted in the quote at the beginning of this chapter, we do not know enough about the extent and real impact (demographic and economic) of seasonal migration, about degrees of attachment of foster children to their households or group living sites (and vice versa) and the accuracy of gathering data on such in censuses and surveys, or about the blending of short-term and long-term resident situations in the same structure (such as the housing of long-term prisoners in local jails or the establishment of semipermanent living quarters in contemporary hotels and motels). Addressing this deficiency is the focus of the remainder of this chapter.
Residence Rules for the 1990 Census
SOURCE: Bureau of the Census (1995:App. 1C).
Include and Exclude Instructions in the 1950 Census
A section labeled “Persons to Enumerate in Census of Population” in the 1950 census enumerator instructions can be parsed as at least 40 separate instructions on how to address living situations, whether on the standard population schedule or on Individual Census Reports (then used for nonresidents and visitors, such as persons in hotels). Excerpted below are the portions of the instructions specifically formatted as persons to enumerate (on the standard schedule) and to exclude:
Include as members of the household:
Exclude as members of the household:
NEEDED RESEARCH ON LIVING SITUATIONS
Conduct of the decennial census requires a major allocation of money and staff. A large share of these resources must necessarily be dedicated to the basic mechanics of the census—the hiring of staff, the development and processing of questionnaires, and the maintenance of support systems like the Master Address File among them. The resources available for research and refinement of techniques to improve the count are relatively scarce, particularly in the years between censuses. Yet the types of living situations described in this report require continued research to inform improvements in enumeration methods. The need to allocate scarce resources motivates the need for serious attention to quantitative information on groups that the Bureau has more typically addressed through qualitative studies.
Finding 5.1: 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.
Fuller Use of Internal Data
We note first the Census Bureau’s unique access to its own microdata. Two of our predecessor panels (National Research Council, 2004b,c) have urged the Census Bureau to continue to mine and analyze the microdata from the 2000 census and its follow-up and coverage evaluation programs, in ways that get beyond the mainly operational focus of the formal evaluation program of the 2000 census. We concur, and add that the Bureau’s major survey programs also provide the basis for further investigation of residence-related questions, whether the surveys are self-initiated (e.g., the Current Population Survey or the American Community Survey) or conducted on behalf of other agencies (e.g., its surveys of the jail and prison population).
Finding 5.2: 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.
Recommendation 5.1: The Census Bureau should conduct and facilitate further research using its detailed census and survey results.
The Committee on National Statistics’ Principles and Practices for a Federal Statistical Agency argues that statistical agencies have a responsibility for in-house analysis of their own data. “There are strong arguments for a statistical agency to have staff whose responsibility is to conduct objective substantive analyses of the data that the agency compiles, such as analyses that assess trends over time or compare population groups”; among these are the ability to use the results of those studies to inform changes to the agency’s data collection program (National Research Council, 2005:40). Historically, the Census Bureau’s focus has been on the production of data and not internal analysis. We urge the Bureau to facilitate ties with external researchers to carry out research objectives, but—pursuant with the practice of in-house analysis—the Census Bureau needs to take a stronger interest in analyzing and interpreting their own data to guide operational improvements. Given their unique access to their own microdata and operational results, there are some problems that the Census Bureau itself must analyze simply because there is no one else who can.
Monitoring Social Trends
In an organizational hierarchy like the Census Bureau’s, methodological changes and developments such as those described above tend not to happen without an internal advocate and an established base within the Bureau. There is also a need for additional breadth in subject-matter specialists whose insight on census and survey data can be marshaled. Emerging social trends and their impact on the accuracy of basic residence information is sufficiently central to the purposes of the census that the topic deserves a visible, active research effort and institutional support by the Census Bureau. Accordingly, we believe that the research effort we suggest should be a standing activity, including input from all parts of the Bureau but coordinated in a central place:
Recommendation 5.2: 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.
The office we envision in this recommendation would perform a variety of important roles:
Analyze Census Bureau data and facilitate cross-divisional ties: In addition to planning evaluations and analyzing experiments connected to the decennial census, the office would analyze data collected from topic modules on the Current Population Survey and the American Community Survey (discussed below in Section 5–B.3). This research office should also tap the field experiences and operational data from surveys conducted by the Bureau on behalf of other agencies, several of which are noted in the previous two chapters: these include the National Nursing Home Surveys, the National Hospital Discharge Survey, and the Census of Juveniles in Residential Placement.
Design new experiments: Based on perceived gaps in knowledge, the office could design new surveys and experiments of the type we describe further in Chapter 8. Over time, the experience of this office in spurring new research would lay a solid foundation for improving census processes on the basis of sound scientific evaluations.
Build and strengthen ties to external research: An important function of the office would be to monitor and cull from the work of the broader research community. External researchers have developed a number of data sources that focus on living arrangements and difficulties in counting people with ties to one or more places: these include the National Survey of Families and Households (Sweet et al., 1988; Sweet and Bumpass, 1996, 2002), the New Immigrant Survey, the Los Angeles Family and Neighborhood Study, and the site-specific surveys of snow-bird and sunbird populations (discussed in Section 4–A.1). Synthesizing the results of researchers’ analyses of these data sources would be an important function of the trends office; more significantly, the office should study and learn from the approaches used in these studies to reach targeted populations and collect data from them. This work could suggest improvements that could be used to improve the accuracy of the census. We revisit the need for Census Bureau ties to external research in Chapter 8.
Basic Research on Living Situations
In this closing section, we suggest selected topics that are particularly ripe for basic research, several of which could be done using extant data resources internal to the Census Bureau. This is a selected list, and should not be interpreted as either a comprehensive list or as a specification of the highest priority research topics.
First, existing 2000 census data should be able to help address remaining questions on the characteristics and reporting patterns of large households, those with seven or more members; the 2000 and other recent censuses only
collected detailed information for six people in a household and asked only for the name of persons 7–12. In some cases, people beyond the six described on the form were found during census follow-up operations, but in others their characteristics had to be filled by imputation. If there is some instinctive tendency to list household members in rough age order, this structural feature of the census form could help explain undercounting of children and babies; it may also be the case that persons not listed in positions 1–6 are those with the weakest ties to the household, and they may be potential duplicate cases. This potential phenomenon is important to understand for the purposes of an accurate count, but is particularly important with regard to the long-form replacement American Community Survey (ACS). Detailed person-level characteristics measured by the social and demographic ACS questions could be distorted by higher levels of imputation required to fill in persons missing from large households.
Recommendation 5.3: 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.
In particular, the Census Bureau should study trends in age reporting by the person number on the census form, the geographic and demographic concentration of these large households, and the characteristics of persons in positions 7 or higher relative to those for which full information is available in positions 1–6.
Other potentially interesting avenues of research connected to the undercount of babies and young children is to examine the actual reported age of babies in the census in comparison with vital statistics data. It would be interesting to see if birth months are missing equally, or whether a lack of March/April births may speak to structural problems in census reporting (see Section 6–F for further discussion). Ultimately, this work could lead to revisions to the structure of the questionnaire, such as the number of persons for whom full information is requested and the number for which only names are listed. Such revision is consistent with an idea we explore in Chapter 6, that the census short form may be too short.
The choice of large households and the potential effect on undercount of the young as a research topic is a useful one because it focuses on a potential source of census omissions. As general guidance to ongoing research on social trends and enumeration methods, it may approach being a truism but we think it important to stress that sources of both duplication (overcount) and omission (undercount) be considered. As we have noted, duplication is often the more tractable problem to analyze, but appreciating the extent to which
social trends and existing census approaches might influence census omissions is vital to overall accuracy.
Recommendation 5.4: 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.
Not all of the important contextual work on residence trends can be achieved by reanalysis of existing data and strengthening ties with external researchers and other survey programs. There is a need for original work as well: the trends office we recommend could design and implement original quantitative work on the basis of identified gaps in existing knowledge. It is surprising that it has been over a decade since the Census Bureau sponsored its LSS; a return to collection of hard data on ambiguous residence situations and questionnaire strategies is long overdue.
Recommendation 5.5: Data similar to those collected by the 1993 LSS 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.
By this recommendation, we emphasize that we do not literally mean that the LSS be replicated and conducted exactly as it was in 1993. The content of the survey should reflect major data needs and, to the extent possible, the survey should be targeted and designed in order to achieve adequate sample coverage in populations of interest. It is also essential that the data from this work be done in a well-documented and publicly accessible way. While we do not advocate an exact replication of the LSS, neither do we think it wise to be overly prescriptive of the shape the work should take; it is for this reason that we suggest a module or supplement to the ACS as a possible vehicle. Current plans for the ACS already include a “methods panel” to test new approaches and questions that would be well suited for additional residence questions; we discuss this further in Section 8–C. A supplemental module to the Current Population Survey (CPS) may also be a useful means for collecting these data, though design differences between the CPS and the census or ACS might make this option less useful for suggesting specific improvements to the census. Regardless of the mechanism by which the data are collected, the crucial thing is that the data be produced, so that future discussions of ambiguous residence situations and possible corrective strategies can be supported by formal quantitative research rather than anecdote.