National Academies Press: OpenBook

Using the American Community Survey: Benefits and Challenges (2007)

Chapter: 3 Working with the ACS: Guidance for Users

« Previous: PART I: Using the American Community Survey, 2 Essentials for Users
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

3
Working with the ACS: Guidance for Users

The American Community Survey (ACS) can benefit decision makers, planners, and analysts in virtually every type of setting, including federal executive and legislative agencies, state and local government agencies, metropolitan planning organizations, nonprofit organizations, professional associations, universities, think tanks, and private businesses in many sectors. The ACS will also be invaluable to educators, students, the media, and the public.

This chapter addresses how users can work with the various ACS products that are planned to become available and the factors to consider when deciding which products to use for particular purposes. Because not every potential application can be included (or indeed foreseen), the chapter highlights key applications for federal, state, and local government agencies, transportation planners, researchers, the media, and the public who currently use long-form-sample data. The specific users and applications that are discussed include:

  • Federal agency users (Section 3-A). Highlighted applications include the use of ACS 1-year, 3-year, and 5-year period estimates for fund allocation to states and localities (3-A.1) and to update the U.S. Department of Housing and Urban Development’s income limits for housing assistance programs (3-A.2).

  • State agency users (Section 3-B). A highlighted application is the use of ACS 5-year period estimates for fund allocation and grants to localities.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • Local government users (Section 3-C). Uses of different ACS estimates are discussed separately for big cities (3-C.1) and small, oversampled jurisdictions (3-C.2). Also discussed is the special case of jurisdictions with large seasonal populations (3-C.3).

  • Transportation planners (Section 3-D). Their applications will rely heavily on the ACS 5-year period estimates and also the public use microdata sample (PUMS) files.

  • Academic and other researchers (Section 3-E). Researchers will make heavy use of the PUMS files and of an ACS summary file, similar to Summary File 3 from the 2000 long-form sample that is currently under development.

  • Media outlets and the public (Section 3-F). These groups will likely make the most use of the ACS 1-year period summary estimates provided in profiles, ranking tables, and change tables.

Whatever their category (federal agency, local government, other), users should review other sections in addition to the one addressed to them. Many of the specific applications discussed—each of which illustrates some but not all issues regarding use of ACS data products to replace long-formsample data products—pertain to more than one category of user.

Section 3-G discusses an issue that affects all users—namely, the fact that new population and housing numbers from the decennial census every 10 years will likely interrupt the time series of ACS estimates. The reason is that the ACS estimates for calendar years ending in 0 through 9 each decade will be calibrated at the level of an individual county (or a group of small counties) to annual population estimates updated from the previous census by records of births and deaths and estimates of net migration. A similar calibration will be made to housing unit totals updated from the previous census. When a new census is taken, the census counts will not necessarily coincide with the updated estimates, thereby producing discontinuities in the ACS time series.

The chapter concludes (Section 3-H) by summarizing the panel’s general guidelines for effective use of the ACS and suggesting ways in which users who expect to work extensively with the ACS small-area data can prepare during the ramp-up period from 2006 to 2010, as the first sets of 1-year, 3-year, and 5-year period estimates become available. Many users are rightly concerned, first and foremost, with how well the ACS can serve as a replacement source of useful and usable estimates for planning, research, public education, and a host of other applications that currently rely on the long-form sample. The examples in this chapter serve principally to address this underlying concern about the functionality of ACS data to meet current needs.

The decoupling of long-form-type data from the once-a-decade census,

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

however, promises to allow the ACS to develop in ways that, while not clear today, will allow this new survey to become much more powerful than the long-form sample could ever be. We urge users to take a long view of the ACS and be open to new uses that were not possible with the long-form sample but that the continuously updated ACS data can support.

The Census Bureau, for its part, needs to provide as much guidance and training as possible to users to help them maximize the upside and minimize the downside of working with this complex data set. As discussed in Chapter 7, the Census Bureau should proactively identify ways to assist the occasional user who will not be in a position to master the ins and outs of the ACS data—for example, by highlighting estimates that meet reasonable standards for precision. The Census Bureau should also support an ongoing education and outreach program for users who plan to work extensively with ACS data, including the staffs of state data centers and other groups whose mission is to assist the broad user community. As discussed in Chapter 4, the Census Bureau should consider the development of new data products that would help many users, such as 3-year period estimates for statistical areas that are larger than census tracts and smaller than public use microdata areas (PUMAs).

3-A
FEDERAL AGENCY USES

Federal government agencies have historically used data from the long-form sample for a wide range of purposes. For at least the past two censuses, the Census Bureau and the U.S. Office of Management and Budget (OMB) have required that each item on the census short and long forms be justified as serving a federal agency need. For the long-form sample, each item had to be needed for federal government use for small areas, often as small as census tracts. Uses were classified into three categories: (1) mandated—that is, the use of census data was specified in legislation; (2) required—that is, data were required by legislation and, although the census was not named as the source, it was the only or the historical source of data; and (3) programmatic—that is, the census data were used for agency program planning, implementation, or evaluation or to provide legal evidence. The same general criteria are being applied with the ACS, although congressional oversight committees have indicated that it is not mandatory to pass legislation in order to add a question to the ACS.1 It should be noted that where laws or regulations specify the use of census long-form-sample estimates, changes in legislation may be required to permit the use of ACS estimates instead.

1

Personal communication, Lynda T. Carlson, Director, Division of Science Resources Statistics, National Science Foundation.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

BOX 3-1

Selected Federal Agency Uses of Census Long-Form-Sample Data

  1. The U.S. Department of Justice uses the long-form-sample data on race, Hispanic origin, educational attainment, language spoken at home, how well English is spoken, and citizenship for census tracts and American Indian areas to implement sections of the Voting Rights Act that deal with bilingual voting assistance.

  2. The U.S. Equal Employment Opportunity Commission uses the data on occupation, industry, and demographic characteristics for ZIP codes and other geographic areas to analyze statistical evidence in class action charges of employment discrimination.

  3. The OMB Statistical and Science Policy Office uses the data on place of work in relation to place of residence, together with population size and density, for counties and places to define metropolitan and micropolitan statistical areas. These areas have many public- and private-sector applications, including use in determining eligibility for some types of federal funding.

  4. The U.S. Department of Health and Human Services uses the data on older people, such as marital status, educational attainment, ancestry, disability status, income, year last worked, and housing characteristics, for counties, cities, and census tracts to measure social isolation and housing needs under the Older Americans Act.

  5. The U.S. Department of Transportation uses the data on disability, means of transportation to work, and automobile ownership for traffic analysis zones (small areas made up of one or more block groups) to monitor compliance with the Federal Transit Act and the Americans with Disabilities Act.

Types of federal agency uses of the 2000 long-form-sample data and, prospectively, of the ACS data vary widely (Citro, 2000a; National Research Council, 2004b:Ch. 2; National Research Council, 1995:Apps. C, G, H, M). Ten selected long-form-sample uses are summarized in Box 3-1; they give a flavor of the importance of these data to the operation of the federal government. The ACS should be able to serve all of these federal agency uses and more, providing more up-to-date information of higher quality than the long-form sample. Some of the issues that must be considered in using ACS estimates for federal applications are illustrated below in the discussion of two specific uses: formula fund allocation (3-A.1) and determination of income limits for housing assistance programs (3-A.2).

3-A.1
Allocation of Federal Funds

In sheer dollar terms, perhaps the most important use by federal agencies of long-form-sample data is to allocate billions of dollars of federal funds annually to states and localities (National Research Council, 2000b, 2003a; U.S. General Services Administration, 2006). Long-form-sample

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  1. The U.S. Department of Housing and Urban Development uses the data on rent and utilities, number of bedrooms, plumbing facilities, kitchen facilities, type of heating fuel, and date when the occupant moved into the unit to determine fair market rents for a base year for some metropolitan areas and nonmetropolitan counties. American Housing Survey data and telephone surveys are used to estimate base-year fair market rents for the remaining areas. Fair market rents, updated yearly from the shelter component of local consumer price indexes and telephone surveys, are used to administer rental housing subsidies and to analyze housing costs relative to household income.

  2. The Bureau of Labor Statistics uses the data on sex, age, race, Hispanic origin, labor force status, occupation, industry, and class of worker to develop state-level labor force projections, which are used by program planners, policy makers, job training administrators, and career counselors.

  3. The Federal Reserve Board uses the data on race, Hispanic origin, and the year a structure was built for census tracts to report on the record of financial institutions in meeting the credit needs of low- to moderate-income neighborhoods under the Home Mortgage Disclosure Act and Community Reinvestment Act.

  4. The U.S. Department of Veterans Affairs uses the data on veteran status and other characteristics of veterans for counties and ZIP code areas to assess changes in the veteran population and to allocate resources, such as outreach specialists and employment and training directors.

  5. The U.S. Department of Agriculture uses the data on farm acreage and sales to distribute agricultural research and extension funds to states.

data are used in two ways in allocation formulas: directly, in that long-form-sample estimates provide one or more factors in a formula, or indirectly, in that the formula relies on estimates for which long-form-sample data are one input to an estimation process that also uses other data sources.2 Whether formulas use long-form-sample estimates directly or indirectly has implications for how proactive the responsible program agency needs to be in deciding how best to use ACS estimates in place of long-form-sample estimates.

3-A.1.a.
Use of Long-Form-Sample Estimates in Fund Allocation Formulas

Most federal allocation formulas that incorporate long-form-sample data use the long-form-sample estimates directly; see Box 3-2 for seven ex-

2

Allocation formulas that use long-form-sample estimates (or estimates that incorporate long-form-sample data) may also include other factors that are often based on administrative records, such as per pupil expenditures or taxable resources.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

BOX 3-2

Selected Uses of Long-Form-Sample Estimates in Federal Fund Allocation Formulas

  1. Special Education Grants to States ($10.6 billion obligated in fiscal 2005): Allocates funds to states for the education of handicapped children in part by a formula that includes long-form-sample estimates of the number of children in the age ranges mandated by the state’s program and the number of children in poverty in those age ranges.

  2. Head Start ($6.7 billion obligated in fiscal 2005): Allocates funds to states according to long-form-sample estimates of the number of children ages 0–4 living in poor families. Organizations that operate Head Start programs use long-form-sample data as part of their applications to the U.S. Department of Health and Human Services for funding (within the limit of the funds allocated to their state).

  3. Community Development Block Grants, Entitlement Grants and State’s Program ($4.1 billion authorized in fiscal 2005): Allocates 70 percent of funds to large jurisdictions (metropolitan counties with 200,000 or more people and cities with 50,000 or more people) and 30 percent of funds to the remaining areas of states on the basis of the larger amount computed under two formulas. One formula uses long-form-sample estimates of total population, poverty population, and overcrowded housing units; the other formula uses long-form-sample estimates of total population, poverty population, and housing units built before 1940.

  4. Home Investment Partnerships Program ($1.9 billion authorized in fiscal 2005): Allocates funds to states, cities, urban counties, and consortia of local governments by a formula that uses various long-form-sample estimates, such as the estimated number of rental units built before 1950 occupied by poor families.

  5. Workforce Investment Act Adult and Youth Activities Programs ($1.9 billion obligated in fiscal 2005): Allocate funds to states, which reallocate most funds to local areas, by formulas that include long-form-sample estimates of unemployment and economic disadvantage for youths and adults.

  6. Title V Maternal and Child Health Services Block Grant to the States ($586 million obligated in fiscal 2005): Allocates funds to states as the sum of the state share of funds received for eight antecedent programs as of 1981 plus a share of any funds appropriated above the fiscal year 1983 level according to the state’s number of poor children under age 18 estimated from the long-form sample.

  7. The New Freedom Program, enacted August 2005 in the Safe, Accountable, Flexible, Transportation Equity Act: A Legacy for Users (SAFETEA-LU, P.L. 109-59): Allocated $78 million in fiscal 2006 for improved transportation services for people with disabilities. Funds are allocated to urbanized areas with 200,000 or more people (60 percent of the funds), and to states for smaller urbanized areas (20 percent) and for nonurbanized areas (20 percent). Within each group, funds are allocated to urbanized areas and states on the basis of the number of people with disabilities.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

amples. Long-form-sample estimates enter indirectly into the allocation of funds under Title I of the No Child Left Behind Act (estimated $12.7 billion obligated in fiscal 2005). This program allocates funds to school districts to meet the needs of educationally disadvantaged children by formulas that include estimates of poor school-age children. In the past these estimates were obtained from the most recent census long-form sample; currently, more up-to-date estimates are obtained from statistical models developed by the Census Bureau in its Small Area Income and Poverty Estimates (SAIPE) program.3

The SAIPE state- and county-level models include long-form-sample poverty estimates as one input together with more up-to-date information from administrative records to predict school-age poverty from a 3-year average of data from the Current Population Survey Annual Social and Economic Supplement (CPS ASEC). The school district-level model uses the previous census long-form-sample estimates of within-county school district shares of poor school-age children to apply to the updated county model estimates of the number of poor school-age children. The SAIPE program produces annual estimates with a 2-year lag between release and the estimates’ income reference year; the lag is due to delays in acquiring administrative records that are required for the modeling.

3-A.1.b
Using ACS Estimates in Formulas

Because the 2010 census will not include a long-form sample, policy makers and program managers must develop strategies for introducing ACS estimates into funding program allocation formulas that previously used long-form-sample estimates and decide whether such a change will require legislation or can be handled by regulation. The primary benefits of using ACS estimates are that they will be more timely and up-to-date and probably of higher quality than estimates from the long-form sample, so that the resulting fund allocations will more accurately reflect the distribution of needs among eligible areas.4 Still, the ACS estimates will have higher sampling error than long-form-sample estimates.


Role of Policy Makers The role that policy makers and program managers play in decisions about the use of ACS estimates in allocation formulas

3

See National Research Council, 2000a; http://www.census.gov/hhes/www/saipe/saipe.html.

4

This discussion does not address whether the variables in a formula (in the absence of data quality concerns) produce the most equitable fund distributions in light of a program’s original goals (see National Research Council, 2003a). The need to replace long-form-sample estimates with ACS estimates could trigger reconsideration of the variables and other features in a formula, but that is outside the panel’s charge.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

depends at least in part on whether the estimates will enter into a formula indirectly or directly. Indirect uses will require less in-depth consideration by program and policy people because the statistical agency that produces the relevant estimates will presumably tackle the matter. Thus, the Census Bureau SAIPE staff will presumably determine effective ways of including ACS data in their model-based estimates of poor school-age children that are used in the allocation of education funds to school districts under the No Child Left Behind Act.

The Bureau of Economic Analysis (BEA) is already incorporating ACS data into its county-level per capita income estimates, which could be considered for possible use in fund allocation. At present, only the BEA state-level per capita income estimates, which do not require 2000 long-form-sample (or ACS) data, are used in federal fund allocation programs, including the largest program—Medicaid ($193 billion of federal funds obligated in fiscal 2005)—and other programs that use the Medicaid formula.

BEA develops county (and state) per capita income estimates from federal and state administrative records, censuses and surveys, and census-based population estimates (as denominators). Currently, BEA is moving to use the ACS, in place of the 2000 long-form sample, as a source of data on intercounty commuting. This information is needed to convert estimates of per capita income by county of workplace to those by county of residence. The BEA estimates are produced annually for counties about 15 months after the end of a year.5

When ACS estimates are to replace long-form-sample estimates directly in a fund allocation formula, then program and policy people must be more involved. Factors in choosing which ACS period estimates to use (1-year, 3-year, or 5-year) include not only the extent of sampling error, but also the desired frequency with which funds are to be reallocated among areas and the types and population sizes of eligible geographic areas. Of course, during the ramp-up period between 2005 and 2010, agencies’ choices are constrained by whether the estimates that best serve their needs are available. For example, if 5-year period estimates must be used to obtain an acceptable level of precision, then agencies will need to rely on the long-form-sample estimates until 2010 when ACS 5-year period estimates become available for the period 2005–2009.


Currency, Precision, and Stability Considerations In determining which ACS estimates to use in an allocation formula (assuming they are available for all eligible areas), decision makers should identify key characteristics that the estimates must satisfy. If currency of the information is paramount, so that areas with the greatest present need receive the most funding, then

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

1-year period ACS estimates will be preferable to 3-year or 5-year period estimates, and 3-year period estimates will be preferable to 5-year period estimates. However, 1-year (or 3-year) period estimates may not be sufficiently precise—that is, may not have low enough sampling error—for fund allocation purposes. If estimates are not precise, then nontrivial changes in funding allocations from year to year may be an artifact of sampling error.6

A related consideration is the weight to give to currency for the most equitable allocations versus the practical arguments for moderating the magnitude of year-to-year changes to facilitate program planning and implementation. Many programs moderate fluctuations in program allocations through features of the formula. For example, under a hold-harmless provision, every locality is entitled to receive at least as many dollars as a specified percentage—which could be 100 percent—of its prior-year dollars.

Such legislative provisions have drawbacks, in that their use can delay the responsiveness of the funding formula to changes in need and also create inequitable allocations that are an artifact of sampling error in the estimates. For example, if legislation sets a threshold for eligibility, such as a minimum number of poor school-age children, and an area exceeds that threshold in a particular year because the estimate is greater than the threshold level due to sampling error, it will erroneously receive funding at that time. Moreover, the application of a hold-harmless provision will enable the area to retain funding in subsequent years, even though it was not eligible in the first place. An alternative approach to achieve more stable funding streams, while still responding to changes in need, is to eliminate thresholds and hold-harmless provisions and instead smooth the estimates themselves—for example, by using 3-year period estimates rather than 1-year period estimates for allocations to states (see Zaslavsky and Schirm, 2002). Implementation of this approach could require changes in legislation.


Geographic Area Considerations Yet another consideration in the selection of ACS estimates for fund allocation is the types and population sizes of geographic areas that are eligible for funding. Some formulas apply to a single type of geographic area, such as states, while others include several types of areas, such as states, cities, and counties, and still others have population size thresholds that may vary by type of area.

Consider first a formula allocation program, such as Special Education Grants to States, which uses state-level estimates of all children and poor

6

See Box 2-5 for definitions of sampling error and related terms, such as coefficient of variation and margin of error.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

children in specific age ranges, leaving it to state agencies to make further allocations to localities. In this instance, the most straightforward method for taking advantage of the ACS would simply be to substitute up-to-date ACS 1-year period estimates for the long-outdated 2000 census long-form-sample estimates in the formula. The ACS 1-year period estimates should have low sampling error for all 50 states and the District of Columbia. For example, estimates of poor school-age children may have a coefficient of variation of less than 8 percent for the smallest states, with about 600,000 people (see Table 2-7a), while the coefficient of variation of these estimates may be only 1 percent for the largest states, with 20 million people. Moreover, the Special Education Grants Program has minimum funding provisions that would moderate year-to-year fluctuations in allocation amounts from the use of annually updated ACS 1-year period estimates in place of the once-a-decade long-form-sample estimates. Should it be deemed desirable to further smooth funding amounts, the Special Education Grants Program could average 2 years of 1-year period estimates or use 3-year period estimates, which should have very low sampling error for all 50 states and the District of Columbia.

Programs like Community Development Block Grants and Home Investment Partnerships, however, provide funds to different types of governmental units, some of which are smaller in population size than the cutoff of 65,000 people or more for ACS 1-year period estimates. For these programs, it will not be possible to take the simple approach outlined above because ACS 1-year period estimates will not be available for all eligible areas. Moreover, while ACS 3-year period estimates may be available for all eligible areas, they may not be sufficiently precise for some of them. For example, should the needed estimates represent a group as small as poor school-age children, then the 3-year period estimates will not have a reasonably small coefficient of variation until the eligible area has a population of at least 80,000 people (see Table 2-7a).7

For such programs as Community Development Block Grants, for which governmental units as small as 50,000 people are eligible for funding, agencies must carefully balance the need for more up-to-date information from using 3-year period estimates against precision requirements that will be better satisfied with 5-year period estimates. For programs for which governmental units must have at least 100,000 people to be eligible for

7

Table 2-7a should be used only as a very rough guide to expected levels of sampling error for estimates for different size areas from ACS 1-year, 3-year, and 5-year period estimates. The sampling error will differ from that shown in the table for a characteristic that is a different percentage of the population from poor school-age children (as seen in Table 2-8). The sampling error will also depend on the sample size that the ACS achieves in the field for the particular governmental unit.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

funding, agencies must trade off the timeliness of 1-year period estimates and the greater precision of 3-year period estimates.

When agencies decide that there is no choice but to use 5-year period estimates from the ACS in a funding formula in order to gain sufficient precision, they should be aware that inequities may result. For example, two areas may have the same 5-year period poverty rate and therefore receive the same allocation, even though one area may have a sharply increasing poverty rate and the other area a sharply decreasing poverty rate over the period. Even in this case, however, the use of ACS 5-year period estimates would represent an improvement over the continued use of the increasingly out-of-date 2000 long-form-sample estimates.

At present, the only federal funding program that makes allocations to areas with fewer than 50,000 people is the No Child Left Behind Act, which allocates funds to school districts, varying in size from a few hundred to several million people (see Table 2-4). The SAIPE estimates that are used for the allocations are more up to date than the direct long-form-sample estimates later in the decade, but they rely on statistical models. The incorporation of ACS data into the SAIPE county and school district models should make it possible to improve their timeliness and precision.


Consistency of Period Estimates In trading off such considerations as currency and precision, in no instance should agencies use in their allocation formulas a mix of different periods of ACS estimates—for example, 1-year (or 2-year) period estimates for larger areas and 3-year or 5-year period estimates for smaller areas—in an attempt to equalize the sampling error across areas. The reason has to do with equity: formulas generally allocate shares of a fixed pie, so that the data used in the allocation should reference the same time period. Otherwise, inequitable outcomes may occur. For example, consider a large county and a medium-sized city, both of which experience rapidly increasing poverty over 5 years. If in a poverty-based formula, 1-year period estimates are used for the large county and 3-year period estimates are used for the medium-sized city, then the county will likely receive more than its fair share of funds over the 5 years compared with the city because the 1-year period estimates will likely exhibit more growth in poverty than the 3-year period estimates.

3-A.2
Determination of Median Incomes for Counties

The U.S. Department of Housing and Urban Development (HUD) obligates $27 billion annually for assisted housing programs in which families that have incomes below specified limits are eligible to live in public housing or receive rent subsidies. The income limits are determined separately for every metropolitan area and nonmetropolitan county as a function of

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

median income. Historically, HUD has used census long-form-sample median family income estimates, adjusted at the national level to agree with estimates from the CPS ASEC for the census income year, as the starting point to develop current fiscal year estimates for each area. To update the long-form-sample estimates, HUD uses the most recent Bureau of Labor Statistics (BLS) wage and salary data for counties, adjusted to match median income estimates for the nine census geographic divisions from the most recent CPS. As a final step, HUD projects the median family income estimates to the middle of the fiscal year for which the agency is setting housing assistance income limits.

The advent of the ACS means that HUD will no longer need to update long-form-sample median family income estimates from data sources, such as BLS person-level wage data, that do not reflect the same concept of total family income. The use of ACS county-level median family income estimates to determine area-specific eligibility for subsidized housing, however, raises at least three important issues: (1) whether achieving comparable levels of precision across areas is preferable to using the same periodicity of ACS estimates (1-year, 3-year, or 5-year) for all areas; (2) the possible effects on the accuracy of ACS income estimates from the moving reference period (respondents are asked about the prior 12 months rather than a consistent prior calendar year); and (3) the possible effects on the accuracy of ACS income estimates from the Census Bureau’s procedure for adjusting income amounts for inflation. (See ORC Macro, 2002:162–171, for a fuller discussion of these and other issues.)

3-A.2.a
Period Estimates for 1, 3, or 5 Years?

HUD requires median family income estimates each year for all 3,000-plus counties in the United States. One-year period estimates of median family income will probably be reasonably precise for counties with at least 50,000 people, and 3-year period estimates will probably be reasonably precise for counties with at least 20,000 people. (Estimates of median family income are about twice as precise and therefore have only about half the coefficient of variation of estimates of poor school-age children—see Table 2-7a.) However, 1-year period estimates will be available only for counties (and other governmental and statistical areas) with at least 65,000 people, yet three-fourths of counties are smaller than that. Moreover, two-fifths of counties have fewer than 20,000 people so that 5-year period estimates will be the only available source for about 1,300 counties (see Tables 2-4 and 2-5).

A study conducted for HUD by ORC Macro (2002:169) suggested that HUD might want to use 1-year period ACS median family income estimates for counties with 200,000 or more people, 3-year period estimates

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

(when they become available) for counties with 65,000 to 200,000 people, and 5-year period estimates (when they become available) for the remaining three-fourths of counties. This strategy is conservative with regard to sampling error. A reason to be conservative is that HUD is concerned not only with having estimates that are as up-to-date as possible, but also with reducing year-to-year fluctuations in median family income estimates that are due to sampling error.

The previous discussion of using ACS estimates in fund allocation formulas concluded that estimates for different periods should not be used in the same formula because the resulting fund allocations could be inequitable. The HUD use of median income estimates, however, is different in that HUD is not allocating shares of a fixed budget allotment; instead, it is determining an eligibility threshold for an entitlement. Families living in a metropolitan area or a nonmetropolitan county that have incomes below a specified percentage of the median income for that area are entitled to subsidized housing, and the median income levels in other areas are not relevant to this determination. (In practice, entitled families may be put on a waiting list because not enough housing is available.) Given that housing assistance is allocated to individual families on the basis of their incomes as a percentage of the median for their area, it makes sense to use the estimate for each metropolitan area or nonmetropolitan county that has an acceptable level of sampling error and is as up-to-date as possible.

3-A.2.b
Moving Reference Periods

Because the ACS is conducted on a continuous monthly basis, the questionnaire items change in their reference period across the year. Many questions (see Table 2-2) refer to the time when the respondent fills out the questionnaire, which could be any date from January to December of a calendar year. Questions on income ask for amounts received in the 12 months prior to when the respondent fills out the questionnaire. Consequently, the ACS 1-year period income estimates will include reference periods that span a full 23 months: for 2005 income estimates, for example, the reference periods range from January–December 2004 for people who responded in January 2005 to December 2004–November 2005 for people who responded in December 2005.

There has been little research on the effects on accuracy of reporting income amounts with a moving reference period of the past 12 months compared with the fixed reference period of the previous calendar year that is used in the long-form sample and the CPS ASEC. A split-sample experiment with mail responses to the ACS questionnaire in October–December 1997 produced the unexpected result of no significant differences in median total income of individuals who were asked to report for the preceding cal-

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

endar year (January–December 1996) and those who were asked to report for the past 12 months, covering October 1996–September 1997, November 1996–October 1997, and December 1996–November 1997 (Posey and Welniak, 1998). What factors explain this result—for example, whether respondents tend to annualize their current income or to report income for the previous calendar year regardless of the reporting period—are not known. Carefully designed research will be needed to assess the effects of the ACS reference period on income statistics, such as research that compares an external measure of income from administrative records with survey responses for the same individuals.

3-A.2.c
Inflation Adjustments

For completeness, this section discusses inflation adjustments not only for income, but also for housing amounts. The latter amounts include housing value, monthly contract rent, monthly gross rent (contract rent plus utilities), and monthly selected owners’ housing costs (mortgage payments, utilities, taxes, property insurance).8


Income To put income amounts that are reported for differing 12-month reference periods on a comparable calendar-year basis, the Census Bureau expresses them in constant dollar terms by using the national consumer price index for urban consumers-research series (CPI-U-RS) for the latest calendar year covered by an estimate.9 For 1-year period income estimates for 2005, for example, each reported amount on a person record is adjusted by the ratio of the annual average CPI for 2005 divided by the average of the monthly CPIs for the particular 12-month reporting period for that person. For 3-year period estimates for, say, 2005–2007, the incomes for people sampled in 2005 and 2006 (which have already been adjusted to calendar 2005 or 2006 on a 1-year basis) are adjusted to calendar year 2007 by the ratio of the annual average CPI for 2007 divided by the annual average CPI for 2005 or 2006, as the case may be (see Table 3-1 for how this adjustment is carried out).

8

To create monthly gross rent and selected owners’ housing costs, the amounts reported for some costs for either the prior 12 months or as “annual” amounts—see Table 2-2—are converted to monthly amounts.

9

“The Bureau of Labor Statistics (BLS) has made numerous improvements to the Consumer Price Index (CPI) over the past quarter-century … [but] historical price index series are not adjusted to reflect the improvements. Many researchers … expressed an interest in having a historical series that was measured consistently over the entire period. Accordingly, the Consumer Price Index research series using current methods (CPI-U-RS) presents an estimate of the CPI for all Urban Consumers (CPI-U) from 1978 to present that incorporates most of the improvements made over that time span into the entire series” (http://www.bls.gov/cpi/cpiurstx.htm).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-1 Hypothetical Inflation Adjustments for Person Income in the ACS

Data: Consumer Price Index for All Urban Consumers (CPI-U), rounded (1983–1984 = 100)

Year

Jan.

Feb.

Mar.

Apr.

May

June

July

Aug.

Sept.

Oct.

Nov.

Dec.

Average

2004

185

186

187

187

189

190

189

190

190

191

191

190

188.9

2005

191

192

193

195

194

195

195

196

199

199

198

197

195.3

2006

198

199

200

201

202

203

204

205

206

207

208

209

203.5

2007

210

211

211

212

213

214

215

216

216

217

218

219

214.5

(1) Adjustment Factors (x) for 2005 ACS 1-Year Period Person Income (x is applied to prior 12 months’ reported income)

January 2005 sample persons (income reported for 01/04–12/04)

 

x = [195.3/((185 + … + 190)/12)] = 1.034

February 2005 sample persons (income reported for 02/04–01/05)

 

x = [195.3/((186 + … + 191)/12)] = 1.032

.

 

 

 

 

.

 

.

 

 

 

 

 

 

.

 

 

 

 

.

 

.

 

 

 

 

 

 

.

 

 

 

 

.

 

.

 

 

 

 

 

 

November 2005 sample persons (income reported for 11/04–10/05)

 

x = [195.3/((191 + … + 199)/12)] = 1.006

December 2005 sample persons (income reported for 12/04–11/05)

 

x = [195.3/((190 + … + 198)/12)] = 1.003

(2) Adjustment Factors (x) for 2006 ACS 1-Year Period Person Income

For each monthly sample as in (1) x = [203.5/(average of factors for previous 12 months)]

(3) Adjustment Factors (x) for 2007 ACS 1-Year Period Person Income

For each monthly sample as in (1) x = [214.5/(average of factors for previous 12 months)]

(4) Adjustment Factors for 2005–2007 ACS 3-Year Period Person Income

a. For 2005 sample persons

x = 214.5/195.3 = 1.09, x is multiplied by the adjusted 2005 income (1)

b. For 2006 sample persons

x = 214.5/203.5 = 1.05, x is multiplied by the adjusted 2006 income (2)

c. For 2007 sample persons

x = 214.5/214.5 = 1.00, x is multiplied by the adjusted 2007 income (3)

(5) 2005–2007 ACS 3-Year Period Person Income Estimates for All Persons

Calculated as ( 4.a + 4.b + 4.c)/3

SOURCE: See http://www.bls.gov for monthly CPIs through February 2006; other months are hypothetical.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

This adjustment expresses all of the reported income amounts for a given period (1 year, 3 year, or 5 year) in a comparable manner with regard to purchasing power as of the most recent calendar year in the period. Such an adjustment should not be confused with a current-year estimate. For example, an inflation-adjusted 5-year period median income estimate covering years 2005–2009 is not an estimate of median income for the latest year (2009); instead, it is an estimate of the median of all of the reported income amounts over the 5 years expressed in 2009 dollars.

It is possible that for many applications users may prefer an inflation adjustment to the most recent calendar year to no adjustment at all. For some applications, users may find that an inflation adjustment to the latest year is not adequate. For example, users frequently wish to compare ACS income estimates with those from other household surveys. Yet a 1-year period income estimate from, say, the 2005 ACS that expresses income amounts in constant 2005 dollars for reference periods spanning January 2004 through November 2005 is not comparable to an estimate from a survey, such as the 2006 CPS ASEC, that collects all income amounts for the same fixed reference period of calendar 2005. The reasons are that prices are not income, and incomes may grow faster (or slower) than prices.

Turek, Denmead, and James (2005) illustrate the problems of using price change as a proxy for income change when comparing survey estimates. For 1998—a period of strong economic growth—they estimated that the Census Bureau’s inflation adjustment would make up only 22 percent of the difference between average person total income from a simulated 1998 ACS sample compared with average person total income reported for calendar year 1998. The simulations used the Survey of Income and Program Participation, which collects income on a 1-month or 4-month basis over a multiyear period. The analysis compared income amounts reported by people for the 12 months preceding each month in 1998 unadjusted for inflation (average $17,304 person total income), the same income amounts adjusted for inflation to calendar 1998 (average $17,447), and income amounts reported by the same people for all 12 months of 1998 (average ($17,945). Presumably, the difference between the second and third figures occurs because, on average, people received pay raises or returns on assets between their income reporting period and the end of the calendar year that exceeded the rate of inflation (for example, a big raise in June 1998 for an individual who reported income for June 1997–May 1998).

Many applications, such as HUD’s use of county-level median income to determine eligibility for housing assistance programs, require current-year estimates. The ACS inflation-adjusted period estimates will not be optimal for such applications, given that they represent averages over the period expressed in dollars for the latest year in the period instead of estimates for the latest year. The inability of the inflation adjustments to represent

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

latest-year income amounts is likely to be much more pronounced for the ACS 3-year and 5-year period estimates compared with the 1-year period estimates. Yet for the county-level estimates of median income required by HUD, only one-fourth of counties will have 1-year period estimates available, and over 40 percent of counties will have only 5-year period estimates available.

Even when users find inflation-adjusted period income estimates to be reasonably satisfactory for an application, they may prefer adjustments that reflect variations in price changes, such as the use of different price indexes for different geographic areas. However, only limited data are available for this purpose (see Section 4-D.3).

Finally, in the special case of poverty estimates, the Census Bureau’s method for determining poverty status for families and their members does not require adjusting income amounts for inflation. This situation arises because the Census Bureau compares the income of a family (or unrelated individual) for a 12-month reporting period, not adjusted for inflation, to 12-month nominal dollar thresholds by family size and type for that same period. These thresholds are derived from a base-year threshold (1982) using the national CPI, as is done in the official poverty measure. The only difference from the official measure, which uses calendar-year thresholds, is that the threshold for each family is the average of the CPI-adjusted monthly thresholds for that family’s 12-month income reporting period. For a 5-year period estimate, then, the poverty rate is the average rate of everyone in the sample over the 5 years.


Housing For housing amounts, such as value, rent, utilities, property taxes, and others, the Census Bureau makes no inflation adjustments for the 1-year period estimates. When, however, the 1-year period estimates for housing amounts are cumulated over 3 or 5 years, the Census Bureau adjusts them for inflation by using the ratio of the annual average CPI value for the latest year of the period to the annual average CPI value for the year for which the amounts were reported.

The issues that can affect uses of the inflation-adjusted income amounts can also affect the inflation-adjusted housing amounts. The ACS 3-year and 5-year period estimates for rent, housing value, utilities, and other housing amounts expressed in dollars for the latest year are not the same as estimates for the latest year. Moreover, increases (or decreases) in housing amounts often differ across areas and by item—for example, housing values in recent years have increased much more than many other items in the national CPI and have increased much more in some areas than others.

Section 4-D.3 discusses several issues involved in adjusting ACS period estimates of income and housing amounts for inflation. A key question that needs to be resolved by discussion among users and the Census Bureau is

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

the purpose of the adjustment. Assuming that users largely prefer an inflation adjustment to the latest year of the period, then the question becomes one of the specific method(s) to use. Another issue is how to assist users who require current-year estimates rather than averages expressed in latest-year dollars.

3-A.2.d
Alternative Sources of Median Income Estimates

Abstracting from the previous discussion, there are at least three possible approaches for HUD (and other users) to obtain median family income estimates for the previous calendar year that are reasonably precise for all counties:

  1. As suggested by ORC Macro (2002:162–171), HUD could ask the Census Bureau to produce 1-year period ACS median family income estimates for combinations of small counties to accompany the estimates that are published for larger counties. (If PUMA combinations of counties are suitable, then HUD could use the 1-year period estimates that will be regularly produced for PUMAs.)

  2. HUD could plan to use the SAIPE model-based median household income estimates for all counties once the model is modified to incorporate information from the ACS. A drawback of the SAIPE estimates is the 2-year lag between release and the calendar year reference period of the estimates. Also, at present the SAIPE estimates represent a 3-year average, but this may change if the ACS is used as the dependent variable in the model equations in place of the CPS ASEC. An advantage of model-based estimates is that they exhibit less variability in precision across areas than direct estimates (see Bell, 2006, for comparisons for states).

  3. HUD could decide to use ACS 3-year or 5-year period estimates for counties and ask the Census Bureau to develop an alternative method for adjusting income responses in the ACS to reflect HUD’s need for current-year estimates. For example, appropriate year-to-year ratios of the BLS wage data for counties could be applied to the ACS 3-year or 5-year household income estimates, not adjusted for inflation, to produce current-year median income estimates.

3-B
STATE AGENCY USES

State governments have many uses of census long-form-sample data for program planning, implementation, and evaluation that are similar to those of federal government agencies (see Section A above). They also have many

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

uses that are similar to those of local government agencies (see Section 3-C below). Because of these similarities, the only major use by states that we explore in detail here is allocation of state funds to localities in Section 3-B.1. Strategies for using ACS data instate fund allocations are considered in Section 3-B.2.

It is worth noting that many state uses are to respond to requirements of the federal government. For example, HUD requires states and localities to have a Comprehensive Housing Affordability Strategy. This plan includes an assessment of the housing needs of families residing in a jurisdiction that is developed, in part, from long-form-sample data on demographic and housing unit characteristics for individual census tracts in the area. Such applications in the ACS context will require use of the 5-year period estimates for census tracts, which will likely need to be aggregated into larger areas to obtain sufficient precision.

3-B.1
Allocating State Funds to Localities

Under many federal fund allocation programs, states are responsible for distributing most or all of their funds to localities by using long-form-sample data. Many states also allocate their own funds by means of formulas to local jurisdictions, such as counties and school districts (see examples in National Research Council, 2000b:Table 2-1). The most used sources of data for state funding formulas are estimates from the previous long-form sample and state administrative records, such as school lunch data and income tax records.

The problems with long-form-sample estimates, as noted throughout this report, include that they are not timely, that they become increasingly out of date over a decade, and that they suffer from high levels of item nonresponse because long-form data collection takes a back seat to completing the basic census count. The long-form-sample estimates also have large sampling errors for small areas.

Administrative records have problems as well. They may not correspond that closely to the target population for a program—for example, school lunch data, which are often used in state formulas to target funds to school districts with poor children, may not closely track the poverty population because children in families with incomes as high as 185 percent of the poverty threshold are eligible for reduced-price lunches. In addition, program participation may be affected by such factors as outreach activities that operate more strongly in some areas than others. To the extent that this is true, the use of administrative data on school lunch or food stamp participants as a proxy for the poverty population may not give consistent estimates across areas (see National Research Council, 2000a:App. D).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-B.2
Strategies for Using ACS Data in State Fund Allocations

States should consider the use of ACS estimates in place of the data sources they currently use for allocating state funds to localities. The same considerations apply as discussed for federal fund allocation, such as the value placed on having the most up-to-date estimates in contrast to the stability of funding streams and the types and population sizes of eligible geographic areas. It is most likely that states would need to use ACS 5-year period estimates to allocate funds to local areas given that 1-year and even 3-year period estimates are not available or not sufficiently precise for many jurisdictions.

There may be instances in which a state believes it is important that fund allocations (or another application) reflect data that are as current as possible and when reasonably precise 1-year (or 3-year) period estimates are available for many but not all eligible jurisdictions. Should a state find itself in this situation, it could consider using a simple procedure to update the 5-year period estimates for jurisdictions for which they are the only reasonably precise estimates available (see Section B.3 below). The intent would be to put the 5-year period estimates on a comparable basis with 1-year (or 3-year) period estimates and not have to discard the more up-to-date estimates for those jurisdictions for which they are available and sufficiently precise for their intended use. Federal agencies may also be able to use this procedure for some applications.

For other applications, the goal may be currency of estimates, but there may be reason to believe that a simple updating procedure will not give good results because its underlying assumptions about change over time among areas are unrealistic. In such instances, a more advanced form of small-area estimation is called for. Such estimation requires additional data from administrative records or other sources, similarly to the way that the Census Bureau’s SAIPE program uses food stamp and federal income tax data to generate updated county estimates of poor school-age children.

Before deciding to use any type of updating procedure, simple or complex, it is essential to carefully examine the procedure’s underlying assumptions. It may be that less current estimates are preferable to more current estimates produced with an unrealistic updating procedure.

3-B.3
Example of a Simple Updating Procedure

Table 3-2 provides an example of a simple procedure to produce current county-level estimates of poor school-age children for possible use in allocating state funds to counties. The state in this example plans to use ACS 1-year period estimates for as many counties as practicable and to adjust ACS 3-year or 5-year period estimates for the remaining counties to

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-2 Example of Simple Method to Update ACS 5-Year Period Estimates for 2010–2014 to Latest Year (2014), Four Small Counties (A, B, C, D) in State X, Using Data for Two Public Use Microdata Areas (PUMAs)

 

PUMA

County

 

PUMA

County

 

1

 

AB

2

 

CD

Total Population (20% are school-age children)

100,000

50,000

50,000

100,000

50,000

50,000

Estimated Number of Poor School-Age Children

 

 

 

 

 

 

1. 5-year period ACS estimate, 2010–2014

4,000

1,500

2,500

2,000

1,000

1,000

2. 1-year period ACS estimate, 2014

5,000

(not available)

2,100

(not available)

Change in School-Age Poverty

 

 

 

 

 

 

3. Ratio of 2014 PUMA estimate to 2010–2014 PUMA estimate (line 2/line 1)

1.25

(not applicable)

1.05

(not applicable)

Estimated Number of Poor School-Age Children, 2014

 

 

 

 

 

 

4. For PUMAs: ACS 1-year period estimate (line 2) For counties: Simple method, using county ACS 5-year Period estimate and PUMA change ratio (line 1 × line 3)

5,000

1,875

3,125

2,100

1,050

1,050

How well does the simple method to update a 5-year average estimate of poor school-age children to the latest year work?

  • Assume that the actual number of poor school-age children for the four counties in 2014 is 2,100 for County A, 2,900 for County B, 800 for County C, and 1,300 for County D.

  • For Counties A and B in PUMA 1, which both experienced an increase in school-age poor children from the average 5-year estimate to the latest year (1,500 to 2,100 and 2,500 to 2,900, respectively), the simple updating method makes their 5-year period estimates more current.

  • For Counties C and D in PUMA 2, the simple method is less satisfactory. Because County C bucked the overall trend and had a decrease in school-age poor children (from 1,000 to 800), the PUMA 2 change ratio between the 2014 estimate and the 2010–2014 estimate is very small. Consequently, the simple updating method does not capture either the substantial decrease in school-age poor children in County C or the substantial increase in school-age poor children (1,000 to 1,300) in County D.

NOTE: See text on the need to understand and evaluate the assumptions that underlie any modeling procedure, even the simplest, before using a particular procedure to update ACS 5-year (or 3-year) period estimates to 1-year period estimates. The method illustrated assumes that the numbers of poor school-age children grew at the same rate for each county in a PUMA, or, alternatively, that each county’s share of poor school-age children in a PUMA remained the same over time.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

represent the latest year of the period. Say that the state has 1 million total people resident in eight counties: four counties are large, each with 200,000 people, and four counties are small, each with 50,000 people. The four smaller counties make up two PUMAs. (In actuality, two-thirds of counties are smaller than this.)

The state could use 1-year period ACS estimates of poor school-age children for the four large counties directly in the allocations. Also available would be 1-year period estimates for the two PUMAs, individually and combined, which could be used to adjust 5-year period estimates for the four smaller counties to the same 1-year period. The simple updating procedure would apply the ratio for the PUMAs of the 1-year and the 5-year period estimates of poor school-age children to the 5-year period estimates for each county component.

Using separate ratios for the two PUMAs (as in Table 3-2) would better capture differences among the smaller counties than would using a single ratio for the two PUMAs combined, but the combined ratio would be more precise than the two separate ratios. Even using separate ratios, the updated estimates for the counties in PUMA 2 are not as realistic as those for the counties in PUMA 2 because one county in PUMA 2 experienced a decrease in school-age poverty and not an increase as in the other three counties.

The simple procedure works best when it only has to be used—and, therefore, its assumptions only have to be invoked—for a small fraction of the total number of jurisdictions. Because only about half a dozen states have ACS 1-year (or even 3-year) period estimates available for most counties, the procedure may not be widely useful when the goal is to adjust 5-year period estimates for smaller counties to the latest year.

The Census Bureau’s SAIPE program currently uses this type of simple procedure to produce updated estimates of poor school-age children for school districts within counties. In that application, good administrative data are available with which to update the county estimates, but the updating procedure for school districts has to assume that the within-county proportions of poor school-age children for school districts are the same for the estimation year as they are for the previous long-form-sample year. Work is under way that shows promise of improving the currency of SAIPE school district estimates of poor school-age children by using IRS personal income tax data coded to the block level (Maples, 2004). The ACS estimates for school districts may also be helpful in the SAIPE school district-level model.

3-C
LOCAL GOVERNMENT USES

Local governments—counties, cities, towns, townships, school districts, and areas governed by Alaska Native or American Indian tribes—will likely be major users of the ACS, particularly local governments with sizeable

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

populations. To date, local governments have been limited to once-every-10-year updates of socioeconomic characteristics for their area from the decennial long-form sample. The Census Bureau provides updated estimates of total population throughout the decade for counties and places (the updates include age, sex, and race/ethnicity detail for counties), as well as updated estimates of school-age poverty for counties and school districts. In addition, many local governments have their own sources of data from administrative records and, in some cases, their own surveys. However, most jurisdictions rely heavily on the detailed socioeconomic information in the long-form sample for a myriad of applications involving program planning, allocation of resources, location of service facilities (for example, health clinics, police stations, schools), preparation of supporting material to accompany requests for state and federal aid, and understanding of important trends for their jurisdiction in terms of economic growth or decline, changing age, race, and ethnic composition of the population, and the like.

Illustrative applications in which ACS estimates are used in place of long-form-sample estimates are discussed below for large cities (3-C.1), small jurisdictions in a rural state (3-C.2), and jurisdictions with large seasonal populations (3-C.3). These examples highlight some of the important considerations that local governments need to take into account when they begin to work with ACS estimates. They also illustrate that large areas will benefit greatly from the ACS, while areas with fewer than 50,000 people will confront a mixed situation: on the positive side, the ACS estimates will be more current than the long-form sample estimates; on the negative side, the ACS estimates will be imprecise for estimates of many population groups—more imprecise than the long-form-sample estimates.

3-C.1
Large City Applications of the ACS

This section considers strategies for large cities to work with multiple ACS estimates (1-year, 3-year, 5-year) and analyze change over time. It also provides a case study that illustrates how useful ACS estimates can be for subcity-area analyses.

3-C.1.a
Working with Multiple Estimates

Large cities, considered as those with at least 250,000 people (for which 1-year period ACS estimates for small population groups should meet common standards of precision—see Table 2-7a), can benefit from the full set of ACS 1-year, 3-year, and 5-year period estimates. (Such cities are referred to as BIG CITY throughout the text and examples.) The challenge is how to make the most effective use of the various period estimates to understand citywide trends and, at the same time, assess varying neighborhood conditions that are important for program planning and implementation.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

Both 1-year and 3-year period estimates will be available not only for BIG CITY as a whole, but also for its PUMAs, which are defined to have at least 100,000 people. Five-year period estimates will be available for BIG CITY, its PUMAs, and small neighborhood areas—namely, census tracts, which average about 4,000 people, and block groups, which average about 1,500 people. (Cities—in contrast to counties—do not contain separate independent governments, such as towns and school districts, so there are no subcity areas with populations between 20,000 and 100,000 for which 3-year estimates will be provided under current plans.) The 5-year period estimates for census tracts and block groups will be extremely imprecise for many population groups of interest because these areas are so small in population size.10 Hence, users must combine groups of tracts or block groups into larger areas—such as health service areas, school attendance areas, planning districts, and the like—for which 5-year period estimates will be reasonably precise.

Given that 5-year period estimates must be used for subcity areas, there is the issue of which set of estimates to use for BIG CITY as a whole for comparative analysis. In the presence of economic growth (or decline), in-migration (or out-migration) of various population groups, and other social and economic changes, a city’s 5-year period estimates may differ appreciably from its 3-year period estimates, and even more so from its 1-year period estimates. Moreover, some neighborhoods may lag behind or be ahead of the overall city trend (see Table 3-3 for an example).

Which estimate to use for BIG CITY will depend on the application, but many users will want to minimize the confusion caused by using estimates for different periods in any given analysis. One strategy is to use the 1-year period estimates for public and media consumption regarding citywide trends (see Section 3-F). The 5-year period estimates for BIG CITY and user-defined subcity areas would be reserved for detailed analyses that are released at a later time and used primarily by the city’s own staff for planning and related purposes.

Another strategy is to request special tabulations from the Census Bureau of 1-year or, more likely, 3-year period estimates for user-defined subcity areas that meet the Census Bureau’s population thresholds of at least 65,000 people for 1-year period estimates and at least 20,000 people for 3-year period estimates. Cities should give early attention to their possible need for such custom estimates and work with the Census Bureau

10

Research on sampling error by the Census Bureau (Starsinic, 2005) found that ACS estimates for census tracts exhibit much more error compared with long-form-sample estimates than is the case for ACS county estimates compared with long-form-sample estimates. A likely explanation is that census tract estimates, in contrast to county-level estimates, are not adjusted to housing unit or population controls.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-3 School-Age Poverty Rates for BIG CITY/COUNTY and Three Subareas, Illustrative ACS 1-Year, 3-Year, and 5-Year Period Estimates for 2010–2014

 

BIG CITY/COUNTY 250,000 people

Area A* 100,000

B* 65,000

C* 85,000

 

ACS 1-Year Period Estimates of Percent Poor School-Age Children

2010

15.0

10.0

22.0

13.0

2011

16.0

11.0

25.0

12.0

2012

17.0

12.0

28.0

11.0

2013

18.0

12.0

31.0

11.0

2014

20.0

16.0

35.0

9.0

 

ACS 3-Year Period Estimates of Percent Poor School-Age Children

2010–2012

16.0

11.0

25.0

12.0

2011–2013

17.0

11.7

28.0

11.3

2012–2014

18.3

13.3

31.3

10.3

 

ACS 5-Year Period Estimates of Percent Poor School-Age Children

2010–2014

17.2

12.2

28.2

11.2

How do the 1-year, 3-year, and 5-year period estimates compare with each other?

  • School-age poverty increased in BIG CITY/COUNTY, so the 5-year period estimate (2010–2014) of 17.2 percent is lower than the latest 3-year period estimate (2012–2014) of 18.3 percent, which, in turn, is lower than the latest 1-year period estimate (2014) of 20 percent.

  • The same pattern is evident for Areas A and B.

  • School-age poverty decreased for Area C, so the 5-year period estimate (2010–2014) of 11.2 percent is higher than the latest 3-year period estimate (2012–2014) of 10.3 percent, which, in turn, is higher than the latest 1-year period estimate (2014) of 9 percent.

How do the 3-year and 5-year period estimates compare with continuing to use a 2010 census estimate (if 2010 included a long-form sample and provided estimates equal to those shown for the ACS for 2010)?

  • The latest 5-year period estimate more accurately depicts current school-age poverty than would continuing to use a 2010 census estimate.

  • The latest 3-year period estimate even more accurately depicts current school-age poverty than would continuing to use a 2010 census estimate.

*Availability constrains the choice of estimates:

  • In BIG CITY, 1-year and 3-year period estimates will only be available for the city as a whole and for PUMAs with at least 100,000 people, so the 1-year and 3-year period estimates shown for Areas B and C will not be available.

  • In BIG COUNTY, 1-year and 3-year period estimates will be available for the county as whole, PUMAS, and any places or towns with 65,000 or more people; in addition, 3-year period estimates will be available for governmental jurisdictions with at least 20,000 people, but large sampling errors will limit their usefulness for comparisons among areas and over time.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

to develop specifications for the subcity areas and the table content. The subcity areas must be large enough in population size and the table content must not be too detailed if 1-year, or even 3-year, period estimates are to be reasonably precise. (See Section 4-D.4 for a recommendation that the Census Bureau consider producing 3-year and even 1-year period estimates for areas smaller than PUMAs in large cities.)

3-C.1.b
Analyzing Change over Time

In addition to comparative analyses among subcity areas, users will likely want to analyze trends over time for BIG CITY as a whole and for its subareas. The sampling errors for estimates of differences are always larger than the sampling errors for the individual estimates that are being compared. Consequently, users should anticipate that estimates of year-to-year differences will often be very imprecise and should take care to avoid playing up differences that may appear important in policy terms but are, in fact, within the margin of error. In addition, for analyses of year-to-year differences that must use 3-year or 5-year period estimates and not 1-year estimates, there is the problem of how to interpret the results. Yet an investment in learning how to work with multiple years of ACS estimates, which may require seeking statistical advice, should benefit users who want to exploit the continuous availability of updated information for time trend analysis.

The following text highlights selected aspects of using the ACS to measure change over time. Chapter 6 has a technical discussion of measuring change with ACS period estimates and the implications of alternative approaches for the precision and usefulness of the resulting estimates.


Using 1-Year Period Estimates to Estimate Change for BIG CITY as a Whole Consider two consecutive 1-year period ACS estimates of poor school-age children for BIG CITY (which is assumed to have 50,000 school-age children in a total population of 250,000)—for example, 17 percent poor school-age children in 2010 and 19 percent poor school-age children in 2011. An increase of this magnitude for the nation would be a significant change, both statistically and substantively—over 1 million more children would be poor, and the estimate of change would be very precise. However, the increase for BIG CITY in this example is only 1,000 more poor children, and the estimate of change would not be precise: the 90 percent margin of error for the estimated 2 percentage point increase in school-age poverty would likely be about plus or minus 4.6 percentage points compared with about plus or minus 3.2 percentage points for each year’s

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

individual estimate of percentage poor school-age children.11 This means that, on the basis of two successive 1-year period ACS estimates for BIG CITY, the school-age poverty rate may have increased by as much as 6.6 percentage points or decreased by as much as 2 percentage points, or it may have stayed the same. Users cannot conclude whether a change has occurred because the estimates are not precise enough to indicate what has happened.

Only if BIG CITY experiences a large real change is the estimate of the difference between two successive 1-year period ACS estimates likely to be statistically significant. Yet BIG CITY will benefit greatly once a time series of 1-year period ACS estimates is available, because the patterns of yearly change will be informative regarding the existence (or not) of a trend in such characteristics as the percentage of poor school-age children. Consider an example in which BIG CITY had an estimated 15 percent school-age poverty rate in both 2000 and 2010, but poverty increased from 15 to 22 percent in 2005 and then decreased to 15 percent in 2010. Two consecutive long-form-sample estimates for BIG CITY, while quite precise, would completely miss the intercensal dynamics of school-age poverty, whereas a time series of 1-year ACS estimates for the city could track the intercensal trends, even though the year-to-year estimates of change were not precise.

Table 3-4 illustrates changes in school-age poverty rates for BIG CITY (population 250,000) and VERY BIG CITY (population 1 million) over the period 2010 to 2014. The year-to-year differences in school-age poverty rates for BIG CITY (Part A, line 2) are not statistically significant, even though the example purposefully accelerates the increase in school-age poverty over the time period (from a 1 percentage point difference between 2010 and 2011 to a 3 percentage point difference between 2013 and 2014). There is only one significant year-to-year difference for VERY BIG CITY (Part B, line 2), which is the 3 percentage point difference between 2013 and 2014.

As 1-year period estimates accumulate, however, the differences from the first year—2010—are significant by 2014 for BIG CITY (Part A, line 3) and by 2013 for VERY BIG CITY (Part B, line 3). The reason is that the size of the differences between the estimation year and 2010 increases over time (from 1 percentage point between 2010 and 2001 to 2 percentage points between 2010 and 2012, 4 percentage points between 2010 and 2013, and 7 percentage points between 2010 and 2014). It could also be

11

When two estimates are approximately independent, as is the case for two ACS 1-year period estimates (for which, the samples do not overlap), the standard error of an estimate of change is the square root of the sum of the squared standard errors for the two individual estimates. In the example in the text, the standard errors for each year of about 1.91 (2010) and 1.99 (2011) are squared, summed, and the square root taken to give a standard error of the estimate of change of about 2.76. Times 1.65, the 90 percent margin of error is plus or minus 4.55.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-4 Analyzing Trends Over Time for School-Age Poverty Rates, Illustrative ACS 1-Year Period Estimates, 2010–2014, BIG CITY and VERY BIG CITY

A. BIG CITY (250,000 people)

2010

2011

2012

2013

2014

(1) Percent poor school-age children

15.0

16.0

17.0

19.0

22.0

90% MOE

±2.99

±3.07

±3.15

±3.28

±3.47

(2) Difference from prior year

1.0

1.0

2.0

3.0

90% MOE

±4.29

±4.39

±4.55

±4.78

(3) Difference from 2010

1.0

2.0

4.0

7.0

90% MOE

±4.29

±4.34

±4.44

±4.58*

B. VERY BIG CITY (1,000,000 people)

2010

2011

2012

2013

2014

(1) Percent poor school-age children

15.0

16.0

17.0

19.0

22.0

90% MOE

±1.49

±1.53

±1.57

±1.64

±1.73

(2) Difference from prior year

1.0

1.0

2.0

3.0

90% MOE

±2.14

±2.20

±2.27

±2.39*

(3) Difference from 2010

1.0

2.0

4.0

7.0

90% MOE

±2.14

±2.17

±2.22*

±2.29*

* statistically significant at the 90% confidence level.

NOTE: MOE margin of error (refer to Box 2-5).

What can the user conclude about changes in school-age poverty rates over time?

• The year-to-year differences (line 2) for BIG CITY are not statistically significant, even though the example purposefully accelerates the increases in school-age poverty compared with Table 3-3; the only significant 1-year difference for VERY BIG CITY is the 3 percentage point increase in school-age poverty between 2013 and 2014.

• As 1-year estimates accumulate beginning in 2010, however, the differences from 2010 (line 3) are significant by 2014 for BIG CITY and by 2013 for VERY BIG CITY, as the size of the difference increases (from 1 percentage point between 2010 and 2011 to 7 percentage points between 2010 and 2014).

possible to use time-series modeling to improve the statistical power of the analysis (that is, the power to detect statistically significant differences) by taking the entire series into account.


Using 5-Year Period Estimates to Estimate Change for Smaller Cities or Subareas of Large Cities Now consider cities and subcity areas for which there are not precise 1-year or 3-year period estimates for population

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

groups as small as poor school-age children. In this situation, analyses of change must use 5-year period estimates, but comparisons of successive 5-year period estimates will not have the precision that one might assume from the additional sample.

Part A of Table 3-5 compares pairs of successive 5-year period estimates for the rates of school-age poverty in SMALL CITY or BIG CITY SUBAREA (population 50,000, including 10,000 school-age children). Each pair of estimates is 1 year apart. For example, an estimate for 2010–2014 is compared with the corresponding estimate for 2011–2015, and so on through the comparison of an estimate for 2014–2018 with the corresponding estimate for 2015–2019. For simplicity it is assumed that the population size remains constant across the decade.

The individual 5-year period estimates are constructed by assuming that the underlying 1-year period estimates increase each year by 1.2 percentage points from 2010 (15 percent school-age poverty rate) to 2019 (25.8 percent school-age poverty rate). Consequently, the estimated difference between each pair of 5-year period estimates that are 1 year apart is also 1.2 percentage points, with an estimated 90 percent margin of error of about 2.1 percentage points. Consequently, none of these differences is statistically significant for SMALL CITY or BIG CITY SUBAREA: school-age poverty could have increased by more than 3 percent or it could have decreased by as much as 1 percent (1.2 ± about 2.1).

The reason that none of the differences is statistically significant is that each pair of 5-year period estimates being compared shares 4 of 6 years in common. For example, in the comparison between the 2010–2014 estimate and the 2011–2015 estimate, the years 2011, 2012, 2013, and 2014 are shared in common. The only new data in the comparison are for the first and the sixth years—2010 in the 2010–2014 estimate and 2015 in the 2011–2015 estimate.

Statistically, the comparisons between adjacent pairs of 5-year period estimates are the equivalent of taking one-fifth of the 5-year difference between year 1 and year 6 as if one had available the 1-year period estimates for those 2 years (assuming that the population size remains the same over the period—see Chapter 6 for further detail). Thus, in Table 3-5, for the comparison between the estimates for 2010–2014 and 2011–2015, one-fifth of the difference between an assumed 15 percent poor school-age children in 2010 and an assumed 21 percent poor school-age children in 2015 is 1.2 percent (6 percent divided by 5).

Such a comparison will have a large sampling error for an area with only 50,000 people and 10,000 school-age children, which can be seen by considering the sampling errors for the assumed underlying 1-year period estimates of school-age poverty. For example, the assumed estimate of 15 percent poor school-age children in 2010 will have a coefficient of varia-

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-5 Analyzing Trends Over Time for School-Age Poverty Rates, Illustrative ACS 5-Year Period Estimates, SMALL CITY or Subarea of BIG CITY with 50,000 People and 10,000 School-Age Children, 2010–2019

A. Estimating Year-to-Year Change by Comparing Overlapping Pairs of 5-Year Period Estimates, Assuming an Underlying Linear Upward Trend in School-Age Poverty

 

Percent Poor School-Age Children

Difference from Prior Period

Estimate

90% MOE

Estimate

90% MOE

2010–2014

17.4

±3.2

2011–2015

18.6

±3.3

1.2

±2.0

2012–2016

19.8

±3.3

1.2

±2.1

2013–2017

21.0

±3.4

1.2

±2.1

2014–2018

22.2

±3.5

1.2

±2.2

2015–2019

23.4

±3.5

1.2

±2.2

NOTES: MOE = margin of error. The formula for calculating standard errors for estimates of change has been adjusted in the case of overlapping pairs of estimates to take account of the data shared in common; see Table 6-4.

• To create BIG CITY subareas, the user must aggregate 5-year period estimates for census tracts.

• The above 5-year period estimates are assumed to reflect 1-year period estimates of school-age poverty as follows:

2010 15.0% 2015 21.0%

2011 16.2 2016 22.2

2012 17.4 2017 23.4

2013 18.6 2018 24.6

2014 19.8 2019 25.8

What can the user learn from this example (Part A, which compares pairs of overlapping 5-year period estimates, assuming a linear upward trend in school-age poverty)?

• None of the differences between adjacent overlapping pairs of 5-year period estimates for SMALL CITY or a subarea of BIG CITY (each with 50,000 people) is statistically significant.

• The reason is the substantial overlap between adjacent pairs of 5-year period estimates—they share 4 of 6 years in common (for example, 2011, 2012, 2013, and 2014 in the comparison between the 2010–2014 estimate and the 2011–2015 estimate)—in which the only new data are for 2010 and 2015.

• Assuming no change in the size or demographic composition of the population over time, the differences between adjacent pairs of 5-year period estimates, with 4 of 6 years overlapping, are the equivalent of computing one-fifth of the change between year 1 and year 6 as if one had 1-year period estimates for those two years (see text; see also Section 6-C for the mathematical proof).

• One-fifth of the change between years 1 and 6 is likely to be a small number—it is only 1.2 percent in the data shown above (for example, one-fifth of the difference between 15 percent in 2010 and 21 percent in 2015). Consequently, the sampling error relative to the size of the estimate will be large for an area as small as 50,000 people with only 10,000 school-age children (see text).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

B. Estimating Change by Comparing Pairs of 5-Year Period Estimates That Overlap Less and Less, Assuming an Underlying Linear Upward Trend in School-Age Poverty

 

Percent Poor School-Age Children

Difference from Prior Period

Estimate

90% MOE

Estimate

90% MOE

(i) 2 years apart (3 years in common)

 

 

 

 

2010–2014

17.4

±3.2

2012–2016

19.8

±3.3

2.4

±2.9

(ii) 3 years apart (2 years in common)

 

 

 

 

2010–2014

17.4

±3.2

2013–2017

21.0

±3.4

3.6

±3.6

(iii) 4 years apart (1 year in common)

 

 

 

 

2010–2014

17.4

±3.2

2014–2018

22.2

±3.5

4.8

±4.3*

(iv) 5 years apart (0 years in common)

 

 

 

 

2010–2014

17.4

±3.2

2015–2019

23.4

±3.5

6.0

±4.7*

NOTES: * = statistically significant at the 90 percent confidence level. The above 5-year period estimates are assumed to reflect the same 1-year period estimates used in Part A above. What can the user learn from this example (Part B, which compares pairs of 5-year period estimates that overlap less and less, assuming a linear upward trend in school-age poverty)?

• The differences between 5-year estimates that are 2 years apart and 3 years apart are not significant, but the differences between 5-year estimates that are 4 years and 5 years apart are significant.

• The reason is the decreasing extent of overlap between pairs of 5-year period estimates, which adds more new data to the comparison, thereby increasing the precision of the estimated difference (see discussion in the text).

C. Estimating Change by Comparing Pairs of 5-Year Period Estimates That Overlap Less and Less, Assuming a Jump in School-Age Poverty in 2015

 

Percent Poor School-Age Children

Difference from Prior Period

Estimate

90% MOE

Estimate

90% MOE

(o) 1 year apart (4 years in common)

 

 

 

 

2010–2014

17.0

±3.1

2011–2015

18.2

±3.2

1.2

±2.0

(i) 2 years apart (3 years in common)

 

 

 

 

2010–2014

17.0

±3.1

2012–2016

19.4

±3.3

2.4

±2.9

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

 

Percent Poor

School-Age Children

Prior Period

Difference from

Estimate

90% MOE

Estimate

90% MOE

(ii) 3 years apart (2 years in common)

 

 

 

 

2010–2014

17.0

±3.1

2013–2017

20.6

±3.4

3.6

±3.6

(iii) 4 years apart (1 year in common)

 

 

 

 

2010–2014

17.0

±3.1

2014–2018

21.8

±3.5

4.8

±4.3*

(iv) 5 years apart (0 years in common)

 

 

 

 

2010–2014

17.0

±3.1

2015–2019

23.0

±3.5

6.0

±4.7*

NOTES: * = statistically significant at the 90 percent confidence level. The above 5-year period estimates are assumed to reflect 1-year period estimates of school-age poverty estimates as follows:

2010 17% 2015 23%

2011 17 2016 23

2012 17 2017 23

2013 17 2018 23

2014 17 2019 23

What can the user learn from this example (Part C, which compares pairs of 5-year period estimates that overlap less and less, assuming an upward jump in school-age poverty in 2015)?

• The estimated differences between 5-year period estimates that are 1 year apart are the same as in Part A above; the estimated differences that are 2, 3, 4, or 5 years apart are the same as in Part B above.

• The user will need to use auxiliary knowledge, which may include 1-year or 3-year period estimates for larger geographic areas, to distinguish the nonlinear underlying pattern of school-age poverty in Part C from the linear underlying pattern in Parts A and B.

tion of about 27 percent and a 90 percent margin of error of ±6.7 percent, meaning that the 90 percent confidence interval ranges from 8.3 to 21.7 percent (see Tables 2-7a and 2-7b). Consequently, for a difference between the 2010 estimate and the corresponding estimate for 2015 to be statistically significant, that difference must be very large.

Part B of Table 3-5 compares pairs of 5-year period estimates for school-age poverty in SMALL CITY or BIG CITY SUBAREA that overlap less and less, in which the underlying trend is also a steady increase of 1.2 percent in the percentage poor school-age children from one year to the next. The differences between 5-year estimates that have 3 years’ overlap

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

(i) or 2 years’ overlap (ii) are not significant, but the differences between 5-year estimates that have only 1 year’s overlap (iii) or no overlap (iv) are significant. The reason is that the decreasing extent of overlap between pairs of 5-year period estimates adds more new data to the comparison, thereby increasing the precision of the estimated difference.

Imagine that one had available estimates of the difference between each pair of years that are 5 years apart. Then, from Part B of Table 3-5:

  1. The comparison between two 5-year period estimates that overlap by 3 years (instead of 4 years as in Part A) is the equivalent, statistically, of taking two-fifths of the average 5-year difference between the two pairs of years that are not shared in common. For example, in comparing the 5-year period estimates for 2010–2014 and 2012–2016, years 1 and 6 (2010, 2015) and years 2 and 7 (2011, 2016) provide new data not shared in common. The difference in the assumed school-age poverty rates between each of these pairs of years is 6 percent, and two-fifths of the average difference (6 percent) is 2.4 percent, which is the difference shown between the two 5-year period estimates for 2010–2014 and 2012–2016. This difference is more precise than when the overlap between pairs of 5-year period estimates is 4 years and only one pair of years is not shared in common, but not precise enough for statistical significance.

  2. The comparison between two 5-year period estimates that overlap by 2 years (instead of 3 or 4 years) is the equivalent, statistically, of taking three-fifths of the average 5-year difference between the three pairs of years that are not shared in common: for example, years 1 and 6 (2010, 2015), years 2 and 7 (2011, 2016), and years 3 and 8 (2012, 2017). The estimated difference between the two 5-year period estimates, which works out to 3.6 percent, still does not attain statistical significance in this example.

  3. The comparison between two 5-year period estimates that overlap by 1 year (instead of 2, 3, or 4 years) is the equivalent, statistically, of taking four-fifths of the average 5-year difference between the four pairs of years that are not shared in common: years 1 and 6 (2010, 2015), years 2 and 7 (2011, 2016), years 3 and 8 (2012, 2017), and years 4 and 9 (2013, 2018). The estimated difference between the two 5-year period estimates, which works out to 4.8 percent, is statistically significant.

  4. Finally, the comparison between two 5-year period estimates that do not overlap at all is the equivalent, statistically, of taking the average 5-year difference between all five pairs of years that are not shared in common. The estimated difference between the two

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

5-year period estimates, which works out to 6.0 percent, is also statistically significant.

The drawback of making comparisons with 5-year period estimates that overlap very little or not at all is that the user must wait for the second set of estimates to become available for the analysis. SMALL CITY will have 3-year period estimates available, and the wait for a second, non-overlapping 3-year period estimate would not be as long as for a second, nonoverlapping 5-year period estimate. However, unless SMALL CITY experienced a very large increase in school-age poverty over a 6-year period, the comparison of 3-year period estimates that were 3 years apart and did not overlap (for example, 2010–2012 and 2013–2015) would not likely yield a significant result, so the user may need to turn to comparisons of 5-year period estimates.

The example of comparing 5-year period estimates in Table 3-5, Parts A and B, is simplistic because it assumes that the total number of school-age children does not change over the period. Also, the example projects a constant linear increase of 1.2 percentage points each year in school-age poverty from 15.0 percent in 2010 to 25.8 percent in 2019. Of course, poverty (and other characteristics) may change at varying rates and in different directions, and the user will not know the underlying dynamics of year-to-year change in 5-year (or 3-year) period estimates.

Part C of Table 3-5 provides an example that produces the same estimates of differences between 5-year period estimates as in Parts A and B but with a distinctly different underlying trend in the data: in this example, school-age poverty is assumed to be static at 17 percent for the years 2010–2014 when it jumps to 23 percent in 2015 (perhaps because a large employer left town) and remains at that level for the years 2016–2019. The interpretation of the differences between pairs of 5-year period estimates is the same as in Parts A and B—namely, that each difference is a fraction (one-fifth, two-fifths, three-fifths, four-fifths, or five-fifths) of the average difference between the pairs of years that are not shared in common. The user must use other information, however, to differentiate between the linear upward trend in poverty in Parts A and B and the jump in poverty in Part C. Examining 1-year or 3-year period estimates for larger geographic areas, such as counties or PUMAs, may help assess the underlying dynamics of change for SMALL CITY or BIG CITY SUBAREA.

3-C.1.c Case Study

The following case study of a rezoning initiative in a large city illustrates the potential usefulness of up-to-date ACS estimates. It starts by describing how data for a housing planning policy for a neighborhood would

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

be obtained before the advent of the ACS and then indicates how ACS data could be used to better inform the policy makers.


Background Area X is a neighborhood in BIG CITY that for decades has housed primarily working-class Polish, Hispanic, and Orthodox Jewish families. In recent years, increasing numbers of artists, young professionals, and students priced out of the upscale portion of BIG CITY have moved into the area. In response to strong demand for housing in Area X, BIG CITY’s planning department initiated an effort to allow new housing development on underused waterfront land abutting the area. As part of an environmental review before the rezoning could take effect, city planners had to determine whether the introduction of new housing could displace the existing residential population through rising rents. Since the proposal would allow the development of luxury apartments in a neighborhood that consisted mostly of modest worker housing, it seemed likely that existing residents could be priced out of the market by newcomers. However, it also seemed likely that residents who had recently moved to the area already had many of the socioeconomic characteristics expected of residents in the proposed new housing—in short, that demographic change was already well under way and indirect displacement was already occurring in certain parts of the study area.


Data and Analysis Needs In order to determine specifically which populations were potentially vulnerable to displacement, neighborhood-level analysis of socioeconomic and housing data was necessary. In 2000, Area X included 33,000 occupied housing units and 80,000 residents. However, the 2000 census data did not capture the rapid social and economic change that had occurred in the area in recent years, and no post-2000 data were available to evaluate trends. In an attempt to validate anecdotal evidence of change, BIG CITY’s planners supplemented the 2000 census data with other evidence of socioeconomic change, including newspaper articles, new housing construction permits, interviews with brokers about rising rents, surveys of illegal loft conversions, and documented cases of new capital investment and economic activity. BIG CITY did not have an ongoing household survey (as some cities have undertaken at some times), and it did not have the time or resources to conduct one.

Having current and historical statistical data on income, occupation, rents, housing value, and other items to assess changes in the characteristics of area residents would be invaluable for a task that has significant implications for policy and program development. Once the ACS is fully implemented and there is no longer a need to wait 10 years for a new long-form sample, then BIG CITY’s planners would be in a much better position

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

to undertake an analysis of this nature. The example below assumes that the year is 2016.


Strategies for Using the ACS An initial strategy for BIG CITY’s planners to consider, assuming that Area X is contained within a single PUMA and hence represents a large proportion of the PUMA’s population, is to use the ACS 3-year period estimates for the PUMA as a proxy to track population growth and changes in socioeconomic composition of Area X. In fall 2016, 3-year period estimates for the PUMA could be compared for, say, 2007–2009, 2010–2012, and 2013–2015. While 1-year period estimates would also be available for the PUMA, they might not be sufficiently precise for the purpose—see Table 2-7a. A variant of this strategy would be to average two years of 1-year period estimates for the PUMA and compare the 2-year averages for, say, 2006–2007, 2008–2009, 2010–2011, 2012–2013, and 2014–2015.

The analysts would need to consider three potential problems that could affect the results. First, the head count and age, race, and sex composition of the population would likely differ for the PUMA before and after the 2010 census because of inaccuracies in the pre-2010 population controls (see Section G). Second, because the PUMA in this example is a subcity area and not a county, the sampling error of its estimates would likely be higher than if it had benefited from PUMA-level population controls rather than the county-level controls used in the ACS. Third, the PUMA in this example is somewhat larger than Area X, and it is possible that the PUMA population outside Area X differs from the Area X population in ways that could affect the results.

A second strategy would be for the planners to use the ACS 5-year period estimates for an aggregation of the census tracts or block groups making up Area X. The combined 5-year period estimates could then be compared for, say, 2006–2010 and 2011–2015. Again, corrections to the population controls from the 2010 census could distort the precensus and postcensus comparisons.

A combined strategy could make good use of all of the available data. In such a strategy, comparing the 5-year period estimates for the census tracts making up Area X to the 5-year period estimates for the larger PUMA could help assess the validity of using 3-year (or 2- or 1-year) period estimates for the entire PUMA as a proxy for Area X. The advantage of being able to use 2-year or 3-year period estimates is that they will better capture trends than the 5-year period estimates that average the data over a longer time span.

Whichever strategy the planners ultimately select, the availability of ACS estimates would be a vast improvement over the current situation in which indirect or partial measures of change had to suffice. The ACS data

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

would permit BIG CITY to make a much more informed assessment of the extent of displacement of current residents that was already occurring and would likely occur with the rezoning. The results of the analysis would inform policy makers, lawmakers, and advocacy groups about neighborhood change, ultimately affecting which policies would be supported and where limited resources would be spent.

3-C.2
Small Jurisdiction Applications of the ACS

Generally, smaller counties, cities, and other governmental and statistical areas will not benefit as much from the ACS as larger areas, if only because larger areas will have more sets of estimates published for them (1-year, 3-year, and 5-year period estimates for areas with at least 65,000 people, and 3-year and 5-year period estimates for areas with at least 20,000 people). In some states, sizeable proportions of the population live in small counties, cities, towns, and school districts that will have only 5-year period estimates from the ACS. In 2000, for example, the percentages of people living in counties with fewer than 25,000 residents exceeded 20 percent in 7 states: Alaska (22 percent), Arkansas (27 percent), Idaho (25 percent), Montana (34 percent), North Dakota (47 percent), South Dakota (57 percent), and Wyoming (31 percent) (from the 2002 Census of Governments, U.S. Census Bureau, 2002a:Table 6).

Five-year period estimates for areas this small will be subject to large levels of sampling error (refer back to Tables 2-7a, 2-7b, and 2-7c), although the oversampling of housing units in very small areas will help their precision somewhat. Consider the 15 percent of people in North Dakota and 24 percent of people in South Dakota who live in cities with fewer than 1,000 residents. Over a 5-year period, these areas will be sampled initially at rates of 1 in 3 housing units (if they have between 500 and 1,000 residents) or 1 in 2 housing units (if they have fewer than 500 residents), compared with the average ACS initial sampling rate of 1 in 9 housing units (refer back to Table 2-3, Part A). This oversampling will reduce the sampling error of estimates for these areas by about 40-50 percent compared with the sampling error of estimates for areas with a 1 in 9 sampling rate (assuming that the areas have the same combined mail and computer-assisted telephone interviewing [CATI] response rates and therefore the same computer-assisted personal interviewing [CAPI] subsampling rates).

Oversampling also benefits many larger areas that contain very small cities, townships, or school districts. Selecting just one of many such examples, in 2000, Iowa County, Wisconsin, had 22,780 residents living in 11 cities and 14 towns (U.S. Census Bureau, 2002a:Table 16). Careful examination of the population size of each subcounty jurisdiction would be required to determine the effect of oversampling, but it seems likely that the

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

entire county would be sampled at 5-year cumulative rates of 1 in 2 or 1 in 3 households instead of at an average rate of 1 in 9 or less. The effect would be to reduce the sampling error for estimates of the entire county to the point that they could meet common standards of acceptable precision.12

Even with oversampling, however, the 5-year period estimates for very small governmental units will fall far short of common standards of precision for many population groups of interest. For example, the 90 percent confidence interval for an estimate of 15 percent poor school-age children for an oversampled area of 1,500 people, based on the assumptions in Table 2-7c, would likely range from 7 to 23 percent, which is not very informative about the extent of school-age poverty. By contrast, the 90 percent confidence interval for an estimate of 15 percent poor school-age children for an area of 50,000 people would likely range from 12 to 18 percent, which is a considerable improvement.

It is important to remember that the 2000 long-form-sample estimates were also subject to considerable sampling error for small areas. However, they were somewhat more precise than the corresponding estimates from the ACS cumulated over 5 years.

The precision of the 5-year period ACS estimates can be improved by aggregating small areas into larger units. Indeed, this is the recommended strategy for large jurisdictions—namely, to aggregate census tracts and block groups into larger subcity or subcounty areas for such purposes as planning the location of governmental service sites and services. A strategy of aggregation is not as suitable for small governmental jurisdictions, given that each typically provides its own services and is interested in estimates for its jurisdiction alone.

Small jurisdictions could ask the Census Bureau to provide estimates for, say, 8-, or 10-year periods that are more precise than the 5-year period estimates. The drawback of this approach is that lengthening the period of the estimates averages underlying patterns of variation in social and economic phenomena over longer periods and does not produce large gains in precision. For the case of a town of 1,500 people, the 90 percent confidence interval for an estimate of 15 percent poor school-age children would be reduced from 7 to 23 percent for the 5-year period estimate to 8.7 to 21.3 percent for an 8-year period estimate and to 9.3–20.7 percent for a 10-year period estimate (under the assumptions used in Table 2-7c). By comparison, the 90 percent confidence interval for the same estimate from the 2000 long-form sample would be 9.7 to 20.3 percent.

To produce reasonably precise estimates for small population groups in

12

The beneficial effects on sampling error for county estimates that result from oversampling subcounty areas are not as great when the subcounty areas are sampled at varying rates, such as 1 in 2, 1 in 3, 1 in 6, and 1 in 9.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

small jurisdictions would require a significant expansion of the ACS. Many such groups are of interest to users, including not only poor school-age children, as discussed in this report, but also ethnic and language minorities, veterans, and people with disabilities. Increasing the final 5-year ACS sample size (after subsampling for CAPI follow-up) to equal the originally proposed size (which was double the current size—see Section 1-B.3) would certainly help. However, acceptable precision for small groups could still often require aggregating estimates over 8 to 10 years.

Of course, ACS estimates for larger population groups will be more precise than those for small groups, and the 5-year period estimates for some large groups in small jurisdictions may reach acceptable precision, particularly if the jurisdiction’s housing units are oversampled. For example, a 5-year period estimate of 15 percent total poor people in an oversampled jurisdiction of 1,500 people will have a 90 percent confidence interval of 11.4 to 18.6 percent, which is much narrower than the interval of 7.0 to 23.0 percent for poor school-age children.

Small jurisdictions may be able to use the levels and trends in the more precise 5-year period estimates for similar but larger jurisdictions to improve understanding of what is occurring for their jurisdiction. Moreover, small jurisdictions, just as large jurisdictions, will benefit from the fact that ACS multiyear period estimates never become as outdated as the long-form-sample estimates do before they are replaced by estimates from the next census.

3-C.3
Special Case of Seasonal Populations

Some jurisdictions in the United States have large, seasonal fluctuations in population. Examples include many college towns, the west and east coasts of Florida, parts of Arizona, the northern parts of Wisconsin, Minnesota, and Michigan, and the Atlantic beaches. Because of the continuous sampling and data collection for the ACS and its use of a 2-month residence rule instead of the “usual residence” rule of the decennial census, the ACS estimates for an area with seasonal fluctuations in population will likely differ from the long-form-sample estimates for the same area.

Table 3-6 works through a simplified example for a hypothetical county in Florida. This county is assumed to have a year-round population of 100,000, of whom 20,000 (20 percent) are poor, and a winter (December-March) population of 300,000, of whom 35,000 are poor (11.7 percent, averaging the 20 percent year-round poverty rate with a rate of 7.5 percent for the richer, part-time residents). Over the entire year, on average, there were 166,667 people in the county, of whom 25,000 were poor (15 percent poverty rate, averaging the year-round poverty population for 8 months and the winter poverty population for 4 months).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-6 Hypothetical County in Florida with Winter Influx of Residents

County Characteristics, January–December 2010

 

Assumed Distribution

Measured by the ACS (Before Controls)

Year-Round Pop.

Seasonal Population

Total Population

Total

Poor

Total

Poor

Total

Poor

(%)

January

100,000

20,000

200,000

15,000

300,000

35,000

11.7

February

100,000

20,000

200,000

15,000

300,000

35,000

11.7

March

100,000

20,000

200,000

15,000

300,000

35,000

11.7

April

100,000

20,000

100,000

20,000

20.0

May

100,000

20,000

100,000

20.000

20.0

June

100,000

20,000

100,000

20,000

20.0

July

100,000

20,000

100,000

20,000

20.0

August

100,000

20,000

100,000

20,000

20.0

September

100,000

20,000

100,000

20,000

20.0

October

100,000

20,000

100,000

20,000

20.0

November

100,000

20,000

100,000

20,000

20.0

December

100,000

20,000

200,000

15,000

300,000

35,000

11.7

12-month average

100,000

20,000

66,667

5,000

166,667

25,000

15.0

 

(20.0% poor)

(7.5% poor)

 

 

 

NOTE: For ease of presentation, the example unrealistically assumes zero year-round population growth over the year and that all seasonal residents arrive December 1 and leave March 31.

Population Controls, 2010

 

Total

Poor

(%)

April 1, 2010 census estimate

100,000

N.A.

N.A.

July 1, 2010 population estimate

100,000

N.A.

N.A.

NOTE: The July 1, 2010, population estimate updates the 2010 census estimate with administrative records. For ease of presentation, the example unrealistically assumes zero population growth from April to July; actual growth might be a fraction of 1 percent.

Hypothetical Estimates of Total Population and Number and Percent Poor, 2010

 

Total

Poor

(%)

Census long-form-sample estimate, 2010 (based on March–June data with April population control)

100,000

20,000

20.0

ACS 1-year period estimate, 2010 (controlled) (based on 12-month average data with July population control)

100,000

15,000

15.0

ACS 1-year period estimate, 2010 (not controlled) (based on 12-month average data, no control applied)

167,000

25,000

15.0

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

Assuming the winter population has left the area entirely by March 31, the Census Bureau would estimate the county’s April 1 population at 100,000 and its July 1 population at about the same number. Therefore, in this very stylized example, both the ACS and a long-form sample conducted in the same year would provide a total population figure for the county of about 100,000 (since the ACS weighting procedure adjusts the ACS sample to conform to the July 1 county population estimates), but the composition of the population would differ between the two surveys. The long-form sample would provide an estimate of 20,000 poor people (20 percent poverty rate for the year-round population). The ACS would provide a 1-year period estimate of 15,000 poor people (15 percent average for the year-round and seasonal populations combined). Note that the percentage of people in poverty from the ACS estimate reflects the average composition of the population over the year; however, the number of poor people is lower than both the long-form-sample estimate and a 12-month average of the ACS that is not constrained to the July population control.

This example is exaggerated, but it does point up the differences between the long-form sample and the ACS for areas that experience significant seasonal fluctuations of population and for which the socioeconomic characteristics of the seasonal and year-round populations differ appreciably. In these instances, the long-form sample provides the numbers and characteristics of the population as of April 1. The ACS provides comparable population numbers by age, race/ethnicity, and sex based on July 1 postcensal estimates, even though the total population, as well as demographic groups (for example, young and older people), may change during the year. For socioeconomic characteristics, the ACS provides percentages that reflect the average experience of the area over the year; however, the percentages are applied to the July 1 population figures so that the numbers are neither the same as the long-form-sample estimate nor the same as an average estimate from the ACS that is not controlled to the census-based population estimates (see further discussion in Section 4-A.5).

For most areas, this problem will not be significant because seasonal increases (or decreases) in population are a small percentage of the year-round population, or the characteristics of seasonal and year-round residents do not differ appreciably. In areas for which users believe that seasonal differences may be significant, they may wish to make a case to the Census Bureau of the need for tabulations of their population at different times of the year (see Section 7-D.2).

3-D
TRANSPORTATION PLANNING USES

Transportation planners are devoting considerable effort to understanding the ACS, determining how to work with the data, and identifying

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

concerns to raise with the Census Bureau about the data products. Their efforts in this regard go back to the beginning of the ACS (see, for example, Bureau of Transportation Statistics, 1996). With funding from federal transportation agencies, a committee of the National Research Council’s Transportation Research Board organized a conference on “Census Data for Transportation Planning—Preparing for the Future” in May 2005. The conference covered a wide range of topics and issues regarding the opportunities and challenges presented by the advent of the ACS (see http://trb.org/conferences/censusdata/Program.pdf).

The transportation community’s interest in the ACS is explained by the central role that the long-form-sample data have historically played in transportation applications ranging from nationwide program planning and evaluation to local analysis of commuting patterns. Questions on place of work, means of transportation to work, length of commute, and vehicle ownership have been included on the long-form questionnaire for three or more decades, as have questions about disabilities that make it difficult for people to work or to go outside their homes (see Citro, 2000b).

The U.S. Department of Transportation has worked closely with the Census Bureau and with state transportation departments and metropolitan planning organizations to improve the quality of the data on place of work (by, for example, encouraging large employers to inform workers of the addresses to report for particular workplaces) and to develop special tabulations for transportation users. The Census Transportation Planning Package (CTPP) has been produced from censuses beginning in 1970 and includes tabulations of households and workers by place of residence, workers by place of work, and flows between place of residence and place of work for each traffic analysis zone (TAZ). There are a large number of such zones, designated by states and regional transportation agencies, each comprised of one or more blocks, block groups, or census tracts within metropolitan areas.13

Regional and metropolitan transportation planning organizations are also heavy users of the long-form-sample PUMS 5 percent sample files, which in 2000 provided records for 14 million individuals, with geographic identification by state and PUMA (areas of about 100,000 population). The long-form-sample PUMS files are the basis of sophisticated transportation activity modeling systems that contain synthetic population models for a base year and, say, 20 years into the future. The population models are calibrated to control totals for the base year and future years on total households, households by income level, and other characteristics that are estimated by the regional organization at the county or TAZ level. The models are then used to predict activity and travel patterns at the person, household, or trip level.

13

See http://www.trbcensus.com/ctpp.html; National Research Council (1995:App. G).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-D.1
Using the ACS 1-Year PUMS Files

Transportation planners are concerned that the ACS yearly PUMS product will contain only about 3 million person records. This reduction from 14 million persons means that the sampling error of estimates from the ACS 1-year PUMS will be much larger than those of estimates from the 2000 long-form-sample 5 percent PUMS, and estimates from the long-form-sample 5 percent PUMS are already subject to about 1.8 times more sampling error than estimates from the full long-form sample.

As an example, for a PUMA with 50,000 workers, an estimate from the 2000 long-form sample 5 percent PUMS that 15 percent of workers carpooled to get to work would have a 90 percent margin of error of approximately plus or minus 1.6 percent—1.83 times the margin of error of about ±0.9 percent for the full long-form sample (see the fourth row in Table 2-7b). This margin of error equates to a coefficient of variation of 6.5 percent. However, a corresponding estimate from the ACS 1-year PUMS would have a 90 percent margin of error of at least plus or minus 3.6 percent based simply on the difference in the number of records. This margin of error equates to a coefficient of variation of 14.5 percent, which does not meet accepted standards for precision. Moreover, the weights in the ACS PUMS will be more variable than those in the 2000 long-form-sample PUMS due to the subsampling for CAPI follow-up in the ACS. Consequently, estimates from the ACS PUMS will likely be even less precise compared with estimates from the 2000 long-form-sample PUMS than indicated above.14

A possible solution for the smaller size of the ACS PUMS is to combine two or more PUMS. While transportation modelers will not likely want to fully analyze each new PUMS release because of the time and resources that would require, the availability of an annual PUMS will make it possible to periodically check and recalibrate their models. Similarly, the availability of updated ACS 5-year period estimates will make it possible to reestimate control totals for the models at the county and TAZ levels more often than once a decade.

14

The current scheme for selecting the ACS PUMS files draws an equal-probability systematic sample of all ACS housing unit records and their household members in each state, with the records sorted by several characteristics (see the 2005 PUMS accuracy statement at http://factfinder.census.gov/home/en/acs_pums_2005.html). A different selection scheme would retain a higher proportion of the CAPI cases so as to equalize the weights of CAPI and non-CAPI cases, yielding a PUMS that would produce more precise estimates than the current PUMS. This scheme could be extended toward equalizing the weights of all sampled housing unit records within PUMAs.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-D.2
Using the ACS TAZ Data

A concern in using the ACS 5-year period estimates for traffic analysis zones is that the 60-month averages that underlie the estimates may obscure short-term changes in commuting patterns that occur in response to marked changes in the local economy or the transportation infrastructure. To address this concern, transportation planners in large cities and metropolitan areas can examine 3-year or 1-year period estimates for the area as a whole, for PUMAs, and, in some cases for smaller cities and towns. Analyses of these estimates can provide an overall sense of changes in commuting modes and times to work that can inform assessments of the usefulness of the 5-year period estimates.

Precision is also a very serious concern for 5-year period TAZ estimates. Statistical mapping techniques may help transportation planners extract useful information from the estimates in some instances. For example, by geographically displaying such variables as mode of transit to work, where workers live, and where workers work on maps of transportation routes, places of employment, and other local features, planners may see patterns that suggest how to combine TAZ estimates to produce meaningful larger areas that have more precise estimates. (Such statistical mapping techniques may help users in other fields extract value from ACS 5-year period estimates for census tracts and block groups.)

The usefulness of 5-year period TAZ estimates also depends importantly on two other factors: (1) procedures that the Census Bureau uses for imputing missing responses and (2) decisions it makes regarding the data that can be provided while protecting confidentiality. Regarding imputation, the Census Bureau needs to engineer its data processing so that imputations for missing responses to commuting items can be made at the outset at the block level. In the long-form-sample processing, imputations for these items were made initially at the city level and only subsequently, in the CTPP, carried out at the block level.

Regarding confidentiality, the Census Bureau needs to consider carefully the added confidentiality protection afforded by 5-year averages compared with point-in-time estimates. The added protection results from the fact that many people change one or more characteristics of interest over a 5-year period, such as place of work, occupation, place of residence, commuting mode, etc. Consequently, the risk of reidentification of a specific individual in 5-year aggregations is reduced. Taking account of this added protection should enable the Census Bureau to release sufficient information on commuting (and other topics) to be useful at the level of traffic analysis zones, block groups, and census tracts (see Section 4-D.1).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-D.3
Conclusion on Using the ACS for Transportation Planning

While transportation planners face significant challenges in using ACS data for applications for which they have previously used the long-form sample, the frequency of release of updated ACS estimates offers benefits to them. The 1-year and 3-year period estimates can help transportation planners track overall trends in commuting patterns and other aspects of household transportation and alert them to the need for special surveys or other data collections to update their models. The 5-year estimates can provide intercensal checks on local-area transportation patterns that would not be possible with the decennial long-form sample, although estimates for traffic analysis zones will often need to be combined to attain an acceptable level of precision. The ACS PUMS can be used in a variety of ways, and it is issued more frequently than the long-form-sample PUMS.

3-E
ACADEMIC RESEARCH USES

Researchers in universities, colleges, research institutes, and other settings have made extensive use of long-form-sample data to understand key social processes, such as migration flows, changes in marriage patterns and family living arrangements, and the social and economic effects of the aging of the population. They have also used long-form-sample data to develop insights on such important topics as trends in educational attainment, magnitudes and effects of immigration from abroad, and concentrations of people in poverty.

Some research applications have used summary files of detailed tabular data for small areas, such as Summary Files 3 and 4 from the 2000 census. For example, summary files have supported analyses of migration flows among regions, states, counties, and places and of concentrated populations of the poor, minorities, and immigrants from different countries. Summary information on neighborhood characteristics has been appended to the records of respondents to such ongoing research surveys as the Panel Study of Income Dynamics. This additional information has permitted rich contextual analyses of family social and economic dynamics.

Other research applications have used the PUMS files, which have been constructed for most censuses back to 1850 (Ruggles, 2000). PUMS files permit detailed, multivariate analyses on such topics as the interactions among disability, educational attainment, labor force attachment, and income and the characteristics that distinguish people who migrate long distances from those who migrate shorter distances or not at all.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-E.1
Using Summary Files for Research

At present, no equivalent of Summary File 3 or 4 from the 2000 long-form sample exists for the ACS. This lack is a drawback for the research community because summary files permit ready analysis of detailed information across multiple geographic areas, population groups, and subject areas. In contrast, the tables that are available online for the ACS can only be displayed one at a time for a specified type of geographic area within a larger unit—for example, a table of age by sex for one or all towns in a particular county. The detailed tables and single-year and multiyear profiles are also available as spreadsheets through the ACS FTP site, and in that format the data can be manipulated (for example, calculating percentages or adding or subtracting categories), but the spreadsheet contents are limited to a specified geographic area, such as a county or township. An ACS Download Center provides access to up to 50 tables for a geographic summary level, such as all states or all counties. None of these data products are as useful for research purposes as a summary file in the same format as the decennial census summary files.

The Census Bureau recently began work to specify and implement an ACS equivalent of Summary File 3 from the 2000 long-form sample. This is a welcome development, not only for the research community, but also for many other users who require the ability to easily manipulate large amounts of data for multiple areas and population groups. The initial prototype 2005 ACS summary file has just been released and contains all of the detailed tables for every geographic area with 65,000 or more people; eventually, the ACS summary files will be released annually for each year’s 1-year, 3-year, and 5-year period estimates. Users have been invited by the Census Bureau to comment on the prototype summary file.15

Researchers who work with the new product will need to be cognizant of the larger sampling errors of the ACS tables compared with the 2000 long-form-sample tables and develop strategies for effective use of the ACS. Such strategies include combining data for census tracts and block groups into larger areas, collapsing data categories, and combining ACS summary files for nonoverlapping periods. The advantage of the ACS will be that researchers will not need to wait for 10 years to track trends in migration flows and other social, demographic, and economic phenomena.

3-E.2
Using PUMS Files for Research

Many researchers will turn to the ACS 1-year PUMS files for their analyses. The availability of PUMS files year after year will afford much

15

See http://www.census.gov/acs/www/Special/Alerts/Alert44.htm#News2, ACS Alert 44, December 28, 2006.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

flexibility to researchers. For example, in analyzing economic change, they can plan to use two (or more) PUMS files that coincide with different stages of a recession and subsequent recovery. Such an analysis was not possible with the once-every-10-years long-form-sample PUMS. The identification of PUMAs of about 100,000 population on each year’s ACS PUMS file also affords flexibility for analysis.

A drawback of the ACS 1-year PUMS files, as noted above for transportation, is the larger margins of errors compared with the 2000 long-form-sample 5 percent PUMS file. Many research uses of the PUMS data can benefit from combining two or more ACS PUMS files to increase the sample size for the analysis and thereby increase the precision of estimates. Researchers may be able to develop custom PUMS files for particular applications—such as a merged file of two or more 1-year PUMS for analyzing economic returns to education—that can be shared with other researchers.

Researchers will also need to grapple with the different reference periods for different respondents in the ACS PUMS files and develop appropriate analytical strategies. For income amounts for the previous 12 months, the Census Bureau will provide the reported amount, not adjusted for inflation. It will also provide a single inflation factor, which will adjust the values, on average, to July dollars for the latest year covered in a PUMS file (for example, 2005 for the 2005 PUMS file, which contains income reference periods that span January–December 2004 through December 2004–November 2005). A single inflation factor is used because the ACS PUMS files do not indicate the month of interview in order to protect confidentiality. Reconsideration of this decision and inclusion of the month (or season) of interview in the PUMS records and in selected summary tables would greatly increase the analytical value of the files (see Section 4-D.1).

3-F
MEDIA AND GENERAL PUBLIC USES

This section discusses using ACS profiles and rankings, which will be appealing products for occasional users and the media (refer back to Box 2-2). It also discusses comparisons of ACS estimates with other data sources, which can confuse users when differences between the ACS and the other data sources are not understood.

Journalists who frequently use statistical information to track local, regional, and national trends will use not only the ACS profiles and rankings, but also more detailed tables. They, like other involved federal, state, and local data users, will need support from the Census Bureau to understand how to properly apply the data (see Section 7-A).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
3-F.1
Using ACS Profiles and Rankings

ACS products that are likely to be of broad general interest include single-year profiles, providing key 1-year period estimates for governmental and statistical areas with at least 65,000 people; multiyear profiles, providing the same key 1-year period estimates for the current year and four prior years for areas with at least 65,000 people; and single-year ranking tables and charts comparing states and large cities on selected 1-year period estimates.16 These products will be timely and easy to reference. They will be the starting point for press releases from government officials and media articles describing what has occurred in a city, county, metropolitan area, or state since the year before and in comparison with other areas.

For trend analysis using multiyear profiles, public officials and the media must take care to avoid making too much of year-to-year differences that are within the margin of error (see Section 3-C.1.b). Just as the media have educated the public about the margin of error in public opinion polls, so should they take on the responsibility to educate readers about the margin of error from ACS estimates in profiles and other data products. The Census Bureau will provide margins of error for estimates in single-year profiles. In multiyear profiles, it will indicate estimates for each year that are statistically significantly different from the estimates for the current year.17

Similarly, for comparisons across areas using 1-year period estimates, public officials, the media, and readers must learn that, in most cases, the difference between, say, the city with the highest school-age poverty rate and the city with the next highest rate is not necessarily indicative of a real difference or even of the real ordering. In fact, the estimates for 5 or 10 of the cities with the highest rates may be not be statistically different, so that it is appropriate to say only that City A falls into the top, middle, or bottom group of cities rather than to assign it an individual rank-order number. Moreover, when the subsequent year’s period estimates are released, and City A has moved, for example, from number 1 to number 2, 3, 4, or 5 in school-age poverty, the reader should not conclude that school-age poverty has necessarily declined in City A relative to the other cities on the basis of one year’s difference in rankings.

Although sampling error affects such uses of the ACS data as trend analysis and comparative rankings, the regularly updated ACS estimates will be more helpful to users than the once-a-decade estimates from the

16

Multiyear profiles will be published for geographic areas defined according to the latest known boundaries for all years shown.

17

The Census Bureau provides 90 percent margins of error; for agreement with standard statistical practice, it should provide 95 percent margins of error instead (refer back to Box 2-5).

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

census long-form sample. From the example above, it is useful information to know what group of cities—top, middle, or bottom—a particular city is part of and to know, year to year, whether that city has remained about the same in relative ranking or, in contrast, has experienced a major change.

3-F.2
Comparisons with Other Data Sources

Often, estimates will be available not only from the ACS, but also from another data source, and the public, policy officials, and the media will want to know the reasons if the ACS and the other source do not agree. In fact, it is likely that differences will occur between estimates from two data sources because of differences in concepts and definitions, data collection procedures, and other aspects of the two sources.

In addition, users who want to compare 2005 ACS estimates for governmental or statistical areas with 65,000 to 250,000 people with estimates from an earlier period must use a different source—namely, the 2000 long-form sample—as their point of comparison. (The 2005 ACS estimates for areas with 250,000 or more people can be compared with estimates from the Census 2000 Supplementary Survey (C2SS) or any of the 2001–2004 ACS test surveys.) As described in Chapter 2, there are important differences between the long-form sample and the 2005 ACS, involving sample size, population covered, data collection mode, population controls, and others, so that assessment of changes between 2000 and 2005 must be made with great care. In the future, the yearly releases of ACS data will make it possible to assess trends using just the ACS, but the 2000 long-form sample will remain an important comparison source for small areas for some time to come.

With regard to comparisons between the ACS and another source for the same time period, an object lesson is afforded by experience in comparing state estimates of poverty from the CPS ASEC and the ACS supplementary surveys. National estimates from the CPS ASEC are the official poverty estimates for statistical use according to OMB Directive 14. To respond to user needs, the Census Bureau began publishing poverty estimates for states in 1990 from the CPS by averaging 2 and 3 years’ worth of estimates to improve precision. The Census Bureau has also published state poverty estimates from the C2SS and the 2001–2004 ACS test surveys and, now, the 2005 ACS.

Comparisons of trends from the CPS ASEC state poverty estimates averaged over 2 years with those from the C2SS and the ACS 2001–2004 test surveys revealed instances in which the two data sources did not agree on the poverty rate or the direction of change (increase or decrease in poverty). There are many reasons that may explain these differences:

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • Sampling error: A difference between the CPS ASEC and the ACS may not be significant because it falls within the margin of error. Each year’s CPS ASEC sample is about 77,000 housing units, compared with about 600,000 housing units for the C2SS and each of the 2001–2004 ACS test surveys and 2 million housing units for the full ACS (all figures are for responding units).

  • Population coverage: The CPS ASEC covers the civilian noninstitutional population, while the C2SS, the ACS test surveys, and the 2005 ACS cover the civilian and military household population. The 2006 ACS covers cover virtually the entire population, including civilian and military residents of households and institutional and noninstitutional group quarters (refer back to Table 2-1).

  • Residence rules: The CPS ASEC employs a usual residence rule, while the ACS employs a 2-month residence rule.

  • Reference periods: The CPS ASEC reference period for household composition—which is used to determine poverty thresholds—is February, March, or April (these are the months of interview each spring). The CPS ASEC reference period for income is the previous calendar year, which centers on July 1. The ACS reference period for household composition is the month of interview, which extends from January through December. Its reference period for income is the previous 12 months (extending from January of the preceding calendar year to November of the current calendar year) with adjustments made for inflation.

  • Mode of data collection: The CPS ASEC is conducted in person using CAPI for sample cases having their first interview and by telephone using CATI for sample cases having their second (or later) interview. The ACS is a mail survey with CATI and CAPI follow-up.

  • Imputation and weighting procedures: The CPS ASEC procedures for imputing an amount for unreported income are carried out on a national basis, whereas the ACS imputation procedures are carried out state by state, thereby capturing state differences in income patterns. The CPS ASEC population controls are applied for demographic population groups at the national level, and there are no housing unit controls, whereas the ACS population (and housing unit) controls are applied for counties or groups of small counties.

  • Question content: The CPS ASEC includes questions on 50 different sources of income; the ACS asks the standard long-form-sample questions, which include 8 sources of income. Past research has shown that asking more detailed questions elicits more complete reporting of income; however, the 2005 CPS ASEC (2004 income)

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

and the 2004 ACS supplementary survey produced about the same level of total income (Nelson, 2006). There were differences by source (the CPS estimated more wages and less self-employment income compared with the ACS), but the aggregates were very close.

Research is needed to understand the contributions of each of the above factors to differences between the CPS ASEC and the ACS. For users, now that the ACS is in full production with a vastly larger sample size than the CPS ASEC, it seems reasonable that they look to the ACS estimates for states and substate areas. However, users who want to analyze income by source and examine the correlates of income for population groups at the national level should stay with the CPS ASEC, which not only is the source of official income statistics, but also contains a wealth of variables to use in analysis.

3-G
WHAT HAPPENS IN A DECENNIAL YEAR?

An important element of the ACS design is to control each year’s estimates at the level of the county (or group of small counties) by total housing and by total population, categorized by age, sex, race, and Hispanic origin. The population control totals for each year are produced by updating the previous decennial census totals with administrative records on births, deaths, and net migration. The housing unit control totals for each year are produced by updating the previous census totals with housing permit records (see Chapter 5). The use of control totals is important to reduce sampling error in the estimates and to adjust the ACS estimates for possible undercoverage of housing and of the population, which may be particularly pronounced for some demographic groups.

The problem with the population control totals is that they become increasingly prone to error as each year passes from the previous census. While birth and death records are very accurate, there is considerable uncertainty about the quality of estimates of net migration, both net immigration from abroad, including illegal immigration, and net migration flows among counties. In the 2000 census, the estimate of the total U.S. population updated from the 1990 census was 1.8 million people fewer than the 2000 census count of 281.4 million people, and there were significant errors also in estimating the population of subnational areas. The national underestimate, which was particularly large for people ages 18–29 and for minorities, was attributed to an underestimation of illegal immigrants during the economic boom of the last half of the 1990s (National Research Council, 2004b:Table 5.1).

The postcensal housing unit controls are also subject to error, given

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

that housing permit data do not necessarily correspond with actual housing units constructed and occupied and given the problems of estimating demolitions and conversions to nonresidential use. The errors in the housing unit controls may well vary across geographical areas and may also cumulate over time.

While one cannot be sure what the magnitude or direction of the errors will be in the 2000 census-based estimates of the population and housing for 2010, one can be reasonably sure that there will be discrepancies between the estimates and the 2010 census counts and, furthermore, that the discrepancies will be greater for many counties and combinations of counties that are the basis for the ACS weighting controls. For the ACS, this means that there will be a discontinuity in many areas in totals for important demographic groups between 1-year period estimates for years preceding a census and for years including and following a census. This discontinuity will also exist for 3-year and 5-year period estimates between those that completely antedate a census year and those that include and follow a census year (for example, when comparing 5-year period estimates for 2005–2009 and 2010–2014). For 3-year and 5-year period estimates that span a census year (for example, a 5-year period estimate for 2008–2012), the Census Bureau plans to use an average of controls in which the population estimates for precensus years are adjusted to be consistent with the census counts.

One might consider that ACS estimates of percentages, as opposed to levels, would not be affected by the problem of differences in precensus and postcensus population controls. This will be the case, however, only if the discrepancies between the two sets of controls are relatively uniform by demographic category. If the discrepancies differ by category, which is likely, then the percentages will be affected as well (see Table 3-7 for an example).

There is no universal solution for the problem that will result from discrepancies between precensus and postcensus population controls. Users must address the situation for their applications and areas of interest, given that the problem will be more significant for some areas and population groups than others. The Census Bureau can help users in this regard by producing concurrent series of estimates that are based on precensus and postcensus controls. For example, the Census Bureau could produce two series of 1-year period estimates for, say, 2008–2010, in which the first series would use the 2000 census-based controls (the official series for those years), while the second series would backcast the 2010 census-based controls.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

TABLE 3-7 Hypothetical Effect of Decennial Census on ACS 1-Year Period Estimates, BIG CITY, 2008–2012

Year

Population Control

Estimated Number of Poor People (Equal to the Percent Poor Times the Control)

Non-Hispanic

Hispanic

Non-Hispanic (10% Poor)

Hispanic (20% Poor)

Total

ACS, Control Based on 2000 Census

 

 

 

 

 

2008

200,000

50,000

20,000

10,000

30,000

2009

200,000

52,000

20,000

10,400

30,400

2010

200,000

54,000

20,000

10,800

30,800

2010 Census

200,000

64,000

20,000

12,800

32,800

ACS, Control Based on 2010 Census

 

 

 

 

 

2011

200,000

67,000

20,000

13,400

33,400

2012

200,000

70,000

20,000

14,000

34,000

NOTES:

• The population controls for Hispanics and non-Hispanics (and for age, sex, and race groups) are implemented for the ACS by estimation area (county or group of small counties; this example assumes that BIG CITY is its own county). The controls are developed from the previous census updated with administrative records on births, deaths, and net migration.

• For ease of presentation, the example unrealistically assumes constant poverty rates from the ACS for the Hispanic population (20%) and the non-Hispanic population (10%) and that the city experienced no growth in the non-Hispanic population.

• For ease of presentation, the example unrealistically assumes that all of the error in the population controls prior to 2010 pertained to the Hispanic population.

What happened?

• The 2010 census gave the same non-Hispanic count (200,000) as the 2000-based population controls updated to 2010. But the 2010 census gave a different Hispanic count from the updated controls—64,000 instead of 54,000. Consequently, the controls were revised going forward from 2010.

What does the example tell the user?

• The number of Hispanics and non-Hispanics is determined by the population controls, while the percentage of poor people in each group is determined by the ACS. Hence, the number of poor people in each group is the product of the control and the estimated ACS percentage.

• The 2010 census results indicate that the Hispanic population and, consequently, the poverty population grew faster prior to 2010 than previously estimated, so that the ACS estimates of the number of poor Hispanics and total poor were too low for the years 2008–2010.

• Users will not know until the 2020 census is taken the extent of error that may occur in the 2010 census-based population estimates that are used as controls for the ACS in the period 2011–2020.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

3-H
PREPARING TO USE THE ACS

Users can ease the transition from using long-form-sample estimates to using ACS estimates for their applications by becoming knowledgeable about general guidelines for effective use of the ACS (3-H.1) and by taking concrete steps in advance to prepare for the time when 1-year, 3-year, and 5-year period estimates will be released each year (3-H.2).

3-H.1
General Guidelines for ACS Use

Abstracting from the specific applications discussed above, this section provides the panel’s basic general guidelines for appropriate use of ACS estimates.


a. Always examine margins of error before drawing conclusions from a set of estimates. Users should follow this practice for the long-form sample, the ACS, and any other survey on which they rely. More specifically:

  • When using ACS data to estimate a number or percentage for a single area or population group, such as a city or county, the ACS period estimate chosen—1 year, 3 years, or 5 years—should satisfy the precision requirements appropriate for the purpose for which it is being used (refer back to Table 2-7a).

  • Five-year period estimates will not be precise for estimates of small population groups (for example, poor school-age children, poor elderly people, minorities, high school dropouts) for areas with fewer than about 50,000 people, which includes most counties, cities, towns, townships, and school districts, as well as every census tract and block group (refer back to Table 2-7a). Consequently, users should work with 5-year period estimates for small areas only with great care.

  • When it is unduly burdensome to examine numerous individual margins of error—as, for example, when working with a large number of 5-year period estimates for multiple geographic areas—users should at least examine some of the individual error margins to check that the estimates are of adequate precision for their purpose.

b. Review the available information about nonsampling errors for estimates of interest and use this information in interpreting findings from the ACS.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • Research on nonsampling errors that may systematically bias survey estimates upwards or downwards is difficult to conduct, and the available information is rarely complete or definitive about the magnitude of the biases. Hence, users are rarely in a position to adjust estimates of interest to correct for nonsampling errors. Nonetheless, users should acknowledge known nonsampling errors in their uses of the ACS data.

  • As examples of possible biases in the ACS, a comparison of the C2SS with the 2000 long-form sample found significantly lower estimates of median income in the C2SS than in the long-form sample, while comparisons of the C2SS and the 2001–2003 ACS test surveys with the CPS consistently found significantly lower estimates of unemployment in the ACS surveys than in the CPS (see Section 2-B.2.e). Further research is needed to determine which survey is more accurate.

c. Carefully consider the pros and cons of alternative strategies for extracting value from ACS 5-year period estimates for very small areas, such as aggregating small-area estimates into estimates for larger, user-defined areas.

  • Large cities and counties should use ACS 5-year period estimates for census tracts and block groups as building blocks to define larger areas that are meaningful for analysis and for which 5-year period estimates are sufficiently precise. For example, a city might aggregate census tracts into several planning areas, or it might use combinations of block groups that do not necessarily respect census tract boundaries. Statistical mapping techniques may help identify which tracts and block groups would be most useful to combine into subareas for analyzing such phenomena as commuting patterns. For user-defined subareas, a city might ask the Census Bureau to develop 3-year period estimates for large population groups to obtain more information on trends.

  • Small governmental units may not be able to satisfy their data needs by aggregating 5-year period estimates into larger areas. However, with due care they may be able to work with 5-year period estimates for large population groups in their jurisdiction and 5-year period estimates for smaller groups for a larger area, such as their county, to assess changes in the composition of their own area. Small governmental units might also ask the Census Bureau to develop ACS estimates for their area for periods longer than 5 years.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • In addition to the basic strategies outlined above, in the future it may be possible to extract more value from the ACS 5-year period estimates by linking them with administrative records and other sources of local-area information (see Section 7-D.1).

d. When using ACS data to estimate shares of some total, compare estimates among areas or population groups, or assess trends over time, use ACS estimates that pertain to the same time period (1-year, 3-year, or 5-year) for all geographic areas or population groups that are being compared. Do not use a mixture of different period estimates.

  • For example, when determining the share of federal or state program funds that is to be allocated to each county in a state, the ACS estimates that are used will most likely need to be 5-year period estimates. The reason is that 1-year and 3-year period estimates are available for only about one-fourth of counties (refer back to Table 2-5), and it is not equitable to use a mixture of 1-year, 3-year, and 5-year period estimates to determine each county’s share of funds.

  • An exception to the need to use 5-year period estimates for fund allocations to counties is when a state has only a small number of counties that lack 1-year (or 3-year) period estimates. In this case, it may be appropriate to update the 5-year period estimates for the smaller counties by using information for larger areas, so that the equivalent of 1-year (or 3-year) period estimates can be used for all counties in the state. A simple procedure for accomplishing the updating is described in Section B.2 above; the use of this or another procedure depends on the reasonableness of the underlying assumptions.

  • As a matter of good practice, differences that are observed in comparing areas or population groups or in assessing trends over time should be evaluated not only for statistical significance, but also for substantive importance—that is, whether the differences are large enough to matter for policy, planning, or research purposes.

e. When analyzing trends over time for an area or population group, use ACS 1-year period estimates whenever they are available and sufficiently precise for the purpose of interest and be cognizant of changes in geographic area boundaries that may affect comparability. Keep in mind that the sampling error for the estimate of the difference between pairs of 1-year period estimates will be larger than the sampling error of either estimate.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

f. If only 3-year or 5-year period estimates are available or sufficiently precise, use them with care for analyzing trends over time for an area or population group. In general, avoid analyses of changes over time that are based on overlapping period estimates (for example, 5-year period estimates for 2010–2014 and 2011–2015).

  • It is not straightforward to interpret the meaning of differences that are observed between pairs of 3-year or 5-year period estimates: an observed difference may reflect a gradual change over the period, or it may reflect another pattern of change, such as stability in a characteristic followed by a sudden increase or decrease. Examining 1-year (or 3-year) period estimates for a larger area may help determine the appropriate interpretation of differences that are observed between pairs of 5-year period estimates for smaller areas within the larger area.

  • The less that pairs of 5-year (or 3-year) period estimates overlap in time, the more precise will be an estimate of differences between them—for example, a difference observed between estimates for 2010–2014 and 2015–2019 will be more precise than a difference observed between estimates for 2010–2014 and 2011–2015. Indeed, to obtain an acceptable level of precision for analysis of population groups, it will generally be necessary to wait for a second, nonoverlapping estimate to become available to compare to an earlier estimate.

g. Take advantage of the availability of 1-year and 3-year period estimates for PUMAs, which include about 100,000 people, to assist with analyses for smaller areas.

  • As one example, 5-year period estimates for small areas (census tracts in a city, towns in a county, small counties in a state) could be updated by adjusting their 5-year period estimates to the latest 1-year (or 3-year) period estimates for the applicable PUMA, as in Section B.2 above. Such adjustments need to be performed with care.

  • As another example, 1-year period estimates for large cities or counties could be compared with the PUMA estimates for the rest of the state (or the rest of the county in the case of a large city within a very large county).

h. Take care to label ACS estimates, including those for 1 year, 3 years, and 5 years, as period estimates.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • ACS 3-year and 5-year period estimates do not refer to a particular year, such as the end year or the middle year. They are period estimates—averages of characteristics over a 36-month or 60-month period—and should be labeled as such. Otherwise, readers may draw an incorrect inference—for example, assuming that a 5-year period estimate of 15 percent poverty is the rate for the end year, when the end-year rate could be considerably higher or lower.

  • ACS 1-year period estimates are also an average over 12 months (except for the special estimates released in June 2006 for January–August 2005 and September–December 2005 for areas affected by Hurricanes Katrina and Rita).

i. Use ACS 3- and 5-year period estimates for income, housing value, and housing costs with care. To compensate for the differing time periods for which dollar amounts are collected, those amounts are adjusted to a common calendar year by the change in the national CPI. This inflation adjustment expresses all of the reported dollar amounts in a comparable manner with regard to purchasing power as of the most recent calendar year in the period. However, the resulting estimates should not be interpreted as current-year estimates.


j. Use care in comparing ACS estimates with estimates from other data sources, including the 2000 long-form sample and other surveys, and be cognizant of the differences that could affect the comparisons. Such differences may include population coverage, sample size and design, reference periods, residence rules, and interview modes.

3-H.2
Suggestions for Users During the Ramp-up Period

In the next few years, users who plan to make extensive use of the ACS will have an opportunity to prepare for the full range of 1-year, 3-year, and 5-year data products that will be available beginning in 2010. It is important for users—including federal, state, and local agencies, and private organizations—not to stint attention or resources in order to make the most of this opportunity, so that they are well prepared to work with the full richness of the ACS data by 2010.

We outline below some of the steps that the technical staff of an agency can take to ensure that their agency is well prepared to work with the ACS data (see also the recommendations in ORC Macro, 2002:Ch. 10).


a. Steps to prepare agency management:

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
  • Schedule briefings with agency program managers to acquaint them with the ACS and how it will replace and improve on the once-a-decade long-form sample.

  • Make the case for sufficient resources from management to support planning for effective use of the various ACS products, which should save resources in the long run by minimizing inappropriate or ineffective use of the data.

  • As methods are developed to work with the ACS data for key applications, keep agency management informed of solutions.

  • Apprise management of the need to take such actions as seeking legislative authority or modification of regulations to permit ACS data to be used in place of long-form-sample data for particular applications.

b. Steps to determine which data and methods to use for particular applications:

  • Make use of information from the Census Bureau about the likely sampling error for different size areas to determine the most useful ACS estimates for the agency’s application(s). For example, if a city will have 1-year period estimates provided but their sampling error will be high, then the city may want to rely on the 3-year period estimates for planning and program applications.

  • Use the detailed 1-year period estimates that first became available in summer and fall 2006 to help develop the most useful profiles and other products to generate from the detailed 3-year and 5-year period estimates when they become available.

  • Use the training data sets released by the Census Bureau in spring 2007 of 1-year, 3-year, and 5-year period estimates fo 34 ACS test site counties for the years 1999–2005. Use of these data can provide valuable experience in working with multiple sets of estimates prior to the availability of the full set of 3-year and 5-year period estimates.

  • Determine the most appropriate geographic aggregations of 5-year period estimates for subareas of, say, a city or county. For example, the technical staff might divide a city of 500,000 into 10 service areas of approximately 50,000 population and aggregate 5-year period estimates for census tracts and block groups accordingly. Similarly, a large county might aggregate 5-year period estimates for townships into subcounty regions.

  • If resources permit, commission a detailed, comprehensive analysis of the alternatives for using various ACS data products for key

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

applications, similar to the study commissioned by HUD (ORC Macro, 2002).

  • Consider how often to analyze updated ACS estimates in light of the agency’s needs and resources. For example, it is unlikely that updates of small-area data analyses can be conducted more than twice a decade, nor may it be effective or efficient to do so.

  • Determine if any special tabulations will be needed from the Census Bureau, develop detailed specifications for them, and discuss feasibility and costs with the Census Bureau well in advance. For example, areas with large seasonal populations may want to request special tabulations.

  • Determine whether and how additional data sources may be helpful in some applications of ACS data. For example, a state might want to use administrative records information in conjunction with ACS estimates in fund allocation formulas.

  • Determine if model-based or composite estimates that are developed from the ACS and other data sources by statistical agencies could support particular applications, thereby saving on the program agency’s technical resources.

  • Request that the Census Bureau inform users of helpful guides that are developed by State Data Centers and other organizations and individuals to assist users—for example, the recent publication, American Community Survey Data for Community Planning (Taeuber, 2006).

c. Steps to work with public officials, the media, and other constituents:

  • Develop templates for appropriate interpretative language to use in press releases and talking points about each summer’s issuance of the latest ACS estimates from the Census Bureau. Given that the media and public officials will inevitably want to compare trends across time and levels across areas using the most recent estimates regardless of their precision, the agency technical staff should develop suitably cautionary language to include in statements by public officials and in speaking with the media.

  • Develop standard formats for tables to provide to constituent groups (for example, neighborhood advisory commissions or council members in a city or county). Be sure to include appropriate explanatory material about sampling error and other aspects of the data.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

d. Steps to look toward the future:

  • Keep up to date with Census Bureau information on the ACS web site, such as users’ guides and design and methodology reports, and take advantage of training opportunities afforded by the Census Bureau, state data centers, and other organizations.

  • Feed back questions, concerns, and data needs to the state data centers and to the Census Bureau. On one hand, be cognizant that the Census Bureau has a heavy workload in collecting, processing, and disseminating the continuous ACS, but, on the other hand, remind the Census Bureau that the ACS must be an evolving data system that responds to user needs.

  • Liaise with other users with similar interests to develop and evaluate strategies for effective use of ACS data products, and put forward coordinated requests for new and improved data products, training materials, and other support from the Census Bureau.

  • If there is a need for new or modified questions, work with the Census Bureau and stakeholders to determine what is the minimum set of changes that would serve the purpose. The Census Bureau has a protocol for the testing that must be undertaken before a new question can be added to the ACS (see Section 7-C.2).

  • Similarly, work with the Census Bureau and stakeholders to adjust geographic boundaries for census tracts and block groups in ways that reflect population change but minimize discontinuities in local geographic boundaries over time. If, for example, most changes to these small areas involve splitting them to reflect population growth (or, alternatively, combining them to reflect population decline), then it will be easier to use successive 5-year period estimates.

  • Participate in forums in which users share their experiences with analysis and presentation techniques that make effective use of the ACS data for a range of applications.

In conclusion, the ACS will offer not only significant challenges to data users, but also significant benefits. Having more timely and up-to-date information that is likely of higher quality will benefit all applications that previously used the long-form-sample estimates. In the future, there will be opportunities for new uses of the ACS that would never be possible with the long-form sample. Users should take steps during the ramp-up period to prepare for the ACS, anticipate problems, and work together and with the Census Bureau on solutions.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×

This page intentionally left blank.

Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 77
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 78
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 79
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 80
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 81
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 82
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 83
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 84
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 85
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 86
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 87
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 88
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 89
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 90
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 91
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 92
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 93
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 94
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 95
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 96
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 97
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 98
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 99
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 100
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 101
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 102
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 103
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 104
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 105
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 106
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 107
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 108
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 109
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 110
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 111
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 112
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 113
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 114
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 115
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 116
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 117
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 118
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 119
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 120
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 121
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 122
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 123
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 124
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 125
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 126
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 127
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 128
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 129
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 130
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 131
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 132
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 133
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 134
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 135
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 136
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 137
Suggested Citation:"3 Working with the ACS: Guidance for Users." National Research Council. 2007. Using the American Community Survey: Benefits and Challenges. Washington, DC: The National Academies Press. doi: 10.17226/11901.
×
Page 138
Next: PART II: Technical Issues, 4 Sample Design and Survey Operations »
Using the American Community Survey: Benefits and Challenges Get This Book
×
Buy Paperback | $80.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

The American Community Survey (ACS) is a major new initiative from the U.S. Census Bureau designed to provide continuously updated information on the numbers and characteristics of the nation’s people and housing. It replaces the “long form” of the decennial census. Using the American Community Survey covers the basics of how the ACS design and operations differ from the long-form sample; using the ACS for such applications as formula allocation of federal and state funds, transportation planning, and public information; and challenges in working with ACS estimates that cover periods of 12, 36, or 60 months depending on the population size of an area.

This book also recommends priority areas for continued research and development by the U.S. Census Bureau to guide the evolution of the ACS, and provides detailed, comprehensive analysis and guidance for users in federal, state, and local government agencies, academia, and media.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    Switch between the Original Pages, where you can read the report as it appeared in print, and Text Pages for the web version, where you can highlight and search the text.

    « Back Next »
  6. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  7. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  8. ×

    View our suggested citation for this chapter.

    « Back Next »
  9. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!