The panel reviewed uses and users of decennial census data with several objects in mind. The first purpose of the review was to document major uses of the census and identify their data requirements to permit the panel to evaluate the likely impact of changes in methodology. The second purpose was to assess—or at least to inquire into—whether some uses could not be satisfied as well or almost as well by other data collection programs. The third purpose was to examine the sensitivity of each major type of use to the accuracy of the data.
An inescapable conclusion from our review is that, given the multiplicity of important purposes served by the census, major changes in census methodology should not be made without careful consideration of their ramifications for a broad spectrum of uses. At the same time, we believe that investigation of alternative approaches to data collection might reveal opportunities to remove some questions from the census (particularly from the long form) or to make other changes that would free funds for efforts to improve the data that are collected. Examination of the sensitivity of uses to the accuracy of the census is needed to understand the consequences of census errors and to determine the benefit of devoting additional resources to improving the data. How much difference would it make in the distribution of revenue sharing dollars, for example, if the differential net undercount among ethnic groups in the population could be reduced from about 4 percentage points (the apparent difference between blacks and all others in 1980) to 2 percentage points? Would this improvement make more or less difference than improved measurement of per capita income, which also enters into the revenue sharing formula?
The panel’s review of uses of the census stems from the belief that decisions on methodology for a data collection program should consider the various purposes the program is intended to serve. If a statistical program were being designed de novo, the responsible agency might go through the following steps:
- Identify the fundamental purposes the program must serve and the minimal requirements for subject matter and geographic area detail, needed accuracy of the data, and frequency of data collection required to satisfy these purposes;
- Identify secondary purposes that it would be desirable to accommodate and their data requirements;
- Identify methodologies that, at a minimum, can serve the basic purposes;
- Further evaluate those methodologies on other criteria, such as ability to satisfy secondary purposes and public acceptability;
- Determine costs for each methodology of serving the basic purposes and the incremental costs of serving additional purposes; and
- Select the optimal methodology.
In the case of the decennial census, with its long history of serving many uses and users, its unique role in determining political representation, and its operational complexity, methodological choices cannot be nearly as cut- and-dried as the above scheme would suggest. It is not easy to rank uses in order of importance—what may be of marginal direct value to federal officials may be of great value for local planners or business people, and it is not clear how to weight these different assessments. Having made a decision to assign a lower or higher priority to a given use leads to further problems of implementation. On one hand, it is hard to reconcile users to a decision to scale back the level of detail or accuracy provided or to stop serving a need altogether. On the other hand, it is hard to make changes to tried-and-tested procedures to accommodate new uses or to improve the level of detail or accuracy provided, even if cost were no particular object.
The panel did not attempt to resolve these difficult questions, but undertook a more limited review of census uses and users with the objectives set forth at the outset. The chapter begins with a brief overview of the uses of census data in the American past. Subsequent sections review the distinguishing features of the modern census that shape the uses made of the data, give examples of major types of applications, and endeavor to draw implications for census methodology from the data requirements for important uses. The chapter then reviews the limited body of research that has attempted to measure the effects of census data errors on key purposes, such as reapportionment and fund allocation. The concluding section re-
views research on the magnitude of errors introduced in postcensal population estimates compared with errors in the census itself and discusses the implications of the research results for the utility of a mid-decade census.
Originally, the main purpose of the decennial census in the United States was to determine the population count in every state for apportioning seats in the House of Representatives. Very soon, however, the census was expanded to collect additional information beyond basic demographics, and policy makers and analysts began to use the data for many purposes.
In the nineteenth and early twentieth centuries, census data are known to have served at least the following types of uses:
- Scholarly analysis. For example, Frederick Jackson Turner’s landmark work on “The Significance of the Frontier in American History” (1894) rested on analysis of census data.
- Input to public policy decisions. Census results strongly influenced the debate at the turn of the century that culminated in the National Origins Act of 1924, which severely restricted immigration (Conk, 1984:10-13). A noted Civil War historian has suggested that the 1860 census results were a factor leading the South to secede rather than accept growing Northern population—and therefore political—dominance (Nichols, 1948:460-461).
- Use for allocation of federal funds to the states. Between 1887 and 1921, the Congress passed laws providing for allocation of funds to the states for programs of vocational education, agricultural extension, conservation, highways, and public health using formulas that included census population counts. These laws laid the foundation for the grant-in-aid system. By 1930, the total funds distributed by formula amounted to about $100 million, 3 percent of the federal budget (Conk, 1984:18-19).
- Public information and population analysis. From the beginning, Americans have been keenly interested in what the census results show about their own place of residence and how it stacks up against others. Census results have found their way into countless speeches, student themes, and newspaper and magazine articles describing and extolling local areas and reporting changes over time.
All the historical uses of census data described above have their counterparts today. Users of early censuses would be astounded by the extent and depth of analysis made possible by modern computer technology, but they would readily recognize many types of applications.
The modern census in the United States has evolved in response to demands for data to serve a wide range of purposes, many of which are not served by any other data collection program. The need to satisfy particular kinds of important purposes has shaped census methodology, and, conversely, the distinguishing characteristics of the census program have created a set of expectations among users regarding data they look for in the census.
This section organizes a discussion of uses of the present-day census according to three main features that together differentiate the census from other data collection programs: population counts for small areas, small-area and subgroup characteristics, and historical time series. Questions posed are: What is the census currently expected to provide that other data collection vehicles do not? What kinds of benefits do users anticipate from census data as opposed to data from other sources? What are the implications of user expectations for proposed changes in methodology? Appendix 2.1 describes applications that state and local governments—two important user groups—make of census data. Appendix 2.2 depicts the range of uses within a single geographic area—New Jersey—among private, public, and academic users. It also describes the various distribution channels through which census data are made available to users.1
Basic Counts for Small Areas
The census is the source of complete head counts, including basic information about age, race, and sex, and of residential housing counts obtained in a consistent manner throughout the country for small as well as large geographic areas. The census provides counts not only for the nation as a whole and for large areas such as regions, states, and metropolitan areas, but also for counties, congressional districts, cities and towns, and minor civil divisions of counties. In addition, in what represents a vitally important and relatively recent development, the census provides counts for local areas including census tracts, block groups, and city blocks.
Local areas identified in the census are typically quite small in population (as are some political jurisdictions, such as towns and villages—see Bureau of the Census, 1982b:Ch.4). Census tracts—first delineated in several large cities for the 1910 census—generally have between 2,500 and 8,000 residents and are currently identified in every metropolitan area
1 See U.S. House of Representatives (1982) for additional documentation provided by many users from government, private business, and academic institutions of their needs for census data.
and some nonmetropolitan counties. Block groups along with enumeration districts covered the entire nation in 1980 (the former were tabulated where there were city blocks and the latter elsewhere) and averaged about 800 population. There were over 2.5 million city blocks in 1980 identified in urbanized areas, cities of 10,000 or more population outside urbanized areas, and in other areas that contracted with the Census Bureau to tabulate block statistics. By 1990, blocks will be identified in all areas of the nation. All of these types of small areas are often used as “building blocks” in putting together information for nonstandard census areas, such as school districts, neighborhoods, police precincts, urban renewal areas, etc.
In contrast, the largest federal sample survey ever conducted, the 1976 Survey of Income and Education, covered enough households (200,000) to provide reliable data for states and metropolitan areas but not for any smaller areas. Regularly recurring federal surveys, such as the American Housing Survey and the Health Interview Survey, contain just enough households (currently about 40,000) to produce reliable information for large states and metropolitan areas. The Current Population Survey (CPS), which includes about 60,000 households, is now designed to produce estimates for all states and also large metropolitan areas but cannot support estimates for smaller areas. Some localities conduct their own censuses (usually contracting with the Census Bureau) or surveys, but these efforts do not generate comparative data for other areas.
Sample surveys, even the most thoroughly conducted ones, also do not obtain as complete a coverage of the population as the decennial census. The Census Bureau estimates that the Current Population Survey (after imputation for refusals and other cases of nonresponse, but before ratio estimation using census-based current population estimates) covers only 93 percent of the census total (Hansen, 1984:138).
Various administrative records systems can potentially provide complete counts for small as well as large areas, but no currently existing system covers the entire population in a consistent manner. Among large federal systems, Internal Revenue Service (IRS) records, while covering most persons, exclude those who do not file tax returns or who are not listed as dependents and, in addition, overcount persons who both file a return and are reported as a dependent on someone else’s return. Social Security Administration records likewise both undercount, excluding children who have not yet applied for a card and adults who have never worked or applied for a card and are not yet eligible for Medicare, and overcount, including some decedents and persons with more than one social security number. Moreover, the address information needed to determine individuals’ specific place of residence is not fully available from these sources—many IRS addresses represent place of business or legal domicile rather than place of residence, and social security addresses typically are current only for those receiving
benefits (see Alvey and Scheuren, 1982). Other limitations of administrative records include the difficulty of generating data on families and households and the paucity of characteristics information.
Among the major uses of basic counts from the census are the following:
- Reapportionment of the U.S. House of Representatives according to the distribution of population among the states. Title 13 of the U.S. Code includes a provision requiring the Secretary of Commerce to report state population totals to the President within 9 months after Census Day, i.e., by December 31 of the census year.
- Redistricting within states and localities to meet stringent court-mandated criteria for equal size and compactness of election districts and for appropriate representation of race and ethnic groups. Under current law, the Census Bureau is to provide to the states within 1 year after Census Day a computer tape containing small-area population counts. The tapes provided April 1, 1981, contained total population plus race and Hispanic origin for blocks, enumeration districts, and, where specified by the state, precincts.
- Benchmarking of postcensal population estimates. Census counts by age, race, and sex are the starting point for current population estimates produced between census years for geographic areas ranging from the nation as a whole to states, counties, and all 39,000 political jurisdictions recognized for federal revenue sharing.
- Calibration of data from other collection programs. Census-based current population estimates by age, sex, race, and Hispanic origin are the basis for weighting the output from federal surveys such as the Current Population Survey and the Health Interview Survey.
- Calculation of vital rates. Census counts and census-based population estimates by age, race, sex, and geographic area serve as denominators for rates of births, deaths, marriages, and divorces produced for the nation and the states from the vital statistics program.
- Allocation of federal and state dollars to states and localities. A large number of grant-in-aid programs include the total population as one element in the allocation formula. The best known of these programs is general revenue sharing.
- Determination of eligibility for funding from government programs and of local rights and responsibilities. A number of grant programs have thresholds for eligibility; for example, the Job Training Partnership Act generally designates service delivery areas as counties or cities with 200,000 or more population. Most states classify counties and municipalities by size and accord various rights and responsibilities to each size class.
- Public planning and decision making. For example, cities examine census counts by police precinct, school district, fire precinct, and many other kinds of administrative areas built up from census geography such as city blocks to allocate personnel and budget in proportion to population and housing, to redraw administrative areas to equalize demands for basic services, and to serve as a starting point for projecting future public needs.
- Business planning and decision making. Retailers locating sales outlets, for example, compare population density and demographic characteristics in areas surrounding possible sites.
- Comparison and ranking of areas, such as cities and metropolitan centers, by population, for many purposes such as advertising, marketing, and public information. Even the most casual review of the nation’s media quickly reveals the extent of reliance on census and census-based statistics for articles, maps, and graphs on national, regional, and local demographic characteristics (see Rowe in U.S. House of Representatives, 1982:424-428, for data on use of federal statistics in The New York Times).
As noted at the outset of this chapter, the panel believes that the uses of census data should be examined periodically, by the Census Bureau and others, to reassess their importance and the possibilities for meeting them from alternative sources. Some uses of basic small-area counts nay appear unimportant or even frivolous and not worth expenditure of public funds. However, other uses are fundamental to our federal governance (including reapportionment and redistricting) or to the efficient delivery of goods and services in the public and private sectors, and demonstrate why basic small-area counts constitute the heart of the census program.
Most of the important uses of basic census population (and housing) counts cited impose the requirements that data be collected in a complete and comparable way for all manner and size of geographic areas, with consequent implications for proposed changes in methodology. The requirement for comparable data across areas strongly implies the need to obtain estimates of the population more or less at a point in time. The requirement for comparable data argues as well for the need to standardize processes of data treatment and estimation to the extent practicable. The fact that users expect to be able to obtain counts for very small areas, such as blocks and tracts, and to use these counts to reaggregate the data into other kinds of areas such as school districts or police precincts, implies the need to incorporate any estimation or imputations used into the microdata records so that consistent totals can be produced for whatever tabulations are requested.
It is of course possible to challenge these arguments or to state that other considerations must take precedence. However, the panel in subse-
quent chapters evaluating promised changes in methodology justifies the premise that comparability and consistency of census figures are requirements that methodological innovations should satisfy unless there are compelling reasons not to do so.
With regard to the requirement for completeness or accuracy of basic census counts, the picture is somewhat less clear. Ideally, every user would like a completely accurate set of numbers, but it is recognized that it is impossible to obtain a perfectly complete count. The question becomes the tolerable level of accuracy for a particular application. For many uses of the basic counts, such as allocating police personnel in proportion to neighborhood population, the level of accuracy currently embedded in the numbers is probably quite acceptable. With regard to reapportionment, there is evidence, discussed below, that the differential errors in the 1980 census counts may have affected the allocation of one or two seats, a matter of some concern to the states involved.
Evaluation of the need for increased accuracy of census counts for uses in fund allocation formulas is difficult. Most formulas include other factors besides the population counts. The available limited evidence on the effect of errors in the basic counts on equity in fund allocation is reviewed in a separate section below. For many programs that allocate predetermined amounts of public monies, and for other key uses such as determination of political representation, it is the differential errors among population subgroups and geographic areas that cause the most serious concern. Differential errors in coverage of basic age, race, and sex groups also have implications, discussed below, for postcensal population estimates and for important series, such as vital statistics that use census figures in the denominator. With regard to uses of the counts as thresholds, inaccuracy assumes importance in cases in which coverage error places an area on the wrong side of a threshold. Legislation and administrative practices often provide avenues of appeal for areas that believe they have grown enough to cross a threshold or not declined enough to drop below a threshold even though the census numbers say otherwise (see appendixes). In sum, although the picture is mixed, there is ample evidence that errors in the counts matter for important purposes and that methodologies showing promise to reduce errors (see discussion in Chapter 7) should be given serious consideration.
Small-Area and Subgroup Characteristics
In addition to head counts and basic characteristics such as age, race, and sex, the census obtains detailed data for many other characteristics on a comparable basis for small as well as large areas and for subgroups of the population. The 1980 census included in total over 30 population and 30 housing items covering a broad range of topics; most items—26 of
the population and 20 of the housing questions—were asked of samples of households rather than of everyone. Census products, which include computer files as well as printed documents, cross-tabulate these items in a variety of ways. In order to protect the confidentiality of individual responses, more detailed tabulations are provided for larger than for smaller geographic areas. Summary tape file 1 from the 1980 census contains over 300 items of data (such as the count of married women age 15 and older) for individual blocks and enumeration districts and summary tape file 5 has over 1 million data items (such as the number of Hispanic women in a certain age and income category) for all metropolitan areas and large cities and counties.
These data make it possible to carry out a wide variety of comparative studies of geographic areas and population components. To list just some examples, census characteristics data are used on a cross-section basis for:
- Government planning, analysis, and decision making at all levels, including:
— Assigning local agency personnel to currently defined police or fire precincts and redefining precincts using census geographic building blocks and census socioeconomic data on the population and housing of each area;
— Identifying the most “disadvantaged” areas in a city for locating service facilities;
— Conducting traffic planning studies related to peak loadings and based on cross-tabulated information on place of work and place of residence;
— Identifying concentrations of groups that are targets or potential targets of government programs (poor elderly persons living alone, youth without previous work experience, work-disabled persons);
— Allocating funds to states and localities by means of formulas (e.g., age of housing is a factor in one formula used for community development block grants and children in poverty is a factor in the formula for some educational assistance programs); and
— Redesigning major statistical programs, such as the Current Population Survey.
- Business planning and decision making, including:
— Locating retail outlets in terms of market potential based on area socioeconomic characteristics, such as income, occupation, education, home ownership, and housing value;
— Comparing the market potential of different cities, ZIP codes, or census tracts within cities; and
— Assessing the availability of needed occupational skills in different labor market areas.
- Basic and applied socioeconomic research, including:
— Analyzing groups that represent reaggregations, for example, persons in high-tech industries aggregated from detailed industry breakdowns;
— Issue-oriented analyses, for example, study of the assimilation of different immigrant groups or projections of shortages in selected occupations;
— Analysis and legal testimony related to affirmative action and equal employment opportunity programs and challenges to the representativeness of juries; and
— Analysis of relationships, for example, characteristics of persons who moved during the past 5 years, characteristics of families with adult children living at home.
Although questions have been raised about the necessity of having the census collect all the characteristics data, the census does not by any means collect every kind of item that business leaders, government officials, and researchers might want. This is true even though the marginal cost of additional questions is low relative to the large fixed cost of obtaining the count and basic demographic information. Moreover, the items that are collected are not all obtained from every household or tabulated for every area. Over the decades, budget and operational constraints, demands for privacy, and considerations of the burden on the public have led the Census Bureau to a methodology that imposes the following kinds of restrictions on the data collected and tabulated:
- The Census Bureau carefully reviews proposed items to be sure that the need for them justifies expenditure of public tax dollars. While the data are useful for many marketing and business planning purposes, the Census Bureau will not include questions solely for such purposes. For example, questions on number of pets are proposed and turned down virtually every decade. Similarly, questions that were asked in censuses through 1970 on appliances, such as clothes dryer and TV, were eliminated from the 1980 census. In prior decades, these items were justified for analysis of changes in standard of living in different areas of the country, but they no longer are.
- The Census Bureau also carefully reviews items to determine whether they are needed at the block level and hence must be included on the short form administered to every household; whether tabulations for somewhat larger areas are sufficient so that the item
should be included on one or more versions of the long form sent to only a sample of households; or whether some other vehicle (such as the CPS) could provide adequate geographic detail.
- Question detail is limited to what it is judged self-respondents can handle in a reasonable amount of time and with a minimum of confusion. For example, the income question in the census specifies fewer categories than the corresponding questions in the Current Population Survey and many fewer categories than the corresponding questions in the Survey of Income and Program Participation. (The SIPP is administered in person by interviewers and the CPS in person or over the telephone.)
- What is asked of every household is limited to what will fit on two facing pages; the number of items asked of a sample of households is limited to what will fit on two additional pages per person plus a page and a half of housing items. Forms are designed and most questions formulated for machine tabulation.
- Cross-tabulations are limited for smaller areas in order to protect the confidentiality of replies and prevent identification of individuals.
The decennial census does not cover as many subjects or cover specific subjects in as great depth as many surveys, but it provides many more analytically relevant explanatory variables than most administrative records systems. The detail it provides can be cross-tabulated in a multiplicity of ways without adversely affecting reliability or raising confidentiality considerations. The census is virtually the only source of detailed comparable characteristics as well as totals for small areas and small groups in the population.
There are several important implications for methodology stemming from these distinguishing features of the census. Many of these are the same as for the basic counts: for comparative analysis the need to obtain a reading more or less at a point in time, the need to standardize all processes of data treatment and estimation, and the need for consistency across various tabulations and retrievals—whether planned or ad hoc. Finally, there is limited evidence that the relatively small errors in the census population count, though possibly significant for analysis of certain very specific subgroups, are rarely significant for most cross-sectional uses of characteristics data. For these uses, the improvement or adjustment of counts is less significant than the reduction of errors and biases in content (e.g., misreporting of marital status by single mothers, miscoding of occupation and industry, errors in income reporting, etc.). As an example, evidence, discussed below, indicates that errors in the income component of the revenue sharing formula have more impact on fund distribution than errors in the population component.
Historical Time Series
The census in the United States provides updates at 10-year intervals of head counts and characteristics for population groups and areas, permitting analysis of changes over time. (Since the Census Bureau does not follow individuals from census to census, longitudinal micro-level data do not exist.) Many of the kinds of uses referenced above, ranging from local planning to market research to scholarly analysis, gain added significance when they are carried out from census to census on a comparable basis. For example, there is keen interest not only in the level of wages of women relative to men and of blacks relative to whites, but also in changes in the relative levels over time by occupation and geographic location.
The time-series character of the census has additional implications for methodology above and beyond the cross-sectional considerations already discussed. Changes in methodology need to be assessed not only in terms of the cross-sectional dimension but also from the point of view of their likely impact on consistency over time. Put differently, if a change is likely to disrupt comparability with the previous census, the gains in the cross-sectional dimension need to be substantial.
Considerations of comparability over time are important for characteristics; they are important as well for the basic counts. There are some important uses of population data for which variability in accuracy of census population counts, from census to census or among population subgroups, is disturbing. A possible change in coverage of a few percentage points for the black or Hispanic population can introduce considerable uncertainty into comparisons of growth rates for various segments of the population. Moreover, because censuses are used to calibrate postcensal population estimates, which are used as population controls for national surveys such as the CPS, changes in coverage create discontinuities in time series that are difficult to interpret. Possibly more important are the uses of population figures as denominators of rates for which the numerators are independently obtained, for example, mortality and birth rates. Changes in coverage over time interfere with analyses of trends in these rates, and differential coverage of blacks and whites, for example, affects comparisons of death rates for specific diseases.
One consideration regarding periodicity of the census is whether 10 years is the optimal interval. A decennial census is mandated in the Constitution for reapportionment. The Congress passed legislation requiring a mid-decade census in 1985 and every year ending in 5 thereafter; however, funds were never appropriated to carry out the 1985 mid-decade census. It appears to be the case that errors in postcensal estimates dwarf errors in the census itself (see further discussion below) and, therefore, depending on the cost, it might be cost-effective to conduct a mid-decade census to improve the data for purposes such as fund allocation.
This section reviews extant research that attempts to measure the impact of errors in the census on important uses of the data. Typically, this is done by implementing one or more types of “adjustment” of the census figures and comparing results of using the pre- and post-adjusted dataset. Considering the political and economic power that flows from the census through reapportionment, redistricting, and fund allocation, as well as the concern over possible inequities in the distribution of power resulting from coverage and other errors, there has been relatively little research on what difference census errors make for the allocation of votes and funds. Moreover, the research studies reported in the literature and reviewed below are subject to limitations in scope and method, so that their findings must be viewed with caution.
The focus on research directed to reapportionment, redistricting, and fund allocation is not meant to suggest that these are the only important uses of census data or that errors nay not be a problem for other uses. As should be evident from the previous discussion, census data are used for a wide range of research, planning, and public policy purposes. However, virtually no research has been carried out on the effects of census errors for purposes other than allocation. Keyfitz (1980) has expressed the opinion that a considerable margin of error is tolerable for most research and planning purposes. Others have argued that, as an example, the use of census data for establishing and monitoring equal employment opportunity programs places requirements for accurate coding of occupation for age, race, and sex groups in small geographic areas that the census currently does not meet (see Conk, 1981).
Before proceeding to review the published studies on effects of census errors for reapportionment, redistricting, and fund allocation, we should be clear about what is encompassed in the term “census errors” and what is not. There are many kinds of errors in collected data. In the census context, errors include:
- Coverage error (households/persons omitted; households/persons erroneously included);
- Unit nonresponse (households/persons known or believed to exist but lacking forms);
- Item nonresponse (households/persons with one or more items blank); and
- Misresponse (e.g., underreporting or overreporting of income).
In addition, the data become less useful the longer the time interval between collection and release.
Current census methodology includes procedures to attempt to correct for some of the sources of error noted above, specifically, unit nonresponse and item nonresponse. However, these procedures are never completely accurate and may introduce added error. Most research studies have focused on coverage errors in the census; a few have also looked at the interaction of coverage error and misresponse for selected items. Users, such as government and business planners, when asked, have often noted that delays in release of census tabulations much more adversely affect their use of the data than do coverage or content errors.2
What none of the research covers and what the discussion in this chapter does not attempt to address are considerations of “error” in the larger sense. That is, even if a dataset were completely accurate, it could well be the case that the application of the data in a formula resulted in an inequitable allocation because the variables did not in fact relate to the intended purpose (see Keyfitz, 1980).
Finally, the panel recognizes the very difficult problems in attempting to assess the implications for census methodology, particularly with regard to adjustment, of research findings about effects of census errors. Such an assessment rests first on one’s judgment about the quality of the research, specifically as to: (1) the accuracy and completeness of the estimates of errors in the census applied in each study and (2) the completeness and appropriateness of the methodology used for evaluating the effects of applying a certain set of correction factors. (For example, need a study of general revenue sharing replicate all aspects of the complicated formulas to assess adequately the implications of estimated census errors?)
Assuming that the research results appear creditable, one must further make a judgment as to whether the measured effects of census errors are sufficiently important to warrant adjustment, particularly given that any adjustment procedure may itself add error and given the cost associated with developing adjustment procedures and gathering the input data for those procedures. In other words, granting that a dataset can never be completely accurate, one must decide what constitutes sufficient accuracy for particular uses and whether adjustments that can be made represent sufficiently significant improvements. Is it tolerable, for example, to have two congressional seats misapportioned because of coverage errors in the census? Four seats? Six seats? Is it tolerable if research suggests that coverage errors do not affect apportionment, but coverage and content errors result in the states receiving, on average, about 1 percent more or less revenue
2 Based on notes of Constance F. Citro from the session on census undercount, annual meeting of the Association of Public Data Users, October 25-26, 1984, Washington, D.C.
sharing funds than they should? Two percent more or less? Four percent more or less? Is it tolerable if areas with high proportions of blacks who are more poorly counted than whites receive less in federal funds from all the population-based formula programs than they should?
Ultimately, these are political judgments. The panel has concluded from the research discussed below and so stated earlier in this chapter its belief that errors in the census do make a difference for important purposes. Throughout the remainder of the report, the panel supports research and testing of methods that show potential to reduce errors within reasonable cost limits. The panel’s recommendations are directed both to methods for coverage improvement (see Chapter 5) and methods for adjustment of the enumerated counts (see Chapter 7).
Effects of Errors on Reapportionment
Siegel (1975) carried out a study of the implications of coverage errors in the 1970 census for the allocation of congressional seats among the states. The study first developed several sets of estimates of net undercoverage for states, using the technique of synthetic estimation (see Chapter 7) to carry down national estimates of undercount for major population groups. The different estimates included:
- A set that applied what the authors believed to be the best estimates of the national rates of net undercount for age, sex, and race groups (black, white, other) to the counts for each group within each state;
- A set that applied the national rates of net undercount for race groups only; and
- A set that in addition applied the national rate of net undercount for the black population to the population of Spanish heritage within each state.
Each of these sets of estimates produced similar geographic distributions of net undercount rates, with nine states having net undercount rates of 3 percent or higher, compared with the national average of 2.5 percent. Three additional sets were produced as modifications of the second scenario as follows:
- A set that assumed that persons below the poverty line in each race group had twice the net undercount rate of persons not in poverty;
- A set that assumed that the net undercount rate for each race group varied inversely with the level of median family income for the state; and
- Finally, a set that assumed that the net undercount rate for each race group varied inversely with the median years of school completed for that group within each state.
Scenarios (5) and (6) resulted in 11 and 10 states, respectively, with net undercount rates of 3 percent or greater, while only 8 states had net undercount rates of 3 percent or greater under scenario (4).
Correcting the state populations using the net undercount estimates developed under each scenario and running the corrected figures through the currently used apportionment formula—a method called “equal proportions”—gave the following results. Only the fifth and sixth scenarios changed the apportionment from that using the unadjusted census figures. Under scenario (5), Alabama gained one seat and California lost one seat, while, under scenario (6), Alabama gained one seat at the expense of Oklahoma.
Carlucci (1980) noted that a subsequent Census Bureau study (Siegel et al., 1977) that developed alternative estimates of net undercoverage in the 1970 census for states showed a greater impact of coverage errors on apportionment. Adjusting the census figures with one set of estimates developed in this later study produced a change of one seat between Tennessee and Oklahoma, while the use of another set produced changes of two seats involving California, Texas, Ohio, and Oklahoma.
Kadane (1984), in a study of the consequences of coverage errors in the 1980 census for reapportionment, developed estimates of the population by state based on the results of the 1980 Post-Enumeration Program, specifically the PEP series 2-9 estimates (see Chapter 4 for a description of PEP). Application of Kadane’s estimates for reapportionment gave California an additional seat at the expense of Pennsylvania.
Finally, a simulation study performed by Gilford (1983) of congressional apportionment based on different sets of state population estimates from the PEP showed that the results are sensitive to the estimates used. Gilford contends that the PEP results should not be used for adjustment because (p. 2): “adjustment of state population counts can cause counter-intuitive changes in apportionment,” and “the extreme volatility of apportionment results based upon adjusted census counts—attributable solely to the random characteristics of the particular PEP sample selected—renders the PEP unsuitable as a basis for adjusting the census for apportionment purposes.” It should be noted that the counterintuitive changes reported by Gilford are largely the result of the fact that states are not allocated fractional representation. As Siegel (1975:13-14) commented, “Under the method . . . used to determine the number of Congressmen from each State, the shift in the population of a State required to produce a change in the State’s representation, may be merely a few hundred persons or a few hundred thousand
persons, depending on the precise populations of all the States.” Moreover, Gilford used all 12 sets of PEP estimates, some of which are regarded as less plausible than others. There remain the conclusions that: (1) coverage errors in the 1970 and 1980 censuses affected at least one or two congressional seats and (2) considerable uncertainty remains as to the particular states (both winners and losers) that might have been affected. However, with respect to point (2), in Gilford’s analysis for 1980, California gained at least one congressional seat under every scenario explored.
Effects of Errors on Redistricting
Determining the boundaries of congressional election districts, as well as districts for state legislative offices, is a state function. Prior to the “one man, one vote” Supreme Court decisions in the early 1960s (Reynolds v. Simms and Baker v. Carr), the states redistricted when absolutely necessary because reapportionment changed the number of seats and on occasion when the party in power believed it would be advantageous politically. States were notorious in allowing districts to vary greatly in population size. The Supreme Court decisions mandated strict requirements for population equality (no greater than a 1 percent difference in population between the largest and smallest congressional district and no greater than a 10 percent difference among state legislative districts—see Carlucci, 1980) as well as compactness and contiguity of districts. In addition, the Voting Rights Act and growing awareness by minority groups of the effect of the composition of election districts on their political strength led to demands backed up by court actions for equal representation of important population groups. Hence, today, in addition to redistricting due to reapportionment (Bureau of the Census, no date-b:1):
States and localities are forced to redistrict because of challenges brought in court or because the Justice Department clearance mandated by the Voting Rights Act fails to occur. Between 1967 and 1978, some two dozen cases concerned with state or congressional redistricting went to the Supreme Court.
Census small-area data are critical for the modern redistricting process to meet the standards set by the courts. Currently, P.L. 94-171 requires that the Census Bureau transmit small-area population data to each state within 1 year after Census Day. To permit achieving equal population size among districts, the 1980 census P.L. 94-171 computer tapes provided to the states by April 1981 tabulated the population for each city block or enumeration district in unblocked areas. Although not required by law, the Census Bureau added race and Hispanic origin data to the tapes, as
it was clear that states would need these data to justify their plans to the Justice Department and to survive court challenges. The states are currently requesting that the Census Bureau provide P.L. 94-171 data in 1991 with the addition of separate counts of the voting-age population.
It is possible that differential coverage errors among population groups and areas could affect the degree of population equality actually achieved by a redistricting plan based on the decennial census data. The study by Siegel (1975), mentioned above, assessed the likely effects of coverage errors in the 1970 census on the composition of districts within a state or city. Given a predetermined number of seats, Siegel noted that a new legislative district must be carved out of existing districts, one of which must be eliminated. The possibility that adjustment of the census counts would result in an additional seat going to an area within a state or city at the expense of another area depends on the average size of the districts, the differential coverage rates of major population groups, the proportionate distribution along areas of the major groups, and the number of contiguous districts with high undercoverage rates. Siegel’s analysis, assuming different rates of coverage of the white and black populations and different proportions of whites and blacks within areas, indicates that the possibility of a shift in the number of congressional districts for the regions of a state was very small. Siegel estimated that the chances of a shift in state legislative districts or city council districts were somewhat greater, but still small.
Carlucci (1980) applied Siegel’s calculations to New York City, as an example, and judged that an adjustment for undercount in 1970 would not produce additional representation for that city. However, Carlucci pointed out that, if one made the additional assumption that all groups other than whites were undercounted at the same rate as blacks, given that New York City’s population in 1970 was one-third nonwhite, it appeared very likely that the city would have been entitled to another congressional seat plus additional seats in the state legislature.
If the Census Bureau were to adjust the counts for use in redistricting, there is the problem that P.L. 94-171 imposes very tight time constraints for delivery of tabulations to the states. In 1980, even when adjustment had been ruled out, the tight timetable limited the checking that the Census Bureau could accomplish, and the redistricting counts consequently contained processing and geocoding errors (Bureau of the Census, no date-b:2).
The rush to get them out necessitated less than complete checking of all of the thousands of small area counts in each state, and there were errors. The Bureau discovered many of these errors in routine reviews, and the redistrictors discovered many as they began to attempt to draft new districts. In general, these errors were geographic misallocations, that is, people were counted but in the wrong block. Occasionally, processing
errors were discovered and it was necessary to add persons to the count, rather than shift numbers from one block to another.
The Bureau issued a series of count corrections in the months and years following the distribution of the redistricting figures. In some cases, these corrections were enough to make it appear that the distribution achieved in the original plan drawn on the basis of the April l, 1981 numbers was not as good as it should be. Redistrictors began to question how final the P.L. 94-171 numbers were and whether or not the states should draw new plans. The Bureau’s reply was that it had furnished the best numbers it had on April 1, 1981 and the counts were final as of that date, but subsequent reviews had surfaced that would be corrected as soon as they became known. Such a process had taken place following previous censuses. The correction process caused concern in many states, but . . . no state was forced to draw a new plan on the basis of corrections alone.
Not only can processing errors discovered after the fact and differential coverage errors affect the population equality achieved by a redistricting plan, but also the passage of time between censuses obviously affects the relative representation among districts that differ in their rate of population growth. However, it appears that states do not want to draw new plans more often than once after each decennial census because of the difficulty of getting any plan approved by the legislature and through Justice Department clearance and/or court challenges. In fact, members of the Conference of State Legislatures stated that, if the mid-decade census authorized by P.L. 94-521 (but never funded) were to provide small-area statistics, they would go to Congress and seek to have the law changed to preclude the use of such data for redrawing state legislative districts. (The law already includes a prohibition against the use of mid-decade data to reapportion the House of Representatives or to draw new congressional districts—see Bureau of the Census, no date-b.) Nevertheless, it is worth noting that apportionment and districting plans are used for about 10 years, during the course of which substantial population shifts occur whose effect on the “one man, one vote” principle probably dwarfs any effects introduced by errors in the census counts. These effects tend to disadvantage areas of disproportionate population growth.
Effects of Errors on Fund Allocation
From a small beginning around the turn of the century, federal grant-in-aid programs have grown enormously in scope and amount. Prior to 1930, federal grant programs for state and local assistance were limited in purpose, used simple allocation formulas involving factors such as total population or population density, and consumed less than 3 percent of the federal budget. As of the 1980 census, there were over 100 programs that
distributed money to states and localities for a wide range of purposes by means of complex formulas including census or census-based data elements. A conservative estimate of the funds distributed via formula in fiscal 1981 was $80 billion, close to 15 percent of the federal budget (Emery et al., 1980). As of fiscal 1984, funds distributed by formula amounted to about $80 billion, or about 10 percent of the budget (Office of Management and Budget, 1985). In addition, many states have formula-based programs of aid to their local governments. Table 2.1 lists selected federal grant programs and indicates the data elements used in fund distribution. (See Appendix 2.1 for a description of some state grant programs.)
There is no requirement for federal funds to be allocated according to formula comparable to the constitutional mandate for reapportionment or court requirements for redistricting. Indeed, the present administration has worked to reduce the scope and extent of both formula-based and categorical federal grant-in-aid programs. However, it is clear that Congress has become accustomed to use formulas that eliminate the need for case-by-case decisions regarding fund applications from states and localities. It is likely that some formula-based programs will continue and that new programs of state and local assistance will in many cases be formula based. Hence, research on the effects of errors in the census on the distribution of federal funds should contribute importantly to the making of sound choices for decennial census methodology.
Unfortunately, the available research to date in this area is severely limited both in method and in scope. No research has been completed that looks at the total set of grant programs; most research has concentrated on one program—general revenue sharing. Several factors have prevented a comprehensive analysis, including the lack of documentation of the formulas used and the complexity of many formulas. Emery et al. (1980:74, 77) in a study attempting to document all of the formulas in use, noted:
Among the formula grant programs . . . one-fourth failed to report the existence of a formula to OMB while others reported the existence of a formula but did not specify the factors involved. The lack of central documentation and the variability in agency documentation cause a large part of the uncertainty concerning how statistics affect assistance payments. . . . Notwithstanding the considerable vested interest and controversy surrounding the topic, the total number of programs having statistical allocations, the amount of money involved and the quality of data employed in calculating payments are unknown quantities.
With regard to complexity of the formulas, the same authors note (p. 77) that “the simplest allocation formulas involve a calculation of a State’s share of dollars based on the State’s share of the total U.S. population.
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|Adult Education Act (P.L. 89-750)||
Title III: allocates same base amount to all states and then remainder based on state share of persons age 16 and older with less than 4 years of high school completed (excluding persons ages 16-19 currently enrolled).a
|$0.8 (includes career and vocational education)|
|Career Education Incentive Act (P.L. 95-207)||
Allocates funds to states based on state share of population ages 5-18.b
|See adult education|
|Education Consolidation and Improvement Act (P.L. 97-35)||$3.4|
Allocates 87% of available funds to states based on state share of children under 18 in AFDC families and children under 6 in families in poverty.a
|Higher Education Act (P.L. 89-329)||
Title IV-C, Work Study Program: allocates 90% of funds to states as follows: 1/3 based on state share of persons enrolled full time in postsecondary schools; 1/3 based on state share of high school graduates; 1/3 based on state share of related children under 18 in families with income under $3,000.a
|Public Libraries (P.L. 84-597)||
Allocates funds to states based on share of population.b
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|Vocational Education Act of 1963 (P.L. 94-482)||
Title I, Part A, State Vocational Education Programs: allocates 50% of funds to states based on share of population ages 15-19; 20% based on share of population ages 20-24; and 15% based on share of population ages 25-65.b
|See adult education|
|Employment and Training|
|Job Training Partnership Act (P.L. 97-300)||
Eligible Service Delivery Areas (SDAs) must be one or more counties or cities of 200,000 or more current population (or a rural CETA prime sponsor or an exception approved by the governor).b
|$ 3.0 (all programs)|
Title II, Part A, Adult and Youth Programs: allocates 1/3 of funds to states based on state share of unemployed in areas of substantial unemployment; 1/3 based on state share of excess unemployed; 1/3 based on state share of economically disadvantaged population. Uses same formula to allocate 78% of each state’s funds to SDAs.c
Title II, Part B, Summer Youth Employment and Training: uses Part A formulas.
Title III, Dislocated Workers: allocates 1/3 of funds to states based on state share of unemployed; 1/3 based on state share of excess unemployed; and 1/3 based on state share of persons unemployed 15 or more weeks. State-required matching percentage is reduced 10% for each l% higher than average unemployment in previous year.c
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|General Revenue Sharing|
|State and Local Fiscal Assistance Act of 1972 as amended, Title I||
Allocates funds to states according to 1 of 2 formulas: (1) allocates funds based on state share of total population times tax effort (state and local taxes divided by personal income) times ratio of state to national per capita income.b,d (2) allocates 2/9 of funds based on state share of total population; 2/9 based on state share of population divided by per capita income; 2/9 based on state share of urbanized population; 1/6 based on state share of tax effort; and 1/6 based on state share of state income taxes.b,d
Allocates 100% of state funds to general units of government in each state based on local-government-unit share of total population for units in the state times tax effort times the ratio of government unit to state per capita income.b,d
|Community Development Block Grants (Housing and Community Development Act of 1974 as amended, Title I)||
Eligible areas are cities with 50,000 or more population, metropolitan counties with 200,000 or more population, and some nonmetropolitan areas.b
Allocates 80% of funds to cities and counties according to 1 of 2 formulas: (1) allocates 1/4 of funds based on area share of total population for all eligible areas; 1/2 based on area share of persons in poverty; 1/4 based on area share of overcrowded dwelling units with more than l.01 persons per room.a,b (2) allocates 1/5 of funds based on area share of the growth lag for all areas; 3/10 based on area share of persons in poverty; and 1/2 based on area share of older housing built before 1940.a,b
|$ 4.4 (includes Urban Development Action Grants)|
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|Federal Housing Act of 1949 as amended||
Title 502, Housing Assistance Programs: Eligible areas to receive insured and/or guaranteed loans include nonmetropolitan areas with under 10,000 population and areas between 10,000 and 20,000 that face credit-shortages.b
Allocates 3/10 of funds for insured loans to states based on state share of rural population living in inadequate housing; 3/10 based on state share of total population; 3/10 based on state share of rural population in poverty; 1/10 based on state share of per capita housing cost.a,b
States distribute funds for insured loans to districts (groups of counties) and from districts to counties using same formula.
Funds for guaranteed loans are distributed using similar formulas, except that the share of rural households with incomes between $15,000 and $20,000 replaces the rural poverty factor.
|Aid to Families with Dependent Children (AFDC—Social Security Act, Title IV)||
Determines federal matching percentage of state expenditures based on state share of 3-year average per capita income.d
|Low Income Home Energy Assistance (P.L. 97-35)||
Allocates funds to states based on households below the lower living standard income level and below 125% of poverty.a
|Medicaid (Social Security Act, Title XIX)||
Determines federal matching percentage of state expenditures based on state share of 3-year average per capita income.d
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|Construction Grants for Wastewater Treatment Works||
Allocates 1/2 of funds to states based on formula A and 1/2 based on formula B: (A) allocates 1/4 of funds based on state share of total population and remainder based on state share of need (based on construction costs and population projections).b (B) allocates funds based on maximum of state share of total population and state share of needs.b
|Urban Park and Recreation Recovery Program (P.L. 95-625)||
Title X: Eligible areas for funds are central cities of metropolitan areas, places of 40,000 or more population, and counties of 250,000 or more population that score above the median on a composite variable including population density, net change in per capita income, percentage of unemployed, percentage of households with cars, population under 18 and 60 and older, and percentage of population in poverty.a,b
|Community Services Block Grants (P.L. 97-35)||
Allocates funds to states based on state share of population in poverty.a
|Older Americans Act of 1965 as amended (P.L. 89-73)||
States must submit a plan for services and designate Planning and Service Areas (PSAs), generally as counties or groups of counties based on total persons 60 and older and low-income persons 60 and older.a
Outreach required of PSAs with large numbers of persons 60 and older with limited English ability.a
Title III, Parts B and C, Supportive Services and Senior Centers and Nutrition Service: allocates funds based on population 60 and older.b
Some states allocate funds to PSAs based on each PSA’s share of persons 60 and older below poverty.a
|Program||Data Items Used for Allocation and/or Eligibility||Fiscal 1984 Expenditures (billions)|
|Runaway Youth Act (P.L. 96-509)||
Allocates funds to states based on number of children under 19.b
|Social Services Block Grants (Title XX of Social Security Act)||
Allocates funds to states based on state share of total population.b
|Highway Research, Planning and Construction (Title 23, U.S. Code)||
Primary Systems Program: allocates 2/9 of funds to states based on state share of total land area; 2/9 based on state share of rural population (including places under 5,000 outside urbanized areas); 2/9 based on state share of mail delivery route mileage; and 1/3 based on urban population.a
|$11.2 (all programs)|
High-Hazard Locations Program: allocates 3/4 of funds to states based on state share of total population and 1/4 based on state share of public road mileage.b
|Urban Mass Transportation Act (modified by 1982 Surface Transportation Assistance Act)||
Section 5: provides funds for approved projects of metropolitan transportation agencies. Surface transportation entitlement is determined for an urbanized area based on average of its share of total urbanized area population and its share of population density.a
NOTES: Except for AFDC and Medicaid, the programs distribute shares of a fixed amount of funds. This is either because the allocation formula is explicitly share based or because the amounts allocated are proportionately reduced to fit within an appropriations ceiling for the fiscal year. The allocation formula descriptions in the table omit many features affecting fund distribution such as hold-harmless and minimum and maximum amount provisions.
N.A. = Not available.
aDecennial census data are the only reliable source for some or all formula elements.
bCan use census-based current population estimates and/or Current Population Survey (CPS) data for some or all formula elements. CPS data are controlled to census-based current population estimates.
cData source is Bureau of Labor Statistics (BLS) local area unemployment estimates. These are calibrated to the CPS, which is calibrated to census-based current population estimates.
dData source is Bureau of Economic Analysis (BEA) per capita income estimates, based on BEA personal income estimates divided by census-based current population estimates.
SOURCES: Bryce (1980); Emery et al. (1980); Gonzalez (1980); Herriot (1984: various unpublished attached documents such as copies of laws provided by federal agencies); Maurice and Nathan (1982); Office of Management and Budget (1985).
However, most allocation formulas are far more complex, involving more than one statistical factor and constraints such as minimum and maximum awards.”
Two key aspects of fund allocation formulas affect the extent to which errors in the census result in inequitable distribution of grant program monies. First, whether a program distributes funds on a per capita basis or as shares of a fixed total sum is a major determining factor in whether adjustment for census errors will cause a large change in the amount of funds an area receives. Obviously, errors in coverage relate directly to maldistribution of funds under programs that operate on a per capita basis. In contrast, maldistribution of funds under programs that allocate shares of a fixed total will generally occur only if the eligible areas experienced significantly different rates of net undercoverage. The exception is a program with a share-based formula that also includes an eligibility threshold; in this case, coverage errors will directly affect the number of jurisdictions that are eligible to share in the fund allocation and hence will affect the distribution of the fixed total amount. At present, almost all grant programs that use formulas operate to distribute shares of a fixed total, either because their formulas are explicitly share-based or because of ceilings on the total amounts that per-capita-based programs can disburse during a fiscal year. Only a few of these programs include eligibility thresholds (see Table 2.1).
The second important aspect of a program’s formula is whether it includes only population counts or whether there are additional factors. To the extent that other factors dominate the formula, errors in coverage per se have less effect on fund distribution under the program.
Siegel (1975) analyzed the implications of adjusting the census counts in each state for a program that was assumed to allocate $1 billion purely according to each state’s share of total population. He found that, depending on which set of population estimates was used (see description of his scenarios in the discussion of reapportionment above), only 5 to 11 states experienced a 1 percent or greater shift in their fund allotment even though 50 states (including the District of Columbia) had estimated net undercounts of 1 percent or greater under each scenario. The scenario that had the greatest effect modified the national undercount rates by race to take account of median family income; under this scenario 6 states experienced a shift in funds of 1 to 1.9 percent, 4 states a shift from 2 to 2.9 percent, and the District of Columbia a shift in funds of over 4 percent.
Most research in this area has focused on the general revenue sharing program, first authorized in 1972. The program distributed over $5 billion to 39,000 governmental units including states and localities in fiscal 1981 and over $4.5 billion to localities in fiscal 1984 under formulas based on population, per capita income, and tax effort factors. (States no longer receive revenue sharing funds, but the program still determines first the
amount to be allocated in total to the localities in each state and then applies a separate formula to determine the share of each state’s total for specific localities—see Table 2.1.)
Siegel (1975), in an extension of the analysis just described, simulated the distribution of revenue sharing funds among the states. He compared the distributions using unadjusted 1970 census population and income data with distributions using: (1) adjusted population data but assuming that per capita income remained as before (i.e., assuming that uncounted persons had the same income as counted persons); (2) unadjusted population data but per capita income data adjusted to Bureau of Economic Analysis control totals; and (3) adjusted population and per capita income data.
The results showed that simply adjusting population never made any large numbers of changes in funds apportioned to the states under general revenue sharing. Using the basic synthetic adjustment of population by age, race, and sex, the distribution of funds shifted by more than 1 percent for only 5 states and by more than 2 percent for only the District of Columbia. Using a modified population adjustment based on median family income, 8 states experienced a shift of 1 percent or more and 5 states a shift of 2 percent or more. Adjusting per capita income alone resulted in more significant changes—25 states experienced a shift of 1 percent or more and 14 states a 2 percent or greater shift, with 4 of those states experiencing a shift of 6 percent or more in their share of funds. Adjusting population and per capita income together also resulted in a larger number of changes, especially using the modified population adjustment based on median family income together with the income adjustment—under this scenario fully 32 states experienced a shift of 1 percent or more in their fund allocation, 17 states a shift of 2 percent or more, and 5 states a shift of 6 percent or more.
Several studies have examined the effect of census errors on distribution of revenue sharing funds to localities. Siegel (1975:22) noted that “the role of the income component is even more dominant when the General Revenue Sharing formula is applied to counties and Cities,” because the income component in the tax effort factor as well as the per capita income factor requires adjustment. In addition, the formula for allocation to localities includes constraints so that no local area may receive less than 20 percent or greater than 145 percent of the state’s average per capita payment or more than 50 percent of the sum of its taxes and intergovernmental transfers. Siegel (1975:23) concluded that prior studies (Hill and Steffes, 1973; Savage and Windham, 1973; Strauss and Harkins, 1974; Grindley et al., 1974) “all fail to make adequate allowance in the application of the formula for the understatement of the income component or to take account of the apportionment features of the Act.”
Robinson and Siegel (1979) carried out an illustrative study of the effects of 1970 census coverage and income reporting errors on distribution of
general revenue sharing funds to localities within the states of Maryland and New Jersey. The results were similar to the findings in the earlier Siegel study for states in that adjustment of income had a greater effect than adjustment of population on shifts in the distribution of funds; however, the effects of adjustment were greater for local areas than for the states. States and local areas experienced similar average percentage shifts in fund distribution with just the population factor adjusted—0.6 percent for the 50 states and D.C., 1.0 percent for the 155 local jurisdictions in Maryland, and 0.7 percent for the 567 local areas in New Jersey. With income alone adjusted, the average percentage shift in funds was 1.8 percent for the states, 4.1 percent for the Maryland local areas, and 4.4 percent for the New Jersey local areas; while, with both income and population adjusted, these figures became, respectively, 1.9 percent, 8.5 percent, and 9.1 percent. The local areas most affected by adjustment were those not constrained by the minimum and maximum allotments specified in the formula.
Another way of looking at the impact of census errors is on a per capita basis; that is, how much lost revenue from various fund allocation programs each additional uncounted person represents for a state or local area. Maurice and Nathan (1982) undertook to answer this question for three different programs: (1) general revenue sharing, (2) the Community Development Block Grant program, and (3) mass transit subsidies provided under section 5 of the Urban Mass Transportation Act. They investigated the simultaneous impact of a synthetic population adjustment using 1970 census national net undercount rates by race for 573 areas (central cities of standard metropolitan statistical areas or cities with more than 50,000 residents). Over half the cities had estimated net undercount rates of greater than 2 percent and almost one-fifth had estimated net undercount rates exceeding 3 percent with the application of the synthetic adjustment.
Maurice and Nathan (1982:253) note that assertions are often made that each uncounted person represents a significant sum of money lost to a jurisdiction; for example, the New York City planning department estimated that the city would lose $200 per year in federal aid for each resident missed in the census. In contrast, they find (1982:266): “For the majority of cities, the total change in allocation [for the three programs] resulting from an undercount adjustment of population is in the range of plus or minus $5 per uncounted person.” For 18 large cities, the total change ranged from a loss of $11.80 (for Minneapolis) to a gain of $15.40 (for Philadelphia). They explain this result as a consequence of three phenomena: (1) the synthetic method of population adjustment produces small changes in cities’ shares of the national population, (2) population is not the only factor in most fund allocation formulas, and (3) one of the most important programs—community development block grants—includes a population growth lag variable in one formula
used by older distressed cities that gives larger allotments to cities with larger net undercounts.
Maurice and Nathan found greater effects of adjustment for coverage errors in the public service employment portion of the Comprehensive Employment and Training Act (CETA) Program. For selected large cities, the change in fund allotment under this program ranged from a loss of $2 per uncounted person for Los Angeles to a gain of $35 for New Orleans. However, they note that this finding could be sensitive to the assumptions used regarding the labor force status of uncounted persons. They also note that this component of CETA was not included in its successor program, the Job Training Partnership Act.
Finally, most of the studies of the effects of census errors on fund allocation have found that, typically, more jurisdictions “lose” by an adjustment than “gain” compared with the distribution of funds using unadjusted census data. For example, Robinson and Siegel found that 31 states, 114 (of 155) local areas in Maryland, and 315 (of 567) local areas in New Jersey would have been worse off with both population and income adjusted than when unadjusted census data were used for the allocation of general revenue sharing funds. However, this type of analysis suffers from two problems. First, as discussed more fully in Chapter 7, the appropriate standard of comparison in determining winners and losers resulting from an adjustment procedure is not the unadjusted census count but the true population (see also the discussion in Bryce, 1980:119-120). Second, the panel argues in Chapter 7 that an adjustment should be evaluated not by counting the number of areas gaining or losing but by taking into account the population size of each area. (In this regard, the “winners” in Maryland in the Robinson and Siegel study included the city of Baltimore, with 23 percent of the state’s total population.)
Summing up, there appears to be evidence that coverage errors affect fund allocation, but to a relatively small degree. Errors that do have an appreciable effect are those related to income reporting. We should caution, however, that the studies reported in the literature are of limited scope and are not simply generalizable to all federal fund allocation programs. Furthermore, the adjustment techniques used were by and large unsophisticated; adjustment procedures that introduced greater complexity, such as the Siegel procedure that modified national net undercount rates by race to take account of state median family income, generally produced larger effects.
As mentioned previously in this report, the census can be viewed as part of a more comprehensive statistical system providing both census and
postcensal data for the uses described above, including the distribution of funds and other mandated purposes. It is instructive to compare the effect of errors in the census with those arising in the updating process as well as from using outdated census information because more current statistics are not available.
The Census Bureau recently completed an evaluation of the quality of county and subcounty postcensal population estimates and made the results available to the panel. The evaluation was carried out by preparing population estimates for 1980 in the same ways they were made during the postcensal years of the 1970-1979 decade and comparing the estimates with the 1980 census counts (see Starsinic, 1983, for a description of the method and a report comparing 1980 estimates with the census counts for states). The comparisons show that the size of the errors in the postcensal estimates for areas below the state level dwarf those in the census. This is not a reflection on the Census Bureau. A considerable amount of research has been conducted on the methodology for population estimation and the estimates have been improving over the years. However, there are at present inherent limitations in the databases used to prepare the estimates and statistical manipulation can only partially correct for them.
Tables 2.2 through 2.5 (extracted from forthcoming Census Bureau publications) illustrate the nature of the problem. The 1980 estimates for 7.1 percent of the 3,142 counties in the United States had errors of 10 percent or more. The errors tended to be concentrated in the smaller counties: 18.8 percent of counties with population under 5,000 and 9.4 percent of those between 5,000 and 10,000 had errors of 10 percent or greater.
|Percentage of Error||Metropolitan Counties (%)||Nonmetropolitan Counties (%)|
|Less than 1.0||19.8||15.5|
|1.0 to 2.9||33.8||29.2|
|3.0 to 4.9||20.5||23.2|
|5.0 to 9.9||20.2||24.5|
|10.0 or more||5.6||7.5|
Average absolute percentage of error
NOTE: Several different population estimation methods are used by the Census Bureau. These data and the data in Tables 2.3, 2.4, and 2.5 are for the method with the smallest absolute errors in 1980.
SOURCE: Unpublished Bureau of the Census tabulations.
|Population of County||Average Absolute Percentage of Error||Percentage of Counties with Errors of 10.0% or More|
|Less than 5,000||6.1||18.8|
|5,000 to 9,999||4.8||9.4|
|10,000 to 24,999||4.1||7.1|
|25,000 to 49,999||4.0||5.5|
|50,000 to 99,999||3.8||3.2|
|100,000 or more||3.0||2.2|
SOURCE: Unpublished Bureau of the Census tabulations.
|Percentage of Error||Percentage of Places|
|–25.0 or more||6.5|
|–24.9 to –15.0||8.6|
|–14.9 to –10.0||8.2|
|–9.9 to –5.0||12.7|
|–4.9 to –0.1||15.4|
|0.0 to 4.9||14.1|
|5.0 to 9.9||9.7|
|10.0 to 14.9||6.7|
|15.0 to 24.9||7.6|
|25.0 to 49.9||7.0|
|50.0 or more||3.4|
NOTE: There were 35,644 places for which estimates were made and evaluated.
SOURCE: Unpublished Bureau of the Census tabulations.
However, errors of this size were not solely a small county phenomenon. Of the 412 counties with 100,000 or more persons, 2.2 percent were off by 10 percent or more. Of course, the evaluation covered 1980 and the errors accumulate over time, so that these results probably reflect the situation only in the last few years of the decade. Even so, the potentially large impact on uses of the data is disturbing.
The situation is even more serious at the subcounty level. The average absolute percentage error among the 35,644 subcounty areas analyzed was 15.2 percent. As in the case of counties, the smaller areas were subject to
|Population of Area||Number of Areas||Average Absolute Percentage of Error||Percentage of Areas with Errors of:|
|Less Than 10%||10% to 19.9%||20% or More|
|100,000 or more||160||3.9||95.6||4.4||-0-|
SOURCE: Unpublished Bureau of the Census tabulations.
greater errors, with the average percentage error ranging from 35 percent for areas with less than 100 persons to 4 percent for those with over 100,000 population. Forty-eight percent of all areas had errors of 10 percent or greater. Of the 160 areas with 100,000 or more persons, 4.4 percent had errors between 10 and 19 percent. Both positive and negative errors existed. For example, of the 6,012 places with errors of 25 percent or more, the errors were in the negative direction for 2,320 places and in the positive direction for 3,692. The difference in population estimates between some pairs of places could thus be off by more than 50 percent of their populations.
These errors contrast with those in the census, where even the black-white differentials in coverage are not large enough to make it likely for places to be undercounted by more than a few percentage points. A comparison of the two sets of errors suggests that the emphasis on census errors in the past few years has been somewhat misplaced, and that users of the data would have been better served if some of the funds used to reduce undercoverage in the 1980 census could have been used to improve postcensal data.
A detailed analysis of the postcensal estimates is not within the scope of the charge to the panel. However, we strongly urge the Census Bureau to examine the cost-effectiveness of a mid-decade census compared with the cost-effectiveness of the extra effort required to achieve the last one-half to one percent coverage improvement in accuracy of the decennial census. If, as we suspect, a mid-decade census would significantly improve the
usefulness of the data for key purposes, such as allocation of federal and state funds, compared with marginal coverage improvement efforts in the census, this fact should be transmitted to the administration with a strong recommendation that funds be budgeted for a mid-decade program for 1995. We realize that diverting some coverage improvement funds from the decennial census to the mid-decade census will only partly support the latter program. The additional support needed would be more than justified, in our view, if further study demonstrates the value of a mid-decade census for importantly improving overall data quality. A mid-decade census program may also afford operational advantages for census-taking, such as facilitating retention of experienced staff, that would further improve data quality and/or reduce the per person costs.
We recognize that the temper of the times is not conducive to the initiation of new programs, but we believe that statisticians have the responsibility to describe the facts and recommend the actions they believe are sensible. We think it highly likely that reallocation of funds from marginal efforts to achieve small reductions in the decennial census undercount to a mid-decade program would improve overall data accuracy and thus contribute to equitable political representation, fund allocation, and public administration. The panel urges that these issues be thoroughly explored before the 1990 census plans are finalized.
Recommendation 2.1. We recommend that the Census Bureau assess the need for a mid-decade census, particularly by studying the effect of errors in postcensal population estimates compared with errors in the decennial census on major data uses. Unless these studies do not support the value of a mid-decade census, the Census Bureau should proceed with preparations and make every effort to secure funding to conduct a census in 1995.
Government agencies at all levels—federal, state, and local—are heavy users of census data. This appendix reviews typical applications of census data made by state and local agencies.3 At these levels of government, the decennial census is an invaluable and unmatched resource in providing comparable small-area and subgroup data.
State governments use census tabulations in ways that are similar to federal and local uses and in ways that are unique to the states’ role in the federal system. Based on a review of uses specified by a reasonably representative group of states (Alaska, Connecticut, Florida, Georgia, Illinois, Indiana, Iowa, Missouri, Montana, New Jersey, New York, Oregon, Tennessee, Virginia, and Wisconsin), the kinds of applications described below are typical for this level of government.
Use for Redistricting
The states determine the boundaries of congressional election districts, as well as districts for state legislative offices. Under the “one man, one vote” requirements imposed by the courts for equal population size and compactness of districts, small-area census data are essential for the task of redistricting. The chapter text indicated the data requirement for this use of the decennial census figures and reviewed potential problems posed by differential undercoverage and by discovery of other kinds of errors, such as processing mistakes, subsequent to release of the redistricting tabulations 1 year after Census Day.
Use to Classify Local Governments
All the states denominate various categories of local governments, such as municipalities or townships, by population size and accord varying rights and responsibilities to each size class. For example, compensation of county clerks in Missouri is established as a function of population size and assessed valuation. This application uses census figures as thresholds, and
3 Much of the material in this section comes from a survey of federal, state, and local government agencies initiated by the Census Bureau in fall 1982 requesting information on specific needs for subject matter and geographic detail from the census for uses mandated in legislation. The responses are summarized in Herriot (1984). Many agencies indicated other kinds of uses in addition to mandated ones.
hence coverage errors can be important if a locality is put in the wrong size class. However, many state statutes include language that permits localities to submit alternative population counts, for example, from special censuses.
Use to Allocate State Funds
Many states have programs to allocate state monies to localities on the basis of formulas similar to federal programs like general revenue sharing. For example, the State of Alaska has a state revenue sharing program that distributes money to municipalities and unincorporated places. The State of Florida allocates its 2 cents per gallon gasoline tax to counties via a formula that includes three terms for each county:
One-fourth the ratio of the county land area to the state, plus one-fourth the ratio of the county population to the state, plus one-half the ratio of the county gasoline tax dollars to the state.
Most states with a motor vehicle fuel sales tax distribute the receipts using a formula including local population counts. Many states likewise distribute the proceeds of consumption or nuisance taxes, such as pari-mutuel, cigarette, and alcoholic beverage taxes, on the basis of population (Bryce, 1980:112-113). The State of New York allocates funds for building code enforcement to counties and cities using a formula that includes each area’s share of the total non-institutionalized population and of total real property valuation, while Iowa allocates day-care center funds on the basis of numbers of children under age 7 and low-income families. The equity of the distribution of monies under these various state programs is presumably affected by differential undercoverage. The chapter text discusses what is known about the effects of errors in the census count on fund allocation formulas for various federal grant programs.
Use for Equal Employment Opportunity Purposes
Every state in the nation has requirements, in legislation or executive order, for state agencies to implement one or more kinds of equal employment opportunity (EEO) or affirmative action programs with regard to hiring and personnel practices. State agencies make use of census data to establish affirmative action goals and to monitor how well equal employment opportunity programs are meeting their goals. The most common data requirements are for occupation by race and sex for counties and cities. Many states also need data on occupation and industry by age, veteran status, disability, and language spoken. After 1980, the Census Bureau provided a special EEO file that contained detailed occupation
cross-tabulated by sex, race, and Hispanic origin, plus years of school completed tabulated by age, sex, race, and Hispanic origin for counties, cities of 50,000 or more population, and metropolitan areas. The Census Bureau’s Data User Services Division sold over 330 copies of this file directly to users in addition to providing copies to all State Data Centers (from information furnished by Michael Garland, Chief, Data User Services Division).
There are many related applications of census data by the states in the area of antidiscrimination efforts. The State of Missouri anti-redlining statute requires the Department of Commerce to monitor bank compliance using data on the characteristics of the housing stock (number of units, tenure, etc.) and of the population (race and income) by census tract in several cities.
EEO applications of census data require tabulations of groups such as blacks and Hispanics that are known to be covered less well than other groups. Moreover, these applications require additional data such as occupation and income, and it may well be that errors or problems with these items have greater impact on the validity of conclusions drawn or actions taken on the basis of the cross-tabulations than simply differential undercoverage by race.
Use for Implementation of Federal Programs
Many federal programs that distribute funds to states and localities require applications for specific programs or projects rather than simply allocating dollars according to formula. States use census data to support grant applications of all kinds. For example, the State of Florida Department of Health and Rehabilitative Services needs data on the elderly population (persons age 60 and older) by race in each county to justify funds for social and nutrition services programs under the Older Americans Act; the Florida Department of State, Division of Library Services, needs census data on income by age, race, and Hispanic origin for counties and cities for funding under the Library Services and Construction Act.
Use for Statewide Planning
The states use census data for many kinds of planning purposes. Just to name a few examples, the Alaska Department of Natural Resources requires small-area data on population, income, employment by industry, household size, and permanent versus seasonal residence for planning various park and recreation programs. The Florida Department of Transportation has statute-based requirements for census data on population, density, income, auto ownership, and employment by occupation and industry for small areas for statewide transportation planning. The Florida Department
of Education needs census data on age, sex, education, income and poverty by county, and current county population estimates by single years of age for community college, state university, and adult education program planning. The Missouri Department of Agriculture uses county population categorized by age to plan publicity for the Missouri State Fair, and the Department of Mental Health develops measures of prevalence of mental disorders, alcoholism, and drug abuse, and plans service programs using census tract data. The State of Indiana uses county and census tract population and counts of housing units with basements in planning nuclear civil protection.
A related use is to determine workload needs for various state services. For example, under its Omnibus Crime Control and Safe Streets Act, the State of Montana uses census data for counties and cities on sex, race, age, and income to estimate personnel needs and workloads for public safety programs.
Local governments exhibit many of the same kinds of uses of census data as do the states, including use of the data for redistricting.4 If anything, localities have a greater need for census data for very small areas, such as blocks and tracts.
Typical census data uses cited by specific local governments include the following:
- Use for transportation planning, including planning of highways and other commuter transportation modes and forecasting airport demand (Orange County, FL; Pueblo Regional Planning Commission, CO; Corpus Christi, TX; Tri-County Regional Planning Commission, Harrisburg, PA; Lincoln City-Lancaster County Planning Department, NE; City of Detroit, MI).
- Use for planning local building and development projects (Houston, TX; Tri-County Regional Planning Commission, Harrisburg, PA) and for obtaining mortgage revenue bonds (Amarillo, TX).
- Use for services assessment and planning, such as needs assessments for human resources services in local community target areas using data on the elderly living alone, female-headed families, and
4 Much of the material in this section comes from the Census Bureau survey previously cited (see Herriot, 1984). This survey obtained responses from a small number of cities—less than 20—and most of these cities noted that they had relatively few mandated uses of census data. However, the examples of census data use discussed in this section appear to represent typical local applications.
children by census tract (Houston, TX); development of state-mandated community services area plan (Orange County, FL).
- Use to support grant applications for state and federal funds, for example, determination of transit subsidies from the regional transit authority using population and automobile availability by small area (Detroit, MI); applications to the state small communities program using data by block and tract on population, housing, employment, income, and poverty (Tri-County Regional Planning Commission, Harrisburg, PA); support of applications for family planning services project grants using data on ethnicity, age, income, and poverty for women ages 15-44 (County of San Diego Department of Planning and Land Use, CA).
Some formula grant programs, in addition to the categorical programs, place data requirements on localities above and beyond the need for the items that go into the formula. For example, the HUD Community Development Block Grant (CDBG) program has one set of data needs to determine fund allocation, another set to use in a Housing Assistance Plan (HAP) that each locality must develop before the CDBG funds to which the locality is entitled can be released, and yet a third set to monitor the impact of the program on housing for low- and moderate-income, minority, and female-headed households.
This appendix endeavors to sketch a picture of census data uses and users in one geographic location. New Jersey was chosen because of ready availability to the panel of relevant information. Examples of uses from all sectors—public, private, academic—are included.
Before describing users and uses, it will help orient the exposition to identify the various channels for distributing census data within New Jersey. The federal government offers documents, including census publications, for sale through the U.S. Government Printing Office and, by law, makes free reference copies available to the nation’s 1,350 depository libraries. Rowe has estimated (U.S. House of Representatives, 1982:424) that perhaps as much as 50 percent of census data use is by the millions of people who visit libraries every day to obtain needed information on a variety of subjects.
Census data in nonprinted form, including tabulations on computer tape (summary tape files), tabulations on microfilm and microfiche, and samples of individual microdata records (public use microdata sample files), are sold directly by the Census Bureau’s Data User Services Division (DUSD). Census tape files serve a growing need for more elaborate and extensive analysis than printed reports can readily serve. The tapes contain many more data items than can be printed in a manageable set of volumes and offer the advantage that the user can readily reprocess the data using computers. The availability of samples of microdata records (with identifying information removed) has greatly expanded the capabilities for original analysis and retabulation of the census responses to suit the user’s needs.
The Census Bureau has also set up a network of state data centers that receive publications and computer tapes containing the census tabulations for their state for redistribution to users. The typical structure includes a lead agency in the state government that works with the state library and one or more universities to provide a full range of user services, plus a number of affiliates that provide basic reference services throughout the state. Currently, there are data centers in 49 states, Puerto Rico, and the Virgin Islands (Riche, 1984b).
New Jersey is one of the states with an active state data center. The New Jersey State Data Center is housed in the Department of Labor and Industry and works with the New Jersey State Library and with Princeton
University and Rutgers University to provide a full range of processing and reference services to users. The center has as local affiliates the planning boards for each of the state’s 21 counties plus the Delaware Valley Regional Planning Commission. In addition, all county libraries receive State Data Center materials.
Finally, a growing number of private firms are in the business of supplying users with census (and other public) data. A recent survey by American Demographics (Riche, 1984a) identified 68 firms that repackage and resell government statistical data, one of which is located in New Jersey. While these firms handle some general information requests, most of their work is for clients who need specific tabulations or analyses that often require extensive processing of computerized census data. Many of these firms provide a range of services based on census data, such as profiles and projections of local area characteristics for site selection and market analysis; relating client information such as number of accounts to census characteristics for ZIP codes or other areas; and development of sampling frames and designs for local surveys. Other firms specialize in such services as using census data for election campaigning and voter registration drives, affirmative action planning and legal actions, and fulfillment of regulatory requirements. In fact, it is probably the case that these firms serve more users of census tapes than does the Data User Services Division. The DUSD supplies tape copies to users and will prepare special summaries of the confidential microdata tapes, but does not make extracts or special tabulations of publicly available data tapes. The DUSD filled over 5,100 orders for 1980 census computer tapes from 1981 to 1983 (from information furnished by Michael Garland), representing a small fraction of total user orders for tapes and analyses and tabulations produced from the data tapes.
The New Jersey State Data Center lead agency—the Office of Planning and Research (OPR) within the State Department of Labor and Industry—serves a large number of census data users each year. The agency has tracked data requests received by phone and reported that in 1982 phone requests totaled 3,600, rising to over 4,100 in 1983 (from information provided by Connie O. Hughes, director of the State Data Center). The increase resulted despite the policy effective July 1, 1982, of reduced direct service to the general public due to budget cuts incurred by OPR. Fully half the requests in 1983—over 2,000—were for data from the 1980 decennial census. Almost three-fifths of the census requests were received from other government agencies, about one-quarter from businesses, 10 percent from private individuals, and 7 percent from academia. Table 2.6 shows the
|Type of User||Data Source|
|Census of Population and Housing||Economic Censuses||Other Censuses||Other Federal Data||Lead Agency Data||Total|
Percentage of overall total
SOURCE: Connie O. Hughes, director, New Jersey State Data Center, personal communication to Constance F. Citro, March 1984.
distribution of requests by type of data (1980 decennial census, economic censuses, other censuses, other federal data, OPR data) and type of user.
The Princeton University Computer Center, which works closely with the lead agency, reported on a week’s sample of census use in spring 1984 (from information provided by Judith S. Rowe, associate director). Projects that took some amount of staff time included:
- A study of migration patterns and the demographic characteristics of the 1975 residents of the service area of a utility company in Texas, using the public use microdata sample files (PUMS);
- A study of migration from Long Island compared with migration from similar metropolitan areas, including age, occupation, income, and other characteristics of out-migrants, for a Long Island newspaper, using the PUMS;
- Development of profiles from the PUMS of recruitment pools (age, income, race) by district for the four military services;
- Construction of a dataset merging selected summary tape file 1 and summary tape file 3 data for minor civil divisions and unincorporated places on housing and homeowners for a private company that is supplying data to realtors;
- An analysis of voting behavior in Chicago using summary tape file 1 to define neighborhoods along race and ethnic lines for an undergraduate student in political science;
- An analysis of need and ability to pay for home health care using summary tape file 4 tables on age and income for a graduate student at Wharton, employed at a New Jersey hospital; and
- An analysis as part of a continuing study of commuting patterns in New Jersey using the Urban Transportation Package of special tabulations of place of work and journey to work data for a professor in the transportation program.
The Rutgers Center for Computer and Information Services, the other main component of the New Jersey State Data Center, described its activities for 1983 (from information furnished by Gertrude J. Lewis, project leader). The center keeps current copies of Rutgers University Guide to Machine Readable Data Files in all of the university’s libraries, and sophisticated users can access the available files, which include census and other datasets, without the center’s active help. In 1983, almost 30 different departments specifically requested the center’s machine-readable data files. For the decennial census files, the center handles phone calls from many business firms inside and outside the state. Users are encouraged to do their own computing with census data. Three times a year, the center offers seminars on using census files and also offers special seminars on request.
Examples of census data use at Rutgers include:
- Two faculty members in the Department of Sociology compared health needs of the poor with their service utilization using PUMS files and data from the National Center for Health Statistics;
- A faculty member in sociology and urban studies analyzed change in housing prices and characteristics between 1970 and 1980 at the minor civil division level both for research and instruction to undergraduates and graduates;
- A staff member of the affirmative action department used data from the special EEO file to construct figures on availability of minorities and women to determine utilization in the university’s work force;
- A faculty member in the Department of Agriculture/Economics compared state and county population figures among cities and places in the United States for 1970 and 1980;
- A professor in the Graduate School of Management carried out research on travel behavior between 1970 and 1980 with emphasis on transportation and the energy crisis;
- A graduate student in geography for his doctoral thesis used census population and housing characteristics to correlate the rate of subsidies at the census tract level in Manhattan;
- Undergraduate students in geography extracted census data and mapped the data using SAS/GRAPH;
- A graduate student in the School of Criminal Justice correlated census demographic data at the block group level with criminal data for his dissertation;
- A researcher in the Center for Urban Policy Research assisted research personnel throughout the year in accessing American Housing Survey and decennial census data. These projects covered a variety of topics, such as assessing population change for planning boards and studying segregation and integration within the state; and
- A research student in the Department of Agriculture/Economics accessed census data to analyze factors affecting employment change between 1970 and 1980 in rural communities in the United States.
The State of New Jersey regularly uses census data for many purposes, typical of state governments across the country. These uses include:
- Redrawing congressional and state legislative districts.
- Classifying local governments. A review of the state code in 1973 identified over 800 statutes that referenced population data; most of these references classified local governments by size and stipulated the rights and responsibilities of each class. For example, the term of office of street commissioner is 3 years in cities of the second class, with population of 100,000 to 250,000.
- Apportioning state funds to localities. New Jersey apportions motor vehicle fuel and general sales tax dollars to local jurisdictions based on population.
- Apportioning other kinds of services. New Jersey law states that localities may not grant retail liquor licenses in excess of 1 for every 3,000 population nor wholesale liquor licenses in excess of 1 for every 7,500 population, “as shown by the last then preceding Federal census” (although a municipality with fewer than 1,000 population can have one wholesale and one retail license in any case); members of the board of trustees for a community college that serves more than one county are allotted to each county based on population.
- Meeting equal opportunity requirements. New Jersey requires all agencies to develop equal employment opportunity plans and to monitor their progress in meeting EEO goals using data on the civilian labor force by race and sex for the state, counties, and cities; the Department of Banking uses data on housing stock characteristics such as number of units and tenure and on population
by race and income for all incorporated places to enforce the state’s anti-redlining statute.
- Approving applications. The Department of Banking approves applications for bank charters and bank branches based on economic feasibility determined from analysis of population, number and size of households and income by census tract for the area to be served; the Division of Mental Health and Hospitals allocates funds to community agencies according to past performance and need-based plans submitted by each agency that analyze data on age, income, marital status, race, and other characteristics for the census tracts and places served.
This page intentionally left blank.