Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Response and Coverage One of the most important findings of the 1990 census the increasing di- versity of the United States-also identifies one source of its challenges. Diverse housing, living arrangements, and language proficiencies all complicate the task of counting a population that is also increasingly mobile. Many people continue to be persuaded by appeals based on citizenship and a common national duty, and many have the skills and motivation needed to respond to the census mail ques- tionnaire. But sizable and perhaps increasing proportions appear to be motivated primarily by local interests and appeals, demand control over portions of the census process or outcomes, or require specialized help or media in order to participate. Others seem simply unmotivated or distrustful of government efforts to collect information. If the trends documented by the 1990 census continue, as expected, the 2000 census will face even larger obstacles. Measuring the effec- tiveness of census reform may therefore be complicated, because even greatly im- proved procedures may not yield greatly improved outcomes (e.g., in terms of public response to the mailed census questionnaire). A recognition that the popula- tion of the United States is simply and fundamentally becoming ever more difficult to count must be incorporated into planning for 2000 in order to develop viable strategies and the organizational and political consensus to implement them. Much of the Census Bureau's research on response and coverage for the 2000 census addresses two main criticisms raised about the 1990 census: the high unit cost of the census and the persistent differential undercount by race. Increasing the primary mail response rate is vital both to improving data quality and to reducing follow-up costs, thereby conserving resources for the task of reducing the differential undercount. 47
48 COUNTING PEOPLE IN THE INFORMATION AGE Problems of differential coverage can and should be addressed at the count- ing stage of census operations, using methods to improve response and coverage, as well as at other stages. ~ Reducing the differential undercount through improv- ing the initial count necessarily requires a large investment to reach a small proportion of the population that is relatively inaccessible to routine standardized procedures. Under such circumstances, controlling costs and the differential undercount in 2000 may require a massive reorientation-both conceptually and organizationally. The 1990 census attempted to apply a standard, and standard- ized, approach everywhere, but the outcome was still a differential undercount. Furthermore, standardization necessarily broke down in many geographical ar- eas, particularly those in which response to the mail questionnaire was low, as enumerators with only brief training were sent out and ultimately empowered to make last-ditch enumerations, using procedures known as last resort and close- out that permitted contacting persons who were not residents of the household. As discussed in Chapter 2, we recommend pursuing uniform outcomes, tailoring operational methods in controlled ways. The 1990 experience with coverage improvement programs suggests that the real alternative to recognizing the increasing diversity of the country and plan- ning for it may be using different approaches in a haphazard or ill-designed manner. The second phase of the 1990 parolee-probationer check, in which information about parolees and probationers was gathered from administrative lists, was not planned in advance of census operations. The vacant/delete check of housing units that were identified as vacant or nonexistent during nonresponse follow-up required enumerators to determine, months after Census Day, housing unit status on Census Day. Both programs apparently introduced large numbers of erroneous enumerations (Ericksen et al., 1991~. Uncontrolled variation in operational procedures becomes even more problematic under integrated cover- age measurement (see the section on uniformity of treatment in Chapter 2~. We believe that a system could be designed that is flexible enough to control or reduce the differential undercount yet maintain important aspects of standardiza- tion (such as definitions of household membership). It is unlikely that the methods discussed in this chapter for improving census response and coverage will completely eliminate differentials in coverage. How- ever, their use will contribute to two key goals: preserving the credibility of the census and addressing social changes that would otherwise tend to exacerbate differential coverage problems. Also, efforts to improve response and cover- age and, in particular, to reduce differential coverage~uring initial census operations will improve accuracy at intermediate stages and therefore reduce the ~ New approaches to address list development, discussed in Chapter 2, could improve coverage of hard-to-locate housing units, and sampling and statistical estimation can be used to measure and correct for differentials in census coverage, as discussed in Chapter 4.
RESPONSE AND COVERAGE 49 burden on sampling and statistical estimation in producing results from the one- number census (see Chapter 4~. Conversely, response and coverage improvement methods that are too ex- pensive to be implemented on a large scale during the counting stage might be valuable tools in the field operation for integrated coverage measurement. Thus, sampling and statistical estimation can reduce the burden on coverage improve- ment efforts and improve the cost-effectiveness of the counting operation. As noted in Chapter 1, the allocation of resources among the major stages of census data collection will be a critical point in the development of the 2000 census design. RESEARCH ON RESPONSE AND COVERAGE ISSUES TO DATE The various research programs for the 2000 census have approached the two broad goals described above along several fronts. This chapter considers compo- nents of the research program that are concerned primarily with updating, ex- panding, and improving data collection methodologies ways people participate in the census to achieve better initial response and coverage and to control or reduce the differential undercount. The topics and research programs we review address the following issues: improving the implementation of the residence rules to increase the accuracy of coverage within households (referred to here as roster improvement research); increasing the response rate to the paper question- naire (response improvement research); using the telephone to answer questions and to accept interviews from citizens who call in during the initial (mailout) phase of the census, to conduct interviews during nonresponse follow-up, and to conduct reinterviews as part of integrated coverage measurement (use of the telephone); expanding traditional data collection methods to include other auto- mated technologies (using other technologies); developing methods tailored for groups that may be difficult to enumerate (hard-to-enumerate populations, tool kit and planning database); encouraging participation in ways that will be effec- tive in addressing the differential undercount (outreach and promotion); and at- tempting to further develop links with lower governmental units (state and local cooperative ventures). The various research programs aimed at response and coverage issues neces- sarily developed at different rates. The response improvement research, which developed methods for improving the initial response rate, was yielding results before this panel was constituted. It has already contributed very promising techniques for further evaluation in the 1995 census test. The procedures tested in the experiments reviewed below do require some further refinement. But the principal challenges for this research program now are ensuring that the methods can be made fully operational in a census. Reaping the benefits of respondent- friendly census forms, for example, requires continued progress in the develop
so COUNTING PEOPLE IN THE INFORMATION AGE ment of methods for scanning forms and capturing data. The continued research on new methods of fostering should result in a final census form that is substan- tially different from those already tested in the response improvement research. Close and continuing communication among those developing new methods of fostering, those designing ways of presenting census forms, and those familiar with the technical and operational details of producing and processing census forms is needed to ensure that the response improvement research yields the . . . gams it promises. Other portions of the research program on response and coverage issues have developed more slowly but are very promising. Research aimed at improving fostering and targeting barriers to enumeration have the potential to produce important innovations and possibly reduce the differential undercount. The ros- ter improvement research is beginning to produce findings that may lead to redesigned instruments that elicit more accurate responses, particularly from those living in complex households. But this program requires sustained efforts and funding beyond the 1995 census test if its important potential is to be realized by 2000, and the same is true for research that examines application of the tool kit to reach hard-to-enumerate populations. Research about outreach and promotion has barely begun~espite evidence that continuous outreach is important, at least in some geographic areas or with some groups (Bentley and Furrie, 1993~. The very successful research on improving the initial mail response rate has developed strategies that promise to help control costs in 2000. Although in- creasing the initial response rate may leave the differential undercount un- changed-or even exacerbate it somewhat a higher initial response rate means better-quality data, fewer nonrespondents to follow up, and better control over cost. Techniques that control costs also potentially make resources available for reducing the differential undercount. Research should now focus on techniques that have potential to improve coverage within households or to reduce the differ- ential undercount. The current state of promising research on improving fostering, incorporating the telephone and other technologies into the design of the census, evaluating the tool kit, and developing and evaluating outreach and promotion activities makes it clear that research and development in these areas must con- tinue beyond the 1995 census test. Although this development is likely to con- tinue in the normal transition after the mid-decade census test for topics with an obvious operational component (such as use of the telephone, data capture, and design of census forms), sustained research on other basic topics, such as im- provements in fostering, may require that the Census Bureau give more attention to research and development in the latter half of this decade than has been given in the latter half of past decades. The remainder of this chapter discusses and makes specific recommenda- tions on a range of topics related to response and coverage issues. We consider research completed and currently under way and propose directions for future investigations. Four major themes emerge from our discussion.
RESPONSE AND COVERAGE 51 1. Focus on reducing the differential undercount. We recommend planning and limited experimentation now so that the most promising methods for reduc- ing the differential undercount can, where feasible, be given operational trials and experimental and cost evaluation in the 1995 census test. It is important that the proposals embodied in the tool kit be evaluated or tested in 1995. Focusing research on methods that may reduce the differential undercount is important given the limitations of resources and time, and it can be justified by the more advanced state of research on improving mail response rates. Analyses of 1980 and 1990 census data that include operational variables, such as the date on which a case was received and how the form was completed (household informant or enumerator), could supplement ongoing experiments by characterizing past re- sponse patterns. We urge the Census Bureau to devote resources to analyses of existing data and smaller studies whenever potentially valuable information can be obtained through such alternatives to large, expensive experiments. 2. Examine the implications of structured reliance on multiple instruments and response modes. Past censuses have largely relied on a single instrument, the paper questionnaire, which was usually self-administered but was sometimes administered by an enumerator. The 2000 census is likely to use a paper instru- ment and a computer-assisted instrument for nonresponse follow-up. (A differ- e~ computer-assisted instrument would be used for the integrated coverage mea- surement reinterview.) These instruments will also be used in different modes: the paper instrument used in the mailout will usually be self-administered but will sometimes be completed by enumerators (either in an interview or, possibly, with neighbors or by observation in last-ditch operations). The computer-assisted instrument developed for integrated coverage measurement is likely to be used in two modes telephone and face-to-face. The different instruments and response modes can affect results in two important ways: they can affect the likelihood that a household will be enumerated (coverage of households), and they can affect the responses obtained during enumeration (coverage within households). Proposals to make census forms widely available and accessible raise similar substantive issues such procedures may contribute different amounts to overall response in different areas and create operational problems, such as how to match records and eliminate duplicates. 3. Go local. The targeting model and tool kit proposed in the Census Bureau's research program provide for localized, decentralized outreach and enumeration activities. But more may be needed. Because undercounted groups are clustered and because their reasons for not participating may vary greatly by locality, reducing the differential undercount will probably require a major reori- entation in the Census Bureau's practice. Outreach and enumeration activities may need to be more decentralized. Even when they know where to find under- counted groups, staff in national and large regional offices may not have the credibility or contacts needed to motivate members of these groups to participate. Gaining access to local media markets and learning how to localize outreach,
52 COUNTING PEOPLE IN THE INFORMATION AGE promotion, and enumeration should be central to planning for the 1995 census test, as should the development of designs to evaluate the success and to estimate the costs of such efforts. Good procedures for eliminating duplicate records (see Chapter 2) are also needed if the Census Bureau is to be able to manage more flexible enumeration procedures that respond to local needs (e.g., filling out a form in the shopping mall) without losing accuracy. 4. Evaluate alternatives, considering their cost and contributions to reduc- ing the differential undercount. Evaluating and comparing alternative strategies requires both solid research design and attention to costs. Controlled experimen- tal designs could be augmented by studies that use planned variation in methods to provide a comparative basis for assessing the usefulness of operational proce- dures. The environment within which costs are evaluated will be very different in 2000 than it was in 1990. If new design and implementation techniques yield a higher initial mail response, households that are followed up by other means may be less cooperative on average than nonresponding households in 1990. And as the primary response improves, it becomes more difficult to estimate the impact of follow-up techniques in experimental studies, because the exact characteristics of mail nonrespondents are likely to be affected by the census climate and be- cause even optimal follow-up techniques will affect only a small percentage of cases. Nevertheless, efforts to determine the relative costs of alternatives should be an integral part of their evaluation. ROSTER IMPROVEMENT RESEARCH Promotion, outreach, and increasing the number of ways households can be counted (described in later sections) will probably have their greatest effect on coverage of households, not on coverage of people within households. Approxi- mately 32 percent of people who were not enumerated in the 1990 census were in households that were enumerated (Childers, 1993), so improving the quality of coverage within households-households that are increasingly complex and di- verse is crucial. Within households, coverage errors are response errors. Im- proving coverage of persons within households and reducing the contribution of within-household coverage errors to the differential undercount requires re- ducing response error, because it is household respondents who implement the residence rules as they fill out the census form or tale with a Census Bureau enumerator. Improving the quality of the initial count obtained by household rosters that is, reducing omissions and erroneous enumerations has the im- portant general benefit of reducing the variance of coverage measurement. As with other methods for improving coverage, enhancing the quality of coverage in the initial count should help control costs and develop support for the credibility of subsequent estimation. But in addition to these general benefits, methods of roster- ing that respond to changes in the structure of households in the United States promise to help reduce the differential undercount, and at a relatively low cost.
RESPONSE AND COVERAGE 53 Attention to methods of fostering assumes greater importance for 2000 for another reason: the 2000 census will use several different instruments adminis- tered in different modes, and it is important that the different instruments (and modes) used in obtaining the count provide comparable results. Past censuses relied on a single instrument, the paper questionnaire. This instrument was usually completed by a household informant. But during follow-up operations, enumerators might administer the form in an interview or, in extreme cases, complete it by observation. The 1990 census form used an "include-exclude" list of 13 items to instruct the household informant whom to include or exclude from the roster of household members. As implemented in 1990, the include-exclude list was designed to be used in a self-administered instrument, that is, the list was to be read by the respondent who was completing the form. This approach relies on a motivated respondent to review and implement the Census Bureau's defini- tion of who is a member of the household. The include-exclude list thus presents problems even on a self-administered form, because many respondents will not read the list. In addition, it is not clear how one might effectively translate the include- exclude list to another data collection mode. Like respondents who completed the paper instrument themselves, enumerators who administered the 1990 form as an interview probably varied in whether they read the include-exclude list. The problems of formally adapting the include-exclude list for an interview are suggested by the procedures adopted in the Mail and Telephone Mode Test (described in more detail in the next section), in which the include-exclude list was available to the interviewer on a help screen; the rules were not integrated into the structure of the questions that respondents were asked. Although such an implementation may make the results of the self-administered and telephone versions more comparable if neither respondents nor interviewers read the in- clude-exclude list this comparability probably has a cost in validity. Developing improved methods for fostering household members that pro- vide comparable results with different instruments (paper and computer-assisted) and across modes (mail, telephone, and field enumerator) assumes critical impor- tance for the 2000 census. The 2000 census will incorporate a structured reliance on different instruments and modes of data collection: self-administered instru- ments, paper or computer-assisted personal interviewing instruments adminis- tered by enumerators in the field (e.g., as part of a "blitz" enumeration or non- response follow-up), and paper or computer-assisted instruments administered by telephone as part of nonresponse follow-up. (Computer-assisted telephone inter- viewing and computer-assisted personal interviews will also be used as part of integrated coverage measurement, and we turn to this issue in the discussion.) Without careful instrument design, there are likely to be substantively important differences in the results obtained by the different instruments and, in some cases, by the same instrument used in different modes (e.g., paper instruments may be self-administered or used by an enumerator).
54 COUNTING PEOPLE IN THE INFORMATION AGE Planning for the use of different instruments and modes permits better solu- tions than simply accepting ad hoc adaptations of the self-administered form by individual enumerators, as necessarily happened in the past. In producing a final estimate from integrated coverage measurement, it is important to be able to treat counts for different blocks as comparable, even though the proportion of a block that is enumerated by self-administered form, by face-to-face enumeration, or by computer-assisted telephone interview during nonresponse follow-up will vary. Obtaining counts that are comparable across instruments and interview modes requires developing fostering questions that can be implemented similarly on paper and computer-assisted instruments and for which the instrument and mode differences that do remain are tested and understood. Several research projects that are under way address questions that must be answered in order to increase the accuracy of coverage within households, and these projects can also provide insights that will be needed to develop fostering techniques that can be implemented in comparable ways across modes. The Living Situation Survey, cognitive research on residence rules, and the National Coverage Test (the 1994 census test) are essential first steps in this research. But a sustained effort will be needed to develop the instruments that the 2000 census will require. This is particularly true because the results from the Living Situa- tion Survey and the National Coverage Test will not be fully analyzed by the time the instrument for the 1995 census test must be made final. The priorities for future research and other recommendations are discussed at the end of this section. Living Situation Survey and Cognitive Research on Residence Rules Various complex living household situations unrelated people sharing the same living quarters, children in shared custody arrangements, people with no stable place of residence, and others pose special problems for respondents attempting to apply the Census Bureau's residence rules accurately. The Census Bureau initiated research on these problems by considering the roster questions used in the Survey of Income and Program Participation (SIPP). This research found that, on average, more black males were listed as usual residents when respondents were not required to give names of those added to the roster (Kearney et al., 19931. This research also suggested that several characteristics increased the likelihood that someone listed on a roster would be described as a usual resident: contributing money to the household, considering themselves a mem- ber of the household, not staying other places often, and staying in the household many nights in the past month. Many of the issues raised in the SIPP roster research are examined further in the Living Situation Survey (LSS), a national sample of approximately 1,000 households with an oversample of households with minority populations and renters. The LSS interview uses 13 questions to list people with many different kinds of attachments to a household. Subsequent questions then attempt to deter
RESPONSE AND COVERAGE 55 mine which people on the list are usual residents of the household. Data from the LSS should provide a systematic description of a wide range of complex house- hold arrangements and an evaluation of the effectiveness of different strategies for determining who resides in a household. Although these data are only recently available, initial analyses address ques- tions important both to improving coverage and to reducing the differential undercount. For example, asking "Who slept here last night?" and then "Who lives here but wasn't here last night?" identifies approximately 99 percent of those later named as usual residents (Sweet, 1994:10~. These questions appear, however, to capture males, blacks, Hispanics, and those ages 18-29 at a lower rate than others (Sweet, 1994:13~. This research also examines how people understand terms commonly used in fostering household members. Meaning is often flexible and contextual, and this poses potential problems in wording questions for a census. The LSS interview asked respondents about the meaning of live and stay. The cognitive research on residence rules extended this line of inquiry using cognitive interviews with about 30 people to explore how they understood residence concepts. There appears to be some regional variation in the meaning of live and stay: respon- dents in the Northeast and the West judge the words as different more frequently than do respondents in the South and the Midwest. It seems that the words live and stay alone cannot be used to distinguish between permanent and temporary residents. Not surprisingly, technical vocabulary such as usual residence is not used spontaneously by respondents; respondents are likely to interpret technical words as referring to more familiar concepts (Gerber and Bates, 1994~. Full results from the LSS will probably not be complete before the instru- ment for the 1995 census test is final, but combining results that are available with those of the National Coverage Test (see below) could lead to major innova- tions with considerable promise for improving coverage within households and reducing the differential undercount. By building on this research for the 1995 census test, effects on coverage (including erroneous enumerations) can be as- sessed. For example, approximately 5 percent of the people listed in response to each of the two initial roster questions discussed above are not usual residents (Sweet, 1994), so extending these methods to the census requires efficient ways of identifying people who might be listed at more than one housing unit. The results from this research are very promising, but the complexity of fostering will require that research both on instrument design and on how to allocate those listed on a census form to households continues beyond the 1995 census test. National Coverage Test The National Coverage Test compares two different approaches to counting people within households. One instrument is a respondent-friendly form with content similar to that used in 1990 (and the Simplified Questionnaire Test, see
56 COUNTING PEOPLE IN THE INFORMATION AGE below). The other instrument collects an extended roster (e.g., by asking respon- dents to list ". . . everyone living or staying at this address on Saturday January 29" and ". . . anyone else who you consider to be a member of this household") and then asks questions to identify which of the people listed should not be counted as usual residents (e.g., "Is there another place where this person lives or staysail. The first form corresponds to the traditional practice of requiring re- spondents to implement the Census Bureau's residence rules (via the include- exclude list) in deciding who to list. This approach asks people to report them- selves where the rules say they should be counted; that is, it is a de jure enumeration. The second form does not use an include-exclude list and instead requests information about people who lived or stayed at the address on Census Day or who are considered by respondents to be members of the household. The form also requests information that enables determination of whether those on the extended roster satisfy the residence rules (de jure). In the National Coverage Test, a reinterview will be used to estimate the impact of the extended roster form on gross coverage error, which includes erroneous enumerations and missed persons. The effect of the form difference on mail response rates will also be estimated. This innovative and important experi- ment provides the first trial of the "collect de facto, tabulate de jure" method- perhaps more precisely described as "collect de facto and de jure, tabulate de jure" that was discussed as a possibility but not implemented for the 1990 census (Schwede, 1993; CEC Associates, 1987:261. Discussion and Recommendations Better instrument design is needed to help respondents provide more accu- rate answers. Improving the operational form of residence rules in the various instruments is especially attractive because it could both improve the quality of the initial count and reduce the differential undercount among responding house- holds-both at relatively low cost. In addition, the new approach to fostering should be designed to give comparable results whether a household informant completes a paper instrument or an enumerator interviews the informant with a computer-assisted instrument. The results from the LSS and the cognitive research on residence rules, together with the ethnographic evaluations conducted for the 1990 census (see discussion below), can contribute to improving the conceptualizations incorpo- rated in the Census Bureau's residence rules (see, for example, Bureau of the Census, 1987), as well as their application in fostering questions. Because of the timing of these different streams of research, the instrument designed for the National Coverage Test was able to build on the Census Bureau's experience in designing the LSS and the cognitive research on residence rules but was not able to incorporate fully the results of that research, which is still being analyzed. If the results of the National Coverage Test are promising, the development of the
RESPONSE AND COVERAGE 57 extended roster form will require research continuing after the 1995 census test. Improvements in coverage from changes in fostering are likely to be bought at a relatively low cost. In addition to bringing together the results of these parallel lines of research, this extended research program must take on the important task of designing a version of the instrument to be used with computer-assisted tele- phone and personal interviews during nonresponse follow-up and estimating ef- fects of various instruments and modes of administration. For example, if the telephone will be used for nonresponse follow-up in 2000, the comparability of self-administered paper instruments and computer- assisted instruments administered by telephone must be evaluated and measured before the census. This evaluation would not necessarily require a study of the magnitude of the response improvement studies discussed below or the National Coverage Test. Similar tests should determine the effects of having the census form administered by an enumerator, as would happen in a blitz enumeration. After instrument and mode effects are first estimated, it is likely that all instru- ments will require modification and small-scale testing (such as cognitive inter- views), and that mode effects will need to be estimated again. Thus, the schedule for this research must allow time for several cycles. Although the Census Bureau did not develop an instrument simultaneously for a paper self-administered form and computer-assisted interview (as recom- mended in our interim report), it seems likely that the extended roster approach used in the National Coverage Test will be less susceptible to differences in the mode of the instrument than the 1990 form probably was. This is because the instrument does not rely on the respondents' reading of an include-exclude list, but instead asks direct questions about the relevant criteria. Continued develop- ment of the census form should consider that the 2000 census will rely on several instruments and modes of data collection. Because of the way instruments will be used, what might be, strictly speaking, instrument effects and mode effects will be somewhat confounded; of the two, instrument effects are likely to be greater and to have a more profound impact on within-household coverage than pure mode effects that occur when completely identical questions are adminis- tered via different modes. Information on the presence and size of instrument and mode effects on within-household coverage will be important in specifying the models used in mixed-mode coverage measurement methods. The challenge of these new methods of fostering extends beyond that of designing new instruments. Processing programs must be developed to allocate those listed on the rosters for a final count. Tabulating de jure using a roster of persons who "reside" at the address or are otherwise associated with the house- hold and additional information collected at the time of fostering is a substantial task in itself. A roster that increases the likelihood that everyone will be counted almost certainly increases the chance that some will be counted more than once. The extended roster used in the National Coverage Test requests the alternate address of anyone listed who also has another residence. Assuming similar
58 COUNTING PEOPLE IN THE INFORMATION AGE techniques are used in the 1995 census test and then in the census, the computer programs must be developed and tested to make these matches, reconcile ambi- guities, and allocate the count. Continued experimentation with forms that gather additional information (e.g., on number of different residences, frequency of travel) about people listed on an extended roster might lead to improvements in the ability to eliminate duplicate census records. The Living Situation Survey and the cognitive research on residence rules are likely to yield some complex methods that may be more suitable for use in the reinterview for integrated coverage measurement than in the instruments devel- oped for the initial count. The results of this research and the National Coverage Test should be evaluated together, keeping in mind the complementary goals of improving the initial count and developing more intensive methods for integrated coverage measurement. (If the fostering procedure proves feasible but too costly for the counting phase of the census, it might be used as part of the integrated coverage measurement phase.) In summary, when developing and applying residence rules, the Census Bureau should consider both the need to accurately enumerate diverse household structures and the demands of a design that uses multiple instruments and mul- tiple modes. In the 2000 census, it seems likely that households could be enumer- ated by a household respondent completing a form, providing an interview by telephone to an interviewer in a centralized facility, or being interviewed in person by an enumerator. Recommendation 3.1: A program of research extending beyond the 1995 census test should aim to reduce coverage errors within households by reducing response errors (e.g., by using an extended roster form). This research should also evaluate the impact of these new approaches on gross and net coverage errors, as well as assess the effects on coverage of obtaining enumerations using different instrument modalities (e.g., paper and computer-assisted) and different interview modes (e.g., paper instrument completed by household respondent and by enumerator). RESPONSE IMPROVEMENT RESEARCH The most developed component of the Census Bureau's research on re- sponse issues addresses the problem of reducing the unit cost of each response by increasing the number of people who return the mail questionnaire. As we said, although increasing the initial response rate may leave the differential undercount unchanged-or even exacerbate it somewhat a higher initial response rate means better-quality data, fewer nonrespondents to follow up, and better control over cost. For these reasons, an important role has been played by the experi- ments demonstrating that response rates can be improved by better design of the questionnaires; sending advance letters, reminder cards, and replacement ques- tionnaires; and stamping "mandatory" on the envelope.
RESPONSE AND COVERAGE 59 Taken as a whole, the Census Bureau's response improvement research pro- gram attempts a coordinated approach to a wide range of issues, and the program clearly has the potential to mitigate the decline in mail return rates or even to increase the rates. The use of a respondent-friendly questionnaire, an advance notification letter, a reminder postcard, and a replacement questionnaire are all methods that have been widely adopted in surveys, and their extension to the census is both overdue and promising. Simplified Questionnaire Test and Implementation Test The Census Bureau conducted the Simplified Questionnaire Test (SQT) in April 1992. Its primary purpose was to compare the final mail completion rates for four alternative short-form questionnaire designs (booklet, micro, micro with Social Security number, and roster) with the 1990 census short- form questionnaire in a survey environment (Bryant, 19921. The SQT also in- cluded a replacement questionnaire for those addresses that had not returned a form. The overall SQT completion rates were 63.4 percent for the 1990 short form and 66.8 percent for the respondent-friendly version of the 1990 short form. Thus, the test yielded an increase of 3.4 percentage points in response rate using the latter. The Implementation Test (IT) was designed to test the benefits of three components in the mail implementation plan for the short-form sample a pre- notice letter, a stamped return envelope, and a reminder postcard (Bureau of the Census, 1992d). The IT was carried out by the Census Bureau in October 1992. The test yielded a completion rate of 50 percent for the shorter respondent- friendly version and 62.7 percent for the treatment that included a prenotice letter and reminder card. Thus, the IT experience was that use of a prenotice letter and reminder card together produced a gain of 12.7 percentage points in the response rate. On the basis of the SQT and IT results, Census Bureau statisticians esti- mated that the use of a replacement questionnaire increased response by approxi- mately 10 percentage points (Dillman et al., 1994~. Although it is less clear exactly how effective these techniques will be in a census or whether they will reduce the differential undercount, any increase in routine mail responses im- proves data quality and helps to control costs. Mail and Telephone Mode Test The Mail and Telephone Mode Test was designed to examine whether offer- ing respondents the option of calling in a response or receiving a replacement questionnaire increases the response rate. The treatments varied in when the offer was made, and one treatment (the "preference" treatment) allowed respon- dents to call when they first received the form. The experiment was conducted in April 1992, and results suggest that offering people the opportunity to call in to
60 COUNTING PEOPLE IN THE INFORMATION AGE be enumerated does not increase the total response rate (Clark et al., 1993) and that a small percentage who would otherwise mail back their form use the tele- phone when it is offered. This finding suggests that offering people more high- tech response modes (e.g., fax, personal computer), which are similar to the telephone in requiring citizens to take the initiative in responding, are similarly unlikely to increase the total response rate. The telephone is more likely to be a useful tool in nonresponse follow-up and integrated coverage measurement; we discuss these applications at length later in this chapter. Appeals and Long Form Experiment The key substantive issue explored by the Appeals and Long Form Experi- ment (ALFE), conducted in July 1993, is how to better motivate participation. This is a question that also affects other components of the census (see discussion below of outreach and promotion). One component of this experiment tested the effects on short-form response rates of two different motivational appeals, one emphasizing the mandatory nature of the census, the other stressing the benefits of census participation. A second component compared strong and standard assurances of confidentiality. The results obtained with the short form suggest that simply noting on the envelope that participation in the census is required by law could significantly increase response. Results appear similar in both high- response and low-response areas, so that the difference between these strata is not decreased. Debriefings with respondents indicated that the message that the census was mandatory was not perceived more unfavorably than a message about the benefits of participation (Treat, 1993; Singer, 1994~. Of the respondents who were debriefed and who said they remembered seeing the message on the enve- lope, most gave the message a positive rating regardless of experimental treat- ment (mean ratings ranged from 4.42 to 2.78 on a scale from 1 to 10, with 1 being most favorable) (Singer, 1994:9~. These results suggest a very cheap, simple, and effective method for increasing initial mail response. Further testing might con- sider whether the use of the "mandatory" message has a negative, backfire effect on mail response among some groups who were nonrespondents in ALFE and so were not included in the debriefing study. It is conceivable, for example, that the message would be as effective when used only on the replacement question- naire envelope as when used during the initial mailing, while inviting fewer negative reactions. Such issues could be investigated by incorporating simple treatment variations in the 1995 census test or other experiments in which partici- pation is mandatory. In the long-form component of the experiment, a long form that provided a space for each person in the household obtained a better response rate than the 1990 long form, but also had greater nonresponse to the housing questions (Treat, 19931. A form that used a row-column matrix format did not obtain a better response rate than the 1990 long form. A program that combined design work
RESPONSE ANrD COVERAGE 61 with cognitive testing of alternative forms could attempt to build on the advan- tages of the individual-person form and avoid its weaknesses. Spanish Forms Availability Test The Spanish Forms Availability Test (SFAT), conducted in 1993, was the Census Bureau's first attempt to provide forms in a language other than English as an initial response option. Because research on the 1990 census suggested that language barriers may have had a large negative effect on the initial mail re- sponse rate in some areas, the SEAT examined whether providing either bilingual forms or both English and Spanish forms could increase response to the initial mailing. The sample included a stratum in which 15-30 percent of the house- holds were linguistically isolated and a stratum in which more than 30 percent of households were linguistically isolated. Households were defined as linguisti- cally isolated if no one over age 14 spoke English or if no one in the household spoke English very well (Bureau of the Census, 1993a:August). The preliminary results suggest that response to the initial mailing was positively affected by the treatment in the stratum with a higher proportion of linguistically isolated house- holds. This kind of tailoring of response options is discussed further in several sections below. But the results of the SPAT suggest that the targeting of methods that could increase initial mail response in areas with specific barriers should continue to be considered. Another potential benefit of Spanish or other foreign- language questionnaires is that better information about within-unit residents and their characteristics might be obtained from respondents who answer in their language of choice. Discussion The program of research on improving initial mail response rates could lead to an impressive series of innovations in the census. Although exactly how these methods will be received in a census climate remains to be seen, these experi- ments suggest that the mail form to be used in 2000 will be more widely accepted than that used in 1990. Because these treatments were not all tested in combina- tion with one another, we cannot confidently assess their joint result. But the effects of many of the factors that increase response the prenotice letter, the reminder postcard, the replacement questionnaire, a respondent-friendly design with a space for each person, and the mandatory appeal appear to be largely additive (Dillman et al., 1994~. The magnitude of these treatment effects in the environment of an actual census year is unknown. Nevertheless, the strength of the experimental findings suggests that these features are likely to have a signifi- cant positive effect on response rates in the next decennial census. We do not recommend abandoning research on improving initial mail re
62 COUNTING PEOPLE IN THE INFORMATION AGE sponse, given its critical role in controlling costs and in the overall success of the census. It is also necessary to develop the implications of these experiments for a full census, to integrate these results with those of other experiments (for example, the roster improvement research), and to understand better the effects of the length of the questionnaire on response. But efforts to improve initial mail response should focus on planning for key instrument and implementation evalu- ations in the 1995 census test and later tests. As the content of the instrument is made final, the respondent-friendly design of the final form will necessarily change and therefore need some further testing. The effects of a modified ex- tended roster form, for example, on both the initial response rate and the differen- tial undercount could differ from estimates obtained in the SQT or the National Coverage Test; the 1995 census test could augment the results of the earlier tests to include information on the effects of incorporating these innovations in the instrument. The respondent-friendly instrument design also assumes that the needed technologies for capturing data i.e., extracting relevant information from individual census forms will be available, and this work must also be pursued. In addition, the methods to be used for data capture constrain how respon- dent-friendly an instrument can be. Because the technology for data capture to be used in 2000 is still in development, the overall design of the instrument cannot be finalized. The breakthroughs needed for automated data capture of the respon- dent-friendly instruments tested in this research program may not materialize for the 1995 census test or by the year 2000. In that case, the initial response rate improvements projected for the respondent-friendly instruments may prove opti- mistic. Planning for operational implementation in the 1995 census test should draw on both research already conducted at the Census Bureau and other findings reported in the literature for example, on the use of telephone for reminder calls and follow-up-to test the best feasible implementation methods. USE OF THE TELEPHONE It is clear that the telephone will play a much larger role in the 2000 census than it did in 1990 or in previous censuses. New technologies will allow the Census Bureau to greatly expand and automate the 800 number call-in assistance program and to use computer-assisted outbound calling for the first time. Callers to the 800 number will be able to access a wide range of automated services from any telephone in the United States, and the Census Bureau will be able to place outbound calls to mail nonrespondents to prompt them to return the census form or to complete it over the telephone. Telephone interviewing will also play a key role in the integrated coverage measurement program. First offered in 1990, the Census Bureau's 800 number questionnaire assis- tance call-in program fell far short of providing prompt and efficient assistance to callers. According to a technology assessment commissioned by the Census
RESPONSE AND COVERAGE 63 Bureau, 4 million call attempts were recorded during the first week after Census Day, but less than 50 percent were answered due to high peak volumes (Ogden Government Services, 1993b). In total, the Census Bureau estimates that about 7 percent of all households tried to call the 800 number in 1990, with most of the calls concentrated in a short time frame. The shortcomings of the 800 number call-in system quite possibly contributed to the drop in the mail response rate in 1990. Telephone technology has made great strides since 1990, and there is little doubt that the telephone infrastructure will be able to meet the operational re- quirements of the 2000 census design both reliably and efficiently. The one caveat is that the Census Bureau must be able to provide precise estimates of the calling volume and its distribution by time of day and day of week (Ogden Government Services, 1993a). Fortunately, emerging related technologies such as touch-tone data entry, voice recognition, and automated voice recording will make it possible to provide a wide range of services to callers without human intervention, up to and including the conduct of a computer-administered inter- view without assistance from a live operator. A major innovation for the 2000 census will be the use of computer-assisted outbound calling. Outbound calling is possible because of the availability of electronic directory services that can match telephone numbers to addresses. Thus, it will be possible to add telephone numbers to the Master Address File (MAP) for a significant percentage of MAP address listings and to use this resource to make outbound telephone calls both to prompt mail nonrespondents to return their forms and to complete the enumeration by telephone. The use of computer-assisted telephone interviewing (CATI) to complete enumeration with mail nonrespondents in the nonresponse follow-up component of the census offers several advantages over the alternative of follow-up by field enumerators. The cost of a census telephone interview is estimated at $3.55 compared with $10 for an interview completed by a field enumerator (Ogden Government Services, 1993a). These estimates are compatible with the survey research literature, which has consistently found telephone interviews to be less expensive than comparable face-to-face interviews (see, for example, Weeks et al., 19839. Also, the literature clearly indicates that CATI provides enhanced data quality when judged against comparable paper-and-pencil surveys (see, for ex- ample, Weeks, 1992~. In the context of the census, using telephone follow-up when possible would also reduce the number of enumerators required and could potentially improve the quality of their work by enabling the Census Bureau to be . . . . . more selective in its recrultmg process. The integrated coverage measurement program that will be tested in 1995 will also use CATI whenever possible to conduct reinterviews with census re- spondents for purposes of measuring within-household coverage. Households without telephones in the integrated coverage measurement sample will be con- tacted by a field enumerator, who will use a laptop computer to conduct the
64 COUNTING PEOPLE IN THE INFORMATION AGE reinterview. The computer assistance will enhance data quality under either mode, although the use of CATI from a centralized telephone facility will offer significant cost savings over field visits by enumerators. These telephone applications offer the potential for considerable cost savings and data quality improvements, but they also raise a number of operational and methodological issues. We comment on these issues as we describe the new telephone applications in more detail. Inbound Calls We understand that the Census Bureau's goal for the 2000 census is to offer the public a questionnaire assistance call-in program through a single integrated telephone network so that anyone can access the system from any telephone in the United States by calling a single 800 number. Services will be accessed through a menu system, either by touch tone or voice recognition; for example, the caller will be instructed by the system to either press the "1" key or else to say the word "one." The Census Bureau seeks to offer as many services as possible without human intervention. However, there will always be an option for the caller to exit the menu system to speak to an operator. The 800 number will be printed on every census document and widely adver- tised through the media campaign. The services offered will include prerecorded messages that address a long list of general questions, as well as operator assis- tance in completing the census forms. The promotion surrounding the 800 num- ber, however, will focus on questionnaire assistance rather than on calling to complete the form over the telephone. While telephone interviews are less ex- pensive than face-to-face field interviews, the mailback response mode is by far the least expensive of the three modes, and the Census Bureau learned in the Mail and Telephone Mode Test that there is nothing to be gained from offering a telephone response as an alternative to a mail response. Although the Census Bureau will not encourage a telephone interview with an 800 number caller, the interview may be conducted by telephone as a last resort that is, if the caller indicates that he or she will not, or cannot, respond by mail. In this circumstance, the Census Bureau operator will ask the caller if he or she will consent to a computer-administered interview. If the caller agrees, the interview will be conducted using voice recognition, with support from auto- mated voice recording if necessary. If the caller declines the computer-adminis- tered interview, the operator will proceed to administer a CATI interview. Com- puter-administered interviewing with voice recognition technology will probably not be ready for testing in 1995 and therefore should be included in subsequent tests conducted by the Census Bureau. The Census Bureau has commissioned the Spoken Language Research Labo- ratory at Oregon Graduate Institute to undertake research on the use of voice recognition and voice recording to conduct a census interview. Although the
RESPONSE AND COVERAGE 65 results are not yet in, this approach was recommended for development in the Census Bureau's technology assessment (Ogden Government Services, 1993a), and it seems reasonable to assume that the combination of voice recognition and voice recording will be technologically feasible in time for possible use in the 2000 census. When conducting the interview with the 800 caller, the voice recognizes will assess whether it can recognize the caller's speech. If not, it will automatically refer the respondent to a live operator. As a further safeguard, callers will be told at the beginning of the interview that they can switch to a live operator if they encounter problems interacting with the automated system. The voice recording utility is involved when the caller provides long or complicated answers. In this circumstance, the voice recognizes will automati- cally turn on a recorder to capture the respondent's response. This action will not be apparent to the respondent. Later, after the interview has been completed, the recorded responses will be keyed by data entry operators and matched with the respondent's interview record. The Census Bureau is working with the current Federal Telecommunications System (FTS) contractor, AT&T, to develop the 800 number call-in system. FTS is a government network of leased long-distance telephone circuits that are paid for on a monthly contract basis rather than by the minute. A specific FTS contract was awarded to AT&T by the General Services Administration. In the 2000 census, the long-distance carrier will handle the initial receipt of calls, the presentation of the menu system, and the playing of prerecorded messages. The only calls that will be forwarded to the Census Bureau (or its contractors) will be those requiring a live operator or a voice recognition interview. The Census Bureau anticipates using multiple contractors in the 2000 census to provide live operators and computer-administered interviews. In the 1995 census test, plans are to use two telephone centers-one operated by the Census Bureau and one provided by a contractor. For all 800 calls referred to an operator, the callers will be asked for their name, address, and phone number. With this information, the Census Bureau can contact people again by telephone if they do not retun1 the mail questionnaire. For 800 callers who say that they did not receive a form, the operator will check their addresses against the MAP to verify that they can be geocoded. In 2000, this check will be done on-line; in the 1995 census test, it will be done manually after the call is completed. If an address cannot be geocoded, the operator will call the person back to try to get better address information. In 2000, the goal is to be able to verify addresses against the MAF and to geocode the address while the caller is on the phone. With this capability, the operator can probe for a codable address during the initial telephone conversation. We are impressed with the scope of the research program to design and test the optimal 800 number call-in system. Census Bureau staff appear to have explored the relevant issues in depth, and they have sought and re
66 COUNTING PEOPLE IN THE INFORMATION AGE ceived expert advice from knowledgeable sources both from within and from outside the Census Bureau. We commend the Census Bureau for recognizing the importance of developing a state-of-the-art 800 number call-in system for the 2000 census and for launching a thorough research and development program to achieve this goal. In implementing such a system, however, there are issues that we believe will deserve further consideration. One obvious concern is the public's reaction to yet another menu-driven call-routing system. We understand that AT&T has conducted research on the public's tolerance of menu systems and has developed guidelines on the number and length of menus. However, the technology is still very new, and we suspect that a caller's patience when interacting with call- routing menus is highly correlated with the importance he or she attaches to the purpose of the call. Consequently, the limited research in this area may reflect the idiosyncratic features of the specific applications investigated, and the find- ings may lack generalizability to other applications. Therefore, we encourage the Census Bureau to undertake research designed to test and evaluate the public's reaction to alternative menu configurations. If possible, experiments addressing this issue should be included in the 1995 census test. Another concern is the public's willingness to participate in an interview administered by a computer instead of a live interviewer. Although callers may initially agree to the interview, they may get frustrated or upset as the interview progresses and simply hang up, even if they are given the option of summoning a live interviewer. A related concern is the possibility of differential mode effects between computer-administered and interviewer-administered interviews. For example, respondents cannot easily ask questions when they do not understand an item on the census form if they are interacting with a computer instead of with an interviewer. These issues need to be resolved through the Census Bureau's research program. The 1995 census test will provide a unique opportunity to collect data on the operational and cost aspects of the call-in system. To gain maximum benefit from the test, the Census Bureau will need to develop and implement a compre- hensive monitoring plan that will capture relevant data on all aspects of the system. Some examples of items that will require monitoring include the number and timing of call attempts, network processing of the calls, disposition of each attempt, frequency of use of menus and menu options, reaction to the offer of a computer-administered interview, results of the computer-administered interview, comparison of the computer-administered interview and the operator-adminis- tered interview, reaction to the request for the caller's name, address, and phone number, and timing and cost data on all facets of the system. The Census Bureau should also consider conducting a follow-up survey of caller satisfaction for a sample of 800 number calls. In all likelihood, the Census Bureau will have to contract with several ven- dors in the 2000 census to create a network of regional telephone centers capable
RESPONSE AND COVERAGE 67 of handling the anticipated telephone workload. The 1995 census test will pro- vide experience in working with an external telephone center and will inform the development of appropriate oversight and quality control procedures. The 1995 census test will also provide operational and cost data that can be used to estimate the number of telephone centers required for 2000 and the approximate cost involved in building a system to serve the entire nation. Recommendation 3.2: The Census Bureau should use the 1995 census test and subsequent tests to inform the design of the 800 number call-in system for the 2000 census. The Census Bureau should focus on the public's response to the menu-driven call-routing system, acceptance of the computer-administered interview, possible differential mode effects between a computer-administered interview and one administered by an interviewer, and the technical feasibility of administering interviews using voice recognition and voice recording. The Census Bureau should also develop and implement a monitoring system in these tests to collect operational and cost data on the call-in program. Availability of Telephone Numbers for Outbound Calling The Census Bureau is conducting research on the availability of telephone numbers for use in making outbound calls. This is important because the use of outbound calling to prompt people either to return their census forms by mail or to provide information over the telephone hinges on the availability of telephone numbers for MAP addresses. Commercial companies offer electronic directory services that could be used to match addresses on the MAP and provide associated telephone numbers. These companies buy files of telephone numbers from the regional Bell telephone com- panies. However, the telephone companies can release only listed numbers; state regulations prohibit the release of unlisted telephone numbers. To expand their coverage, the commercial companies supplement the Bell directory files with address and telephone number lists purchased from third parties (such as credit card companies). The first phase of the Census Bureau's research program in this area was designed to measure the ability of the commercial companies to match addresses and provide accurate telephone numbers. The Census Bureau selected one such company, MetroMail, and sent the company a sample of addresses drawn from the addresses used in the Census Bureau's response improvement research. MetroMail succeeded in matching 35 to 40 percent of the addresses and provided the telephone numbers for 25 percent of the sample addresses. In the next phase of this research, the Census Bureau will select about 20,000 addresses from households that did not respond to the National Coverage Test and will send the selected addresses to MetroMail, which will attempt to match
68 COUNTING PEOPLE IN THE INFORMATION AGE the addresses and provide the telephone numbers. The Census Bureau will then select a sample of about 1,500 matched cases with telephone numbers, call the numbers, and try to complete the interview. This research will measure the accuracy of the address matches and telephone numbers. Interestingly, the study will also measure the willingness of nonrespondents to a census-type mail survey to participate in a nonrespondent telephone follow-up survey; the results will be of interest to those involved in designing the telephone nonresponse follow-up survey for the 1995 census test. The third phase of this research will involve an evaluation of various ad- dress-matching algorithms. The Census Bureau plans to purchase the directory files from the four regional Bell telephone companies that serve the four 1995 census test sites and then apply various address-matching algorithms to deter- mine whether the results achieved by the commercial companies can be im- proved. The fourth phase of this research program will involve legal research on the right to privacy of telephone subscribers. We understand that there are no federal regulations concerning the release of unlisted telephone numbers. Rather, the privacy guarantees that telephone companies give their subscribers who request an unlisted number are based on agreements between the local companies and the state regulatory agencies. In certain circumstances, it may be possible for the federal government to obtain an exemption from the state regulations. The Cen- sus Bureau is considering asking one of the states involved in the 1995 census test for a one-time exemption so that unlisted numbers could be provided for use in conducting nonresponse follow-up by telephone. The research program outlined above addresses the major issues in this important area. Some suggestions follow that we believe would enhance this program. First, we suggest that the Census Bureau gain more experience with the companies providing electronic directory services. The matching services are relatively inexpensive and could prove to be very cost-effective when compared with the cost of field follow-ups by enumerators. At a minimum, the MAFs for the 1995 census test sites should be sent to several companies for processing. The Census Bureau could then compare the results achieved by the individual companies and could add as many telephone numbers to the MAP as possible for use in the 1995 census test. The investigation of address-matching algorithms is interesting. If the com- panies providing electronic directory services will supply the Census Bureau with the algorithms they use, these algorithms could be evaluated as well. The product of this research could be the development of an optimal matching proto- col that could then be given to the commercial companies for their use in process- ing the MAFs. The legal research is probably worthwhile; however, it raises a concern about the potential effect on public relations of calling people with unlisted telephone numbers to prompt them to return the census form or to complete a
RESPONSE AND COVERAGE 69 telephone interview. Some people may consider such a call an unwarranted invasion of their privacy and may refuse to participate in the census. If the Census Bureau is successful in acquiring unlisted telephone numbers from the regional Bell telephone companies, the results obtained from calling these house- holds should be carefully monitored and compared with results of calls to house- holds with listed numbers. Recommendation 3.3: The Census Bureau should expand the research program involving the acquisition of telephone numbers for MAF ad- dresses by working with more companies that offer electronic directory services and developing an optimal protocol for matching addresses. If the Census Bureau is able to acquire unlisted telephone numbers for a 1995 census test site, it should carefully monitor the results obtained from calling households with unlisted numbers. Outbound Reminder Calls New technology now exists that makes it possible to conduct computerized outbound calling. Using what is called predictive dialing technology, the com- puter dials numbers automatically and senses when a person answers. At that point, the system can either play a prerecorded message or else route the call to a waiting operator. Predictive dialing improves the efficiency of large-volume calling efforts by automating both the dialing process and eliminating nonwork- ing and no-answer numbers. Predictive dialing can also eliminate human inter- vention altogether if prerecorded messages are used. The Census Bureau has considered using predictive dialing for making prompting calls in lieu of sending a reminder postcard. Given the anticipated volume of calls, the Census Bureau would have to use prerecorded messages instead of live operators, although the people who are called would be given the option of requesting a live operator. There are several problems, however, with using predictive dialing for out- bound reminder calls. The Census Bureau would like to reach an adult household member with the prerecorded message, but the current systems have trouble distinguishing when a child answers instead of an adult. This objective suggests the need for some type of automated screening procedure, which could be prob- lematic. Another problem is that some states prohibit calls with recorded mes- sages, and 12 states require the caller to ask permission before playing a recorded message. The MAP is likely to contain telephone numbers for only a small percentage of the addresses. Consequently, the use of random-digit dialing may be required to increase the coverage of households with telephones. However, random-digit dialing will also increase costs and reduce efficiency because most samples of random telephone numbers contain a large proportion of nonworking or nonresi
70 COUNTING PEOPLE IN THE INFORMATION AGE dential numbers. Even with a computerized system, there is some cost associated with calling out-of-scope numbers. Finally, there is concern about the relative effectiveness of a computerized phone message versus a reminder postcard. The latter is a well-established mail survey procedure that has been proved effective in the Census Bureau's own research as well as in the wider survey community. We know of no research indicating that a computerized phone message is more effective. The Census Bureau would need to undertake such research before deciding to use the phone message in lieu of the postcard. We understand that computerized outbound calling will not be included in the 1995 census test but will remain on the Census Bureau's long-range research agenda. In view of the problems associated with this procedure, we concur with this position. Use of CATI for Nonresponse Follow-up The Census Bureau plans to use computer-assisted telephone interviewing for nonresponse follow-up in the 1995 census test. Nonresponse follow-up will occur after the full set of mail survey procedures (advance letter, initial question- naire mailing, reminder postcard mailing, and replacement questionnaire mailing to nonrespondents). After the mail operations have been completed, the frame for nonresponse follow-up will be constructed (all mail nonrespondents as of the mail survey cutoff date), and a sample of nonrespondent households will be selected. The sample is expected to be about 33 percent of the total nonrespondent universe. Sample cases that have telephone numbers will be sent to the two telephone centers for follow-up calling. Cases without telephone numbers will be sent directly to the field offices for assignment to field enumerators. The field enumerators will also be assigned the residual of the telephone cases that could not be completed by telephone for other reasons (e.g., refusal, inability to con- tact). The 1995 census test offers a unique opportunity to examine a number of methodological, operational, and cost issues related to the use of CATI for nonresponse follow-up. A key task is to determine which calling protocols are most successful in obtaining the census data from the mail nonrespondents in the follow-up sample. There are several options when implementing telephone fol- low-up. For example, the interviewer could either prompt the contacted person to return the census form by mail or attempt to obtain the data over the telephone in a CATI interview. If a prompting approach is used, the interviewer would pre- sumably enter a follow-up date into the case management system, and the system would then schedule another follow-up call if the contacted person's form is not returned by that date. If the person no longer has the census form, the interviewer could either attempt to complete the interview or enter an order into the system to have a replacement form sent to the person. In this case, the system would
RESPONSE AND COVERAGE 71 schedule the case for a follow-up call if the form is not received by a specified date. There are other possible permutations for various types of cases. The point is that the Census Bureau will need to develop, implement, and evaluate calling protocols in the 1995 census test in order to develop the optimal set for use in the 2000 census. The optimal timing of calls to nonresponding households also warrants in- vestigation. There is a considerable body of literature on this topic (see, for example, Weeks, 1988; Kulka and Weeks, 1988; Weeks et al., 1987~; however, every survey application has its idiosyncrasies, and the Census Bureau would be well advised to build into the design of the l99S census test some controlled experiments in this area. Another question is when to send cases to field enumerators for follow-up by a personal visit. That is, at what point does it become more efficient to send the cases to the field instead of continuing to work the case by telephone? A related issue is whether to try to work the telephone cases in geographic clusters so that the residual can be sent to the field in clusters, thus saving on field travel costs. The use of both CATI interviews and paper-and-pencil field interviews in the nonresponse follow-up component of the 1995 census test raises concern about the potential for mode effects. The two interview modes are radically different in two respects: the use of the telephone versus a face-to-face interview, and the use of computer assistance versus an unassisted interviewer-administered question- naire. An important operational issue is the design of the automated case manage- ment system. This system must be able to initialize cases, schedule them for calling using preprogrammed protocols, prioritize cases across and within calling queues, assign cases to appropriate interviewers, close out completed cases, iden- tify cases that are ready to be assigned to the field, and produce a variety of monitoring reports. CATI case management systems vary considerably in their design and capabilities (see Weeks, 1988, for a discussion of various systems'. The Census Bureau should consider investigating a variety of extant systems, including those currently used in its own telephone centers, to determine the optimal design for use in the 1995 census test. As noted above, the 1995 census test will give the Census Bureau an oppor- tunity to gain experience in working with an outside telephone center. The test will also provide data that can be used to develop workload and cost estimates in planning the 2000 census. The Census Bureau should take care to design and implement a comprehensive monitoring system that will collect these operational and cost data. Using CATI in the Integrated Coverage Measurement Program The Census Bureau is developing an integrated coverage measurement (ICM) design that involves an independent reinterview with a sample of census respon
72 COUNTING PEOPLE IN THE INFORMATION AGE dents followed immediately by a comparison with the original census data and a reconciliation of differences. For several reasons, such a design suggests an ideal application for CATI. First, the telephone number for sample cases is likely to be available from the initial census form. Second, the data from the first interview can be preloaded into the computer, and the comparison of the first and second interviews can be done by computer. Finally, the computer program can be written to script customized questions and probes for use by the interviewer in reconciling the two interviews. In current plans for the 1995 census test, ICM interviews with no-phone households will be conducted via computer-assisted personal interviewing (CAPI), which involves equipping enumerators with laptop computers. Both CATI and CAPI will presumably be conducted using the same interview pro- gram. However, because CATI will be conducted by telephone and CAPI will be conducted by personal visit, the possibility for mode effects will also exist in this phase of census operations. The use of CATI in the ICM program offers the potential for significant cost savings in comparison to CAPI. However, the Census Bureau will need to design and implement a monitoring system that is capable of collecting the data neces- sary to quantify the relative cost of the two modes and to inform cost modeling for the 2000 census. OTHER AUTOMATED RESPONSE TECHNOLOGIES The Census Bureau is also evaluating a number of other automated response technologies. These are briefly reviewed below. As noted above, the Census Bureau plans to use CAPI (as well as CATI) in the ICM component of the 1995 census test. However, CAPI will not be used to conduct field follow-up interviews in the nonresponse follow-up component. There are several reasons for this decision. First, it would probably be prohibi- tively expensive to equip an army of field enumerators with laptop computers for the very short time the equipment would be used. Second, the use of CAPI would make it more difficult to recruit qualified enumerators, and those who were selected would have to receive an expanded training program. Finally, Census Bureau staff expect that they will continue to learn about CAPI through the Census Bureau's ongoing demographic surveys that are converting to CAPI (e.g., the Current Population Survey) and through the limited use of CAPI in the ICM program. The Census Bureau does plan to equip field staff with pen-based computers in the 1995 census test to update address lists and maps and to compile the independent listing of addresses in the ICM blocks that will be matched to the census address list (see Chapter 4~. The computers will be preloaded with geocode information so that addresses will be automatically geocoded as they are listed.
RESPONSE AND COVERAGE 73 The multimedia kiosk is another response technology under consideration. A prototype is currently under development, and the Census Bureau may test this technology in one of the 1995 census test sites. The Census Bureau is part of a consortium with several other federal agencies (the Social Security Administra- tion, the Internal Revenue Service, the Postal Service, and the Department of Veterans Affairs) considering a potential joint investment in a network of kiosks. Kiosks would be located in federal buildings, post offices, Social Security of- fices, etc., to provide information about government services. The Census Bu- reau could use this network to provide information to the public during the decennial census. The technology assessment of interactive cable television commissioned by the Census Bureau concluded that this technology will not be mature enough by 2000 to warrant consideration as a data collection method (Ogden Government Services, 1993a). However, interactive cable television is changing rapidly, and the Census Bureau plans to continue to monitor the progress of this technology. The technology assessment of personal computers (PCs) is also not encour- aging. It is estimated that less than 40 percent of households will have PCs by the year 2000, and only 12 percent will have modems (Ogden Government Services, 1993a). Consequently, the Census Bureau does not plan to experiment with PCs as a response option in the 1995 census test. However, this response mode remains on the long-range research agenda. One possible project would be to establish a computer bulletin board through an existing on-line service. The technology assessment of reporting from businesses recommended that the Census Bureau conduct a national survey of businesses to assess their willing- ness to promote the census and to allow their employees to use business tele- phones, fax machines, and modem-equipped PCs as response modes. The Cen- sus Bureau will not pursue this activity in 1995, but the option remains on the long-range research agenda. In our opinion, the Census Bureau has made a commendable effort to iden- tify and evaluate alternative response modes. The technology assessments have been comprehensive and thorough, and the Census Bureau's responses have been appropriate. We encourage the continuation of this line of forward-looking re- search. HARD-TO-ENUMERATE POPULATIONS As noted above, neither improvements in questionnaire design and imple- mentation nor the development of automated response technologies is likely to address particular difficulties that historically have hampered attempts to count certain subpopulations. In this section, we review what the Census Bureau has learned in recent years about the problems of enumerating inner-city and rural low-income populations, immigrants, internal migrants, and homeless people ("persons with no usual residences. We then discuss some of the remedies the
74 COUNTING PEOPLE IN THE INFORMATION AGE Census Bureau is considering for evaluation during the 1995 census test, with a view toward reduction of the differential undercount in the year 2000. Challenges in Counting Poor and Migrant Populations In the course of the 1990 census, the Census Bureau conducted an alternative enumeration in 29 sample areas of the United States and Puerto Rico called the Ethnographic Evaluation Project. The areas were chosen for their high concen- tration of particular minority subpopulations, both urban and rural, as well as the presence of a large number of undocumented immigrants who are known to fear participating in the census. Experienced ethnographers familiar with the target populations conducted the alternative enumeration. They were able to improve the count by using their existing relationships with members of these communi- ties, familiarity with the native languages of the immigrants, and their knowledge of the particular housing practices of poor populations both native and immi- grant. Results of these special census projects were compared with the enumera- tions of the same areas in the 1990 census. From this comparison the Census Bureau learned a great deal about the sources of undercount and overcount. Substantial numbers of erroneous enu- merations and omissions occurred at the sites, and the net undercount rates varied considerably (Brownrigg and de la Puente, 1993~. The two extreme cases re- ported a net undercount of 47 percent and a net overcount of 53 percent; the median net coverage was a small undercount of approximately 1 percent. Clearly, the sites present very difficult counting conditions that create both types of cen- sus error. These errors occasionally cancel out within a geographic area, but they are more likely to result in surpluses of omissions or erroneous enumerations that produce the large variations in net undercount that were observed in 1990. For the affected communities, undercounting is a source of great policy concern because of implications for political representation and allocation of federal and state program funds. The primary task now should be to capitalize on this research and to implement the valuable policy suggestions that emerged from this study in designing the 1995 census test and developing the tool kit of methods for the 2000 census field offices. An excellent summary of the 29 ethnographic coverage reports (de la Puente, 1993) identifies five sources of undercount or overcount: (1) irregular and com- plex living arrangements, (2) irregular housing, (3) residential mobility, (4) dis- trust of government, and (5) limited English proficiency. Almost all of these point to the difference between the underlying norms expressed in the census (nuclear families, discrete households, long-term or permanent residences) and the living arrangements of poor people in this country, whether foreign-born or native. Some of these problems are related to the difficulty of enumerating the population in question: gaining trust, counting all the people in a household, and the like. Others are related to locating dwellings in poor areas of the nation's
RESPONSE AND COVERAGE 75 cities and rural communities and are of great importance in updating the Master Address File. The Census Bureau will need to give serious consideration to the policy options suggested by the ethnographic studies for both of these domains. Irregular and Complex Household Arrangements The first source of enumeration difficulties involves the irregular and com- plex household arrangements that typify poor minority and immigrant communi- ties. Households defined as irregular or complex contain unrelated individuals, people who are mobile or present for no other reason than to share the burden of the rent, and multiple nuclear families. Households structured in this fashion become extremely difficult to enumerate accurately. Census rules of residence, which ask the respondent to identify members of their household in relation to "person l," cannot easily accommodate the composition of these living groups and often contradict the respondent's definitions of family or household. In such cases, residence rules are hard to apply; the individuals who "should" be counted may be absent and thus excluded. As an ethnographer working in rural Marion County, Oregon, noted (cited in de la Puente, 1993:43: If all members are not present, . . . obtaining the data pertaining to persons outside, asleep, at work, or temporarily absent is virtually impossible. It is as if those persons do not exist. These households may also contain unrelated males who have assembled solely to cut the cost of the rent, who work long hours, and share living space in very dense fashion with beds lining the rooms and individuals sleeping in shifts.2 Nothing, other than expediency and common ethnicity, binds these indi- viduals into anything they would define as a household of people related to person 1, with emphasis on the term related. As Rodriguez and Hagan put the matter (cited in de la Puente, 1993:7~: For recent immigrants from Central America and Mexico, household and fami- ly are viewed as the same. Boarders and unrelated individuals are not part of the family and thus not part of the household. At some sites, ethnographers discovered that up to half of the individuals living in complex households of this kind were left out of the census (see, e.g., Romero, 1992~. The same pattern of undercount was likely to develop in Haitian house- holds, which were composed of a "core" family and a series of peripheral indi- viduals who are "just passing through" but sometimes stay for years. Chinese families in New York may be equally perplexed by the census definitions of household and may augment those present in a household with relatives who are 2 In some immigrant communities, this kind of dense subleasing is an important source of income for older immigrants, who build a housing pyramid on the basis of holding a lease and exploiting recent arrivals who have no other options (see Mahler, 1993).
76 COUNTING PEOPLE IN THE INFORMATION AGE family (e.g., adult children) but no longer living in the house. Again, the resi- dence rules on the census forms confuse even the willing participant resulting in undercounts (the Haitian example) and overcounts (the Chinese example). Irregular and complex household structures are generally to be found wher- ever there are high housing prices and a concentrated poor population (native or immigrant). This is an important finding in and of itself, for researchers (perhaps including the ethnographers working on this project) are prone to believe that these patterns are expressions of cultural differences. The regularity of the ethno- graphic findings across ethnic subpopulations and in disparate parts of the United States suggest there may be little that is culturally specific about the formation of irregular households. Rather, irregular households form in response to structural conditions that bring poor people into areas of expensive housing, resulting in these fairly uniform "irregular" practices. If this is true, then it should be possible for the Census Bureau to identify many of these areas in advance and target special resources for enumeration within them. Irregular Housing Irregular housing is the second major source of the undercount, one that presents difficulties in locating housing units, rather than in enumerating the individuals living in them. Brownrigg (1991J estimates that across the 29 sample areas subjected to ethnographic evaluation, perhaps as many as 40 percent of the people who should have been included in the count, but were missed, were missed because the housing unit itself was overlooked or misidentified. Typi- cally, these dwellings were hidden from public view (in backyards or down rural roads) or were illegally built (and often concealed in single-family homes or garages). Irregular housing also causes overcounts because buildings have mul- tiple addresses or multiple entrances or are temporary dwellings that move around but have already been counted elsewhere~.g., trailers in the neighborhood just for the weekend (see de la Puente, 1993:12~. Irregular housing goes hand in hand with irregular and complex households, thus compounding the enumeration prob- lems discussed above. Families are likely to double up in areas where affordable housing is in short supply, stimulating the construction of illegally converted housing.3 The success of the ethnographic teams in locating irregular housing was impressive, particularly when compared with the 1990 census. One case study reported a census undercount of nearly 50 percent, much of which could be attributed to missed housing units. The alternative enumeration was often able to identify these irregular dwellings, drawing on the detailed knowledge of locally based ethnographers and their skill in cultivating informants who could guide 3 Descriptions of these areas leave little doubt that they can be forbidding places for enumerators to survey. Ethnographers were sometimes exposed to threats to their physical safety.
RESPONSE AND COVERAGE 77 them into the nether world of warehouses, back alleys, unscrupulous landlords, crack dens, and other unlikely places for private housing. Residential Mobility Residential mobility was the third reason given for enumeration difficulties in the 29 sample areas. Some populations that the census generally undercounts tend to move often, to reside in one place for a shorter period of time than middle- class citizens think of as the norm, and so to be difficult to enumerate. High residential mobility goes hand in hand with seasonal and low-wage labor mar- kets, patterns of return migration, and the pressures of accommodating large numbers of "peripheral" members in a household (who are likely to move when the strain becomes too great). Difficulties with landlords give rise to eviction, resulting in greater mobility for poor people than others. Mahler (1993:8, quoted in de la Puente, 1993:22) notes that Salvadoran immigrants move as often as three times a year in search of cheaper housing or jobs or because they are reunited with family members. These circumstances not only make it difficult to count individuals but also exacerbate the difficulties that many native-born and immigrant poor people face in applying the residence rules. Are temporary household residents "members"? They may be deemed such if they are also relatives, but otherwise they may be excluded because they are both highly mobile and unrelated. As noted above, experimentation with questions that ask "Who stayed here last night?" has al- ready yielded some interesting results.4 New formulations of the roster questions might identify a significant percentage of individuals who have traditionally been missed in the census. Research from the Living Situation Survey (Sweet, 1994) should inform the development of roster questions for the 1995 census test and the 2000 census. Distrust of Government Distrust of, or ambivalence toward, the government is another cause of undercounts. In the 29 communities surveyed, many members of the target populations believe they have little to gain in cooperating with the census and fear the possible consequences of yielding information particularly if they are among the large number of undocumented immigrants or participants in the underground economy. Contending with this problem proved difficult for the ethnographers who worked on this special project, for they were rarely equipped 4 Cognitive research aimed at uncovering the connotation of phrases such as lived here or stayed here has been helpful in illuminating possible causes of within-household omissions. Regional differences may play a role in divergent personal interpretations of census roster questions, with potential implications for differential response and coverage (see Gerber and Bates, 1994).
78 COUNTING PEOPLE IN THE INFORMATION AGE to survey drug enclaves, shanty towns, and the like. They had to contend with the dangers of the street and, perhaps more important, the hesitance of ordinary (law- abiding) individuals living under these conditions to open their doors to strangers. It is clear that many people in immigrant and native-born minority groups do not believe that the census is confidential. Those who harbor an undocumented individual in the household are especially fearful of discovery through participa- tion in the census; others who are receiving public assistance are concerned about the government finding undeclared partners. It also seems likely that some immigrants would be wary of government initiatives to count individuals and skeptical about confidentiality claims because of their negative experiences with government activities in their native countries. Limited English Proficiency Finally, the ethnographic studies found that lack of English proficiency was a major source of undercounts in some communities. The studies made use of native-speaker enumerators who could conduct the census in Haitian creole, Spanish, Chinese, and a variety of other languages. Multilingualism proved to be enormously important, both enabling effective communication and inducing the respondent to trust the enumerator. In addition, immigrant populations are often characterized by a high incidence of illiteracy in the native tongue. Ethnogra- phers sometimes found that members of their target population were unable to read or write in any language. Illiteracy necessitates even more intervention or interpretation in the enumeration process and undoubtedly introduces standard- ization problems. Coverage was undoubtedly improved in these case studies; consistency may be another matter. Persons With No Usual Residence In recent years, research attention has focused on the problems of enumerat- ing the homeless population, a subset of persistently poor people who have no regular domicile. This population is of particular concern to city agencies who have to provide services and who, in turn, depend on an accurate count in order to budget for their needs. Advocacy groups have also played an important role in drawing attention to homeless people. The Census Bureau adopted new measures for enumerating persons without a usual residence in the 1990 census. Based on prior research, the Census Bureau assumed that these people would be found in one of four situations: (1) living temporarily in someone else's household (often referred to as "doubling up"~; (2) living in facilities such as commercial campgrounds, hotels, and boarding houses or noninstitutional group quarters; (3) staying temporarily in shelters or other emergency facilities (or cheap hotels); or (4) spending the night "in the street" (on sidewalks, in vehicles, in abandoned buildings, etch. The 1990 census
RESPONSE AND COVERAGE 79 counted people in the first situation via the regular household enumeration pro- cess, depending on the wording of the residence rules to "capture" those who were temporarily located in a given household. For the second situation, census procedures for enumerating individuals living in group quarters were invoked. In the third and fourth categories, a special census was conducted in March of 1990 that has come to be known as S-Night, short for Shelter and Street Night (Kalton etal.,l994:2-2~. S-Night convinced the Census Bureau that major methodological problems attend the attempt to enumerate the street population. Using various controls, the Census Bureau determined that enumerators often failed to locate many of the preidentified sites where the homeless congregate, that departures from standard procedures were common, and that enumerators were selective in approaching individuals because of safety concerns. Policy Initiatives for 2000 and Future Censuses It is important to remember that the Ethnographic Evaluation Project was conducted in particularly hard-to-enumerate neighborhoods. Nonetheless, the project sheds light on the causes of undercounts and overcounts and suggests valuable policy initiatives for improving census coverage of such neighborhoods. Similarly, the Census Bureau's efforts to improve methods of counting the nation's homeless population have yielded some important insights that will make the 2000 census more complete, even for those who are the hardest to enumerate. We emphasize that the insights developed thus far should evolve into (a) further research programs, if needed, and (by experiments to be undertaken dur- ing and after the 1995 census test to evaluate the cost-effectiveness of special methods designed to reduce the differential undercount. Strategies suggested for possible use by headquarters and by regional offices make insightful use of the lessons contained in the studies of hard-to-enumerate populations. At present, Census Bureau staff are in the process of evaluating these strategies in order to determine which ones might be incorporated into the 1995 census test. Our discussion and recommendations focus on the following areas of con- cern: (1) creating ongoing local ties, (2) further comparative studies of hard-to- enumerate populations, (3) strategies for reducing differential undercount, (4) counting persons with no usual residence, and (5) cognitive research on race/ ethnic classification. Creating Ongoing Local Ties The ethnographic research suggests that individuals familiar with the native languages, the customs, and the physical layout of the communities in question were far more successful than traditional enumerators in locating and surveying
80 COUNTING PEOPLE IN THE INFORMATION AGE hard-to-count populations. The project directors concluded, and this panel con- curs, that the Census Bureau should create a more effective partnership with locally based, grass-roots organizations in order to address the coverage problem. Some of these organizations have regional and even national representation (e.g., the Urban League, the Mexican American Legal Defense Fund); others are en- tirely local in orientation. These two kinds of ethnic organizations appeal to different groups and would require different partnership strategies with the Cen- sus Bureau, but both must be approached and integrated into the 2000 census. Effective partnerships between the Census Bureau and local leadership will have to be inaugurated well before Census Day. A well-designed continuous outreach, promotion, and enumeration operation in targeted areas may prove cost-effective in comparison with the start-and-stop approach that has been used in past censuses. Developing local ties that work that is, that help to reduce the differential undercount by creating a participatory spirit or simply by using the superior knowledge of local ethnographers is not an overnight operation. If such efforts prove effective as they have, to some extent, for Statistics Canada with regard to the Canadian aboriginal population-they should pay off in terms of reduced undercount and more effective targeting of Census Bureau resources. Implementing an in-depth, localized, network approach will demand an enor- mous change in the culture of the Census Bureau, for which, for perfectly under- standable reasons, centralized control and standardized methods have been para- mount. Involvement of community organizations at the level suggested by the ethnographic studies will mean yielding a degree of control and fostering a sense of trust that would probably be unparalleled in the agency's history. Community leaders speak of the importance of defining the census as "theirs," the critical need to "own" the census, rather than approaching the census as an instrument that belongs to the government or to the nation as a whole. It is important to recognize how revolutionary "going local" on this scale would be. At the same time, the survival of many community organizations depends on the census, because their funding is linked directly to the size of the populations they serve. If for no other reason than self-interest, many Native American, Haitian, African-American, Hispanic, and Pacific Islander social service agen- cies are eager to help. Grass-roots groups have infrastructures that can be of enormous assistance in the enumeration process; they know where to find their people, they have lists of their members, and they are trusted in the community. These are assets that cannot be acquired by outsiders, and they are directly related to a successful enumeration. One concern with grass-roots efforts, however, is the potential for increases in erroneous enumerations and consequent over- counting when the involved organizations may be interested more in the size rather than the accuracy of the count. An interdivisional working group at the Census Bureau has suggested re- gional office activities that involve ongoing local contact. In particular, the working group has recommended that continuous outreach in hard-to-enumerate
RESPONSE AND COVERAGE 81 areas be undertaken at least two years before decennial Census Day (Bureau of the Census, 1994b): The outreach staff would engage in such activities as kdeveloping] an on-going presence in the schools, establishing and maintaining rapport with local media, educating and gaining entree into local community groups, and making speech- es and presentations to any groups [that] could help with the census enumera- tion in the area. Recommendation 3.4: The Census Bureau should consider developing an extensive network of relations between field offices and local commu- nity resources, particularly in hard-to-enumerate areas, and should ex- amine the cost-effectiveness of maintaining this infrastructure in con- tinuous operation between censuses. The Census Bureau should develop and implement pilot programs in conjunction with the 1995 census test in order to gather information about the potential costs and benefits of a large-scale local outreach program. Further Comparative Studies of Hard-to-Enumerate Populations There is a tendency to assume that irregular households are the product of culturally specific notions of family, but the evidence from the case studies suggests a surprising degree of uniformity in the residence practices of the nation's poor, whether immigrant or native born. Further research is needed, and results must be incorporated into the targeting model and into any component of the census design that depends on comprehensive coverage (e.g., as part of integrated coverage measurement). A theoretical understanding of the distribution of hard- to-enumerate populations is needed, one that looks across ethnic and racial groups to the common characteristics of poor populations; that theoretical knowledge is of considerable practical value in directing additional resources to field offices where differential undercounts are likely. The same modeling may help in deter- mining the most effective placement of the standing infrastructure discussed in Recommendation 3.4. Recommendation 3.5: The Census Bureau should conduct further com- parative studies of hard-to-enumerate areas, focusing on those parts of the country where three phenomena coincide: a shortage of affordable housing, a high proportion of undocumented immigrants, and the pres- ence of low-income neighborhoods. Strategies for Reducing Differentials in Coverage Ethnographers who were involved in the 1990 studies generally believe that enumerators who reflect the racial, ethnic, and cultural composition of the target population are better able to conduct the census collection than those who are
82 COUNTING PEOPLE IN THE INFORMATION AGE defined as outsiders. Some of the advantages are trust, communication skills (especially foreign-language skills), knowledge of the housing stock, and cre- ation of the appearance (and reality) of "local ownership" of the census. Commu- nities that are suspicious of the federal government or that see no reward from participation are more likely to be persuaded of the importance of the census if the words come from people they know and have reason to admire or trust. One potential disadvantage with community enumerators is that respondents may be reluctant to give accurate information about more sensitive items, such as in- come, to a friend or a neighbor. Local enumerators are more likely to know the ins and outs of the neighbor- hoods that feature irregular housing, but further training and support for locating hidden housing units will be necessary. The ethnographic studies recommend making more systematic use of letter carriers, not to conduct the count, but as sources for identifying housing units. Rental offices and landlords turn out to be poor sources of information, but letter carriers are disinterested parties whose knowledge is currently underused. Clearly a large-scale program of ethnic enumerators would require a signifi- cant change in traditional training methods. More resources would have to be available to train enumerators, and the process would take longer. It would be foolhardy to hire people and provide them with insufficient training. But the investment would be well worth it if the program contributed to reducing the differential undercount-particularly if it was tied to an ongoing organizational structure (which would not have to be reborn with every decennial census). Ethnographers working in Spanish-speaking areas noted the importance of developing Spanish-language census forms based on a conceptual rather than a literal translation and distributing these forms more effectively than in 1990, when people were required to call a toll-free number to request a Spanish-lan- guage form. They reported frustration on the part of Spanish speakers who were unable to participate even though they are literate in their native tongue. This appears to be a relatively simple and cost-effective change. Every form that can be completed by mail saves on the cost of follow-up. Although this was the only non-English-language form suggestion, further research on the efficacy of trans- lating the form into other languages might be useful. Further research is needed to assess whether the use of written information materials, foreign-language tapes, or interactive video might be effective ways of reaching individuals who are not literate in their native language. Immigrant families working shifts and seasonal jobs are often difficult to enumerate because their schedules are irregular. Ethnographers recommend that enumerators conduct visits in the evening and on weekends, rather than in the middle of workdays, in order to find more people at home. They also noted the importance of moving Census Day so that it falls in the middle of the month; occupants of low-cost housing are more likely to move at the beginning and the end of a month (see also the discussion in Chapter 2~. Hence, a change in the
RESPONSE AND COVERAGE 83 timing of the census itself will reduce differentials in coverage that are attribut- able to residential mobility (see Recommendation 2.3~.5 Canada has already changed its Census Day in 1996 from June 4 to May 14 for essentially the same reasons (Choudhry, 1992~. Officials concluded that some coverage problems could be solved simply by shifting the census reference date from the beginning of the month. Recommendation 3.6: In the 1995 census test, the Census Bureau should include a larger repertoire of foreign-language materials than those cur- rently available in Spanish (both written and audio). In addition, the Census Bureau should conduct more aggressive hiring of community- based enumerators (with due consideration of local concerns about the confidentiality of census responses) and should accommodate greater flexibility in the timing of enumeration by personal visit (i.e., permitting contact during evenings and weekends). Enumerating Persons With No Usual Residence Because of the methodological and operational problems encountered in the 1990 attempt to enumerate the street population (Martin, 1992), the Census Bu- reau is developing a service-based approach to the enumeration of persons with no usual residence (Bureau of the Census, 1994e). The development and testing of a service-based approach was recommended in a methodological review (Kalton et al., 1994) of studies of the homeless populations in several U.S. cities. The review was commissioned by the Census Bureau to evaluate possible strate- gies for enumerating persons with no usual residence in the 2000 census. Similarly designed surveys of the homeless population have been camed out in several major cities Los Angeles, Chicago, and the Washington, D.C., met- ropolitan area. The studies used multiple frames to sample places where people find social services (e.g., shelters and soup kitchens); coverage results ranged from approximately 51 to 92 percent of the estimated homeless population in these areas (Kalton et al., 1994~. These findings suggest that canvassing shelters and soup kitchens may be more cost-effective than a street-based approach to enumerating persons with no usual residence. Street enumeration might instead serve a supplementary role in estimating coverage in sample areas. Some type of coverage evaluation will be needed to judge the completeness of coverage of the homeless population that can be achieved by enumeration at service providers. It will also be important to assess 5 We note that the 1995 census test is set for March 4. It is not clear whether the Census Bureau plans to analyze the impact of moving Census Day from April 1. We hope that some kind of evaluation will be considered, although the March 4 date may still be too early in the month to result in appreciably fewer problems with end-of-month movers.
84 COUNTING PEOPLE IN THE INFORMATION AGE the variation among communities in rates of service utilization because of the implications for census coverage of homeless people via a service-based ap- proach. Results suggest that service usage rates vary considerably across geo- graphic areas and across time of year; possible explanations for this variation include differences in weather and in the level of available services (Kalton et al., 1994). Recommendation 3.7: We endorse the Census Bureau's plans to con- duct, in the 1995 census test, enumeration at service providers (e.g., shelters and soup kitchens) as a method for counting persons with no usual residence (and possibly migrant workers). The Census Bureau should consider conducting enumeration of streets and other public places on a sample basis at each of the test sites for the purpose of coverage assessment. Further research will be needed to determine whether service-based enu- meration is the most cost-effective approach to counting the homeless popula- tion. Finally, we observe that any enumeration method would likely benefit from greater community outreach to identify locations in which persons with no usual residence might be found. Cognitive Research on Race/Ethnic Classification The racial and ethnic categories used in the census must be consistent with federal standards established by the Statistical Policy Office of Office of Man- agement and Budget (OMB). The OMB standards take into account legal re- quirements for race and ethnicity data, as well as user needs; the standards apply to all federal statistical agencies and to federal agencies that maintain administra- tive record systems. Information about race and Hispanic origin is collected on the short form of the decennial census; the long form collects additional informa- tion about such characteristics as ancestry or ethnic origin, country of birth, language, and citizenship status. Our discussion in this section is concerned primarily with the short-form questions that are asked of everyone in the census. Ethnographers noted the increasingly outdated conceptualization of race and ethnic identity embodied in the census. America is becoming a multicultural society, and personal identities are changing along with this trend. The selection of single-race identifiers is proving particularly problematic under these circum- stances. Are immigrants supposed to define themselves by nationality, ethnicity, language group, or physical appearance? Are individuals supposed to select a category that expresses how they view themselves or how others define them? This is a source of great confusion. Laotian nationals of Chinese descent and Hmong origin, for example, have difficulty in knowing how to respond. Spanish- speaking Filipinos are unsure whether they should check "Asian or Pacific Is- lander" for race and then "Hispanic" for origin.
RESPONSE AND COVERAGE 85 Questionnaire items that apply OMB guidelines and ask respondents to claim a single race suggest a conceptualization of race (and ethnicity) as something fixed, inherited, and unambiguous. But contemporary racial and ethnic self- identification seems to be more fluid, situational, and continuum-like in its cogni- tive organization. The United States is starting to look more like Brazil or Tahiti, among the many places in the world where fluid racial and ethnic identity is the norm and where single-category identifications are not easily reconciled with personal conceptions of race. As immigration patterns bring more people from these societies (e.g., the Dominican Republic) to the United States,. patterns of identification are shifting. The Census Bureau should continue to support research on questions about racial and ethnic identity6, and policy makers should clarify the purpose of such questions. Recent and ongoing activities have begun to address these issues. The Census Bureau and Statistics Canada jointly sponsored an international confer- ence on the measurement of ethnicity in April 1992 (Statistics Canada and Bu- reau of the Census, 1993~. OMB has convened an interagency committee to study federal standards on the collection and reporting of race and Hispanic origin data by federal agencies. At the request of OMB, the Committee on National Statistics held a workshop in February 1994 to review federal experi- ence with these standards. Also, the Panel on Census Requirements in the Year 2000 and Beyond is investigating the subject of racial and ethnic classification with regard to content needs in the decennial census and other demographic programs. The complexity of racial and ethnic identity causes methodological problems of racial and ethnic misclassification in the decennial census. The effects of misclassification on differential undercount are not well-understood. But the Census Bureau will continue to feel pressure to solve these problems as long as resource flows to ethnic community organizations depend on the accurate classi- fication and counting of their members and as long as the Voting Rights Act demands single-race reporting. We recognize that these are extremely difficult methodological problems to solve; we suggest, among other things, that the Census Bureau consider experimenting with classification schemes that enable respondents to check off more than one race category. We further suggest that the Census Bureau sponsor additional research on the question of ethnic identity, utilizing the intellectual resources of cognitive anthropology (in which much relevant research has already been done) and sociology. We note in passing the importance of the cognitive research that has accom- panied the Living Situation Survey; this research is being conducted primarily to develop new ways of defining the residence rules (Gerber and Bates, 1994; 6 The Census Bureau has conducted or sponsored studies of the effects of the order of race and Hispanic origin questions on item nonresponse and race reporting (see Bates et al., 1994).
86 COUNTING PEOPLE IN THE INFORMATION AGE Sweet, 1994~. A similar program of research should be undertaken on the subject of racial and ethnic identity. Recommendation 3.8: The Census Bureau should undertake a program of research in cognitive anthropology, sociology, and psychology that will contribute to the development of more acceptable racial and ethnic identification questions. The policy recommendations that emerged from the ethnographic studies are potentially of great value for the 2000 census. We believe that the Census Bureau has made major progress in developing strategies of outreach and enumeration that are based on the ethnographic evidence particularly strategies that involve localized, customized methods of recruiting enumerators drawn from minority populations, providing ongoing outreach support to local minority organizations, and using the greater trust and recognition of these organizations by hard-to- enumerate populations as a basis for a partnership in census-taking. Work must now continue to transform these policy recommendations into operational com- ponents that can be tested and evaluated in the 1995 census test. TOOL KIT AND PLANNING DATABASE The key census design components being developed to improve coverage and consequently reduce the differential undercount of hard-to-enumerate popu- lations are the tool kit and the planning database. The tool kit comprises the set of special enumeration methods and such strategies as questionnaire assistance and targeted outreach and promotion efforts. Some tool-kit methods for ex- ample, team enumeration, blitz tactics, and bilingual enumerators would be available for deployment by regional offices in hard-to-enumerate areas. Use of the tool kit could also involve specialized outreach procedures and decisions at the headquarters level on mailout-mailback procedures, allocation of staff and resources, and differential pay rates and incentives for census workers. The planning database (formerly called the targeting database in Census Bureau documents) would be used either informally or with a predictive model (the targeting model)-in the 2000 census to preidentify geographic areas in which enumeration barriers are likely to be present and deployment of the special enumeration methods in the tool kit might particularly improve coverage. The planning database could be used, for example, to identify linguistically isolated areas to which Spanish forms should be mailed, to target recruiting messages, and to tailor promotion and marketing materials. Development of the tool kit and the planning database draws on the findings of interdivisional working groups at the Census Bureau that were charged with identifying barriers to enumeration and special methods that might overcome these barriers. The final report of the Tool Kit Working Group (Bureau of the Census, 1994b) recommended tools for further consideration for research and
RESPONSE AND COVERAGE 87 possible use in the 2000 census. Seven candidate tools for use by district or regional offices have been identified for testing and evaluation in the 1995 census test (Bureau of the Census, 1994d): 1. Blitz enumeration. The district office would employ crews of specially trained enumerators who would conduct enumeration activities in a very com- pressed time schedule. 2. Use of community-based organizations. The regional office would iden- tify local organizations that are willing to conduct questionnaire assistance or outreach activities. 3. Use of rental companies. In areas with high concentrations of rental units, regional offices would ask moving companies, resident managers, managers of temporary storage space, truck rental companies, and others who provide ser- vices to renters and movers, to participate in census promotion. Participants could display census posters and publicity materials, as well as distribute census promotional items, such as key chains and calendars, to their customers. 4. Use of local facilitators. The district office would have the option of using either paid or volunteer local facilitators, such as community activists, religious leaders, gang leaders, and other recognized local figures, to facilitate the enumeration for example, by introducing enumerators to households, translat- ing when necessary, convincing people to cooperate, and locating hidden living quarters. 5. Promotion focus on confidentiality. The regional or district office would identify geographic areas in which greater attention should be given to confiden- tiality concerns. In such areas, confidentiality would become the key theme in outreach and promotion efforts. The regional or district office would develop messages that speak to the specific fears and concerns of the local population. 6. Assistance centers in large multiunit buildings. District offices would open and staff booths in large multiunit buildings to enumerate residents and assist residents in completing their census questionnaires. 7. Team enumeration. More than one enumerator would work in an area because of concerns about safety or because additional enumerators could assist in locating the units or in persuading respondents to cooperate. An eighth tool, multilingual telephone assistance, was identified but will not be used in 1995 because of insufficient need at the test sites. This method will be developed for use in the 2000 census. The 1995 census test will also include four tools requiring coordination by census headquarters: 1. Urban updatelenumerate. Headquarters would work with regional of- fices to identify selected areas for urban update/enumerate methodologies instead of attempting to conduct regular mailout-mailback operations. Enumerators would canvass selected areas to update the address list and to enumerate people they find while canvassing.
88 COUNTING PEOPLE IN THE INFORMATION AGE 2. Questionnaires in languages other than English and Spanish. Headquar- ters would develop questionnaires as well as guides in languages other than English and Spanish. Regional offices would be responsible for using these materials with appropriate populations during nonresponse follow-up and update/ enumerate. 3. Unaddressed questionnaires. Census questionnaires in various languages would be available in places where the historically undercounted tend to congre- gate. 4. Mailout of Spanish-language questionnaires. Headquarters would mail questionnaires in Spanish to areas with high concentrations of linguistically iso- lated Spanish-speaking households. We strongly encourage the Census Bureau to examine these tool-kit methods in the 1995 census test. Some of the above tools have been used in past censuses (e.g., blitz and team enumeration, use of community-based organizations), but their cost-effectiveness is not well documented. The Census Bureau has little or no research experience with most of the proposed tools, and testing in 1995 should therefore provide important imformation. We note, however, that the application of the tool kit needs further specifica- tion. It should not be assumed, for example, that tools could only be used in efforts to improve the initial counts in particular geographic areas. Some tools, especially if they are expensive to implement, might be used most effectively during nonresponse follow-up or integrated coverage measurement. The applica- tion of tool-kit methods in different stages of census operations would require coordination. Similarly, some tools may be more appropriate for identifying and including housing units; others may be suited for motivating and reaching indi- viduals. The planning database has an important role to play in guiding application of the tool kit to ensure that tools are used systematically and only when needed and that their use is recorded so that, when relevant, this information can be taken into account in integrated coverage measurement. Current plans for the 1995 census test call for a tool to be assigned to a given area if that area exceeds predefined threshold values (e.g., mail nonresponse rates) that are set using judgments from experienced field personnel. The Tool Kit Working Group's report (Bureau of the Census, 1994b) at- tempts to identify the role that the tools would play as components of a full census design and to suggest the need for their evaluation. But these suggestions are not yet fully developed, and there does not appear to be a formal evaluation plan for all of these methods or for the smaller set included in the 1995 census test. The need for formal evaluation of the components of the tool kit is similar to that for evaluating the effects of different instruments or modes. When multiple data collection strategies are used, particularly in conjunction with nonresponse
RESPONSE AND COVERAGE 89 follow-up and integrated coverage measurement, it is important to be able to estimate the impact of the different strategies. The first step is more complete definition of the tools themselves. Subse- quent evaluation of tool-kit methods should include information on their effec- tiveness (yield, impact on the differential undercount) and cost. Information on cost will be needed to choose among competing strategies. Although full experi- mental evaluation of components of the tool kit used in the 1995 census test is probably not possible, it should still be possible to plan comparisons and varia- tion across sites or areas within sites. Posttest evaluation surveys that ask people whether they were aware of outreach and promotion efforts can help in evaluat- ing the effectiveness of such efforts. Information on effectiveness and cost will be needed to inform decisions after 1995. At present, the planning database includes block-level tabulations of 1990 census content, mail return rates, and information from administrative records. We understand that the Census Bureau intends to add recently acquired crime statistics (compiled at the census tract level) and more current administrative records data to the targeting database for the 1995 census test. The effectiveness of the planning database will depend on its being maintained with up-to-date information. Commercial databases and local administrative record systems may be good sources of block-level supplemental information. Both the tool kit and the planning database represent substantial develop- ment efforts, and developing a formal predictive model for use with the database requires even larger investments. It seems very likely that the development and testing of the tool kit, planning database, and targeting model will continue after the 1995 census test, and planning should take this schedule into account. Ex- perimental evaluation of some components of the tool kit (e.g., blitz enumera- tion) could be incorporated into census tests after 1995 or carried out in conjunc- tion with special-purpose tests, special censuses, or current Census Bureau surveys. OUTREACH AND PROMOTION The outreach and promotion program for the 1990 census was the most intensive to date for a decennial census. For the first time, a multifaceted mass media campaign addressed several traditionally undercounted groups as well as the general public. The community-based Census Awareness and Products Pro- gram was enhanced and began operations farther in advance of Census Day than in 1980. Programs were conducted that worked through national civic and reli- gious organizations, schools, Head Start agencies, governmental units, and busi- ness organizations. Based on an analysis of the data collected in the 1990 Outreach and Evalua- tion Survey, Bates and Whitford (1991) concluded that the Census Bureau's 1990 outreach and promotion program achieved many of its goals. The Advertising
90 COUNTING PEOPLE IN THE INFORMATION AGE Council's mass media campaign received wide exposure, and it achieved signifi- cant media presence around Census Day in the six media markets in which the coverage was monitored. By Census Day, over 90 percent of the population had recently heard or read something about the census, although the campaign was less effective in reaching blacks than whites and Hispanics. Ironically, the differ- ential success of national outreach in raising levels of census awareness may have had the unintended effect of exacerbating problems of differential coverage. This experience underscores the importance of outreach to local communities with hard-to-enumerate populations. Outreach and publicity may also help the census mail response rate; data from the Outreach and Evaluation Survey indicate that in 1990 the mail return rates of respondents with high awareness of census operations and knowledge of census uses were 15 to 20 percentage points higher than respondents with low awareness and knowledge (Bates and Whitford, 1991~. (We cite these results with the standard caution that the question of whether these associations reflect causal relationships cannot be answered without considering potential interaction effects and controlling for other variables in a designed experiment.) Neverthe- less, the mail response rate in 1990 was 10 percentage points below that in the 1980 census, and the differential undercount between blacks and others was the highest since the Census Bureau began estimating coverage in 1940 (U.S. Gen- eral Accounting Office, 1992~. Although there are undoubtedly many social factors that contributed to the response rate decline, it is also possible that the Census Bureau's 1990 outreach and promotion campaign did a better job of announcing the census than it did of persuading people to participate. Regard- less, it seems clear that an even more intensive and effective-outreach and promotion program will need to be designed, tested, and successfully imple- mented in the 2000 census if the Census Bureau is to prevent further erosion of census participation rates that began declining in 1970. Responsibility for Decennial Census Outreach and Promotion The Census Bureau does not have a single, permanent office that is respon- sible for decennial census outreach and promotion. Instead, in the past two censuses, it has opened a temporary Census Promotion Office two years prior to the census and then closed the office after census operations were completed. It appears that the primary function of this office is to coordinate the media cam- paign with the Advertising Council and to oversee various national outreach programs conducted to promote the upcoming census. (The Census Bureau has a permanent Public Information Office, but it is not directly involved in decennial census promotion.) In every census since 1950, the Census Bureau has relied on the Advertising Council to design and conduct the media campaign. (Working through advertis- ing agencies that volunteer their time, the Advertising Council regularly conducts
RESPONSE AND COVERAGE 91 mass media advertising campaigns on a pro bono basis for government and nonprofit agencies.) In 1990 the Advertising Council for the first time designed multiple campaigns to target selected minorities (blacks, Hispanics, and Asian and Pacific Islanders) as well as the general public. The principal national outreach program to undercounted groups is the Na- tional Services Program. Conducted by the Data User Services Division, the program is a continuing outreach and data dissemination program aimed at na- tional organizations that represent undercounted minorities. Its goal is to secure the support of the national organizations and their local and regional chapters in efforts to encourage participation by the minority communities they represent. The Census Bureau also conducts a variety of national outreach activities close to Census Day; these include programs that target a variety of organizations, schools, governmental units and agencies, and private-sector corporations. The principal program for local outreach in 1990 was the Census Awareness and Products Program. Administered by the Census Bureau's Field Division, this program was typically activated at least one year prior to Census Day in local areas and deactivated shortly thereafter. In 1990 there were about 280 staff across the country, working out of the regional field offices. The Census Bureau's census outreach and promotion program does not in- clude a structured plan for conducting research and development work. The absence of a permanent office responsible for census outreach and promotion militates against sustaining an ongoing research program during the decade be- tween censuses. Also, the Advertising Council does not permit its clients to undertake or commission media research on their own. The Census Bureau does, however, undertake retrospective research to evaluate census outreach and pro- motion. Examples from the 1990 census include the Outreach Evaluation Sur- vey, the Telephone Survey of Census Participation, the Survey of 1990 Census Participation, and the National Service Program Structured Debriefings. The Census Bureau's Public Information Office has recently funded some focus groups with young black, Hispanic, and Asian males with the hope of identifying ways to reach these groups and motivate them to participate in the census. The Year 2000 Research and Development staff is also involved in outreach and is considering a joint venture with state and local governments to improve outreach at the local level. We are concerned that the responsibility for outreach and promotion is split among several different units within the Census Bureau. We believe that the effectiveness and efficiency of the Census Bureau's census outreach and promo- tion program could be improved if a permanent office were established and staffed with advertising and public relations professionals. This office would be responsible for planning, researching, and developing all outreach and promotion activities and for overseeing the implementation of the decennial census pro- gram. It should be a permanent and nonpartisan office to provide continuity between censuses and to monitor an ongoing national and local outreach and
92 COUNTING PEOPLE IN THE INFORMATION AGE promotion program during the decade. Because the Census Bureau already has a permanent Public Information Office, we suggest that the Census Bureau con- sider expanding the mission of this office to include responsibility for the decen- nial census outreach and promotion program. Although the establishment of a permanent decennial census outreach and promotion office would centralize the overall responsibility for outreach and promotion activities, it does not follow that such activities would become more focused on the national rather than the local level. Nor should they, as we argue below; a principal objective of the new central office should be to enhance outreach efforts at the local level. Just as the centralized Field Division is respon- sible for all data collection operations conducted by the regional offices, so a centralized census outreach and promotion office would design, monitor, and support all field outreach activities conducted through those regional offices. The one outreach and promotion activity that should not be consolidated under the purview of the new office is evaluation research. To preserve indepen- dence, this type of research should continue to be undertaken by other units within the Census Bureau, such as the Center for Survey Methods Research. Recommendation 3.9: The Census Bureau should assign overall respon sibility for decennial census outreach and promotion to a centralized, permanent office. The Census Bureau should consider expanding the mission of the extant Public Information Office to include this charge. Evaluation of outreach and promotion programs should be conducted by an independent unit within the Census Bureau. National Media Campaign An important issue confronting the Census Bureau is whether to continue to rely on the Advertising Council to design and implement the national media campaign. As noted above, the Advertising Council's rules prohibit clients from commissioning any media research on their own. It also does not permit clients to supplement the pro bono campaign with any paid advertising. Thus, the Census Bureau is entirely dependent on the Advertising Council's pro bono campaign for its media research and advertising. Because the Advertising Council depends on volunteer labor, we are con cerned that its pro bono campaign may not include the same level of media research that large commercial advertisers have found beneficial. For example, the leading advertising agencies evaluate commercials by using sophisticated technology that continuously measures a focus group's response to a proposed commercial message. However, this type of research is expensive and may not be included within the scope of a pro bono campaign. We also understand that there is a growing feeling among certain charitable organizations that national public service announcement campaigns are losing their effectiveness. One theory about this change is that U.S. society has become
RESPONSE AND COVERAGE 93 so diverse and complex that people are increasingly narrowing their focus to the community in which they live. As a consequence, there is a movement among pro bono advertisers away from the national campaigns toward increased use of local campaigns. Certainly, the Census Bureau must continue to rely on public service an- nouncements run on a pro bono basis; it would be prohibitively expensive to launch an equivalent campaign on a strictly fee-for-service basis. However, it could discontinue use of the Advertising Council and instead work directly with local and regional agencies. This approach would allow the Census Bureau to undertake a paid media research program to identify the most effective advertis- ing messages. It would also mean that the national campaign would include a collection of local and regional campaigns that might be more effective in reach- ing the communities to which they are directed. Finally, this approach would allow the Census Bureau to supplement pro bono advertising with paid advertising especially in hard-to-enumerate com- munities where a pro bono campaign may not provide sufficient or appropriate media exposure (Committee on National Statistics, 1978~. We believe that the additional cost involved may be more than offset by the increase in response among the traditionally undercounted groups, with a corresponding reduction in the differential undercount. We note in this regard that Statistics Canada is a strong proponent of paid advertising; it was discontinued in Canada in 1986 as a cost-saving measure, but then promptly reinstated after a decline in participation. Recommendation 3.10: The Census Bureau should evaluate the costs and benefits of alternatives to the use of the Advertising Council to conduct the 2000 census media campaign. Some alternative options are working directly with local and regional agencies, undertaking paid media research, and supplementing pro bono advertising with paid ad- vertising in hard-to-enumerate localities. Cooperative Ventures With State and Local Governments The Census Bureau has undertaken a number of initiatives designed to im- prove the level of cooperation between its field operations and state and local governments, critical nodes in the Census Bureau's efforts to go local. These efforts span a range of tasks, from outreach and promotion of the decennial census to the use of state or locally maintained administrative records for the purpose of improving coverage (Collins, 1994~. Some of these efforts will be put in place during the 1995 census test in order to assess their effectiveness, imple- mented by "census advisors" who will be placed in local offices in the test communities in order to oversee liaison programs of all kinds. We understand that, in the 1995 census test sites of Oakland, California, and the six parishes in northwestern Louisiana, the census advisors will be district office employees of
94 COUNTING PEOPLE IN THE INFORMATION AGE the Census Bureau. Discussions are ongoing regarding whether the census advi- sors in the remaining two test sites Patterson, New Jersey, and New Haven, Connecticut will be district office staff or local people appointed by the mayor or city council. With respect to outreach and promotion activities, current plans build on the 1990 experience, during which "complete count committees" formed in various localities to boost the coverage of the decennial census. These committees were responsible for tailoring local press coverage of the census and for maintaining contact with grass-roots organizations and advocacy groups with particular knowledge of hard-to-enumerate populations. For the 1995 census test, these activities will be expanded, formalized, and coordinated by the census advisors. This effort is of great importance, for the more effective the Census Bureau's ties to local organizations involved in service delivery to poor people, migrants, immigrants, and other groups traditionally undercounted, the better the coverage will be. Future census tests may provide an opportunity to assess how much complete count committees and other outreach activities cost and how much they contribute to better coverage. Census advisors will also be responsible for coordinating local information and personnel resources that can assist in augmenting the Master Address File. Local officials will be sworn in so that they can be given access to Census Bureau address lists and held accountable for maintaining confidentiality.7 Thereafter, the Census Bureau plans to (a) share census address lists with state and local governments officials, particularly those in city planning departments; (b) solicit block boundary suggestions from local officials familiar with the contours of neighborhoods; (c) hire city employees to work on address list verification; (d) work with knowledgeable local officials on administrative records that are lo- cally generated and maintained that can be used to improve coverage; and (e) develop an effective program for office space procurement, a perennial problem for the census staff. These strategies will, we believe, capitalize on the knowl- edge base that resides at the local level, improving outreach, promotion, and the quality of address lists and encourage much-needed federal-state-local coopera- tion. The Census Bureau plans to enlist the support of local officials to help in the planning of the 1995 census test, particularly where blitz enumeration is con- cerned. City employees who know the hard-to-count populations in their locali- ties will be able to assist enumeration teams, making it more likely that poor people, immigrants, and internal migrants will be counted the first time. These individuals will be working under the direction of the census advisors during the test and will, in this capacity, become part of the planning apparatus for using the 7 For the 2000 census, legislative changes will be required in order to make it possible to expand this cooperative program nationwide.
RESPONSE AND COVERAGE 95 tool-kit strategies for had-to-enumerate groups. Clear guidelines and role defini- tions may be needed to ensure that the involvement of local officials with en- forcement responsibilities is accomplished in a manner that preserves the integ- r~ty of the census operation. In an effort to evaluate the Census Bureau's program of cooperative ventures with state and local governments, a series of debriefings with participants are being planned for the 1995 census test. The debriefings will serve to obtain information from program participants, state and local government officials, and the Census Bureau staff from the regional offices on their reactions to the pro- gram and what effects they experienced on timing, operations, and logistics. The Census Bureau will also meet with representatives from localities that elected not to participate in the program to discuss their reasons for not participating as well as review the reasons why some decided to participate only in certain aspects of the program. Cooperative ventures of this kind are responsive to our general call for a strategy of going local, in the sense that they make use of existing ties and local expertise to encourage cooperation with the census. We believe they should be seen as complementary to programs already in place (particularly the Census Awareness and Products Program), rather than substitutions for them. Recommendation 3.11: The Census Bureau should evaluate the pro- grams for state and local cooperation that will be overseen by census advisors in the 1995 census test areas in order to collect from these experimental initiatives those programs most likely to (a) reduce the cost of the decennial census (particularly by improving mail response rates) and (b) reduce the differential undercount. Preservation of the Census Awareness and Products Program should, however, be a high priority, not to be superseded by this new initiative for improving state and local cooperation.