2

Needs for Small-Area Income and Poverty Estimates

Regularly updated small-area estimates of income and poverty are increasingly in demand for federal and state programs. Formulas that include such estimates are used to allocate billions of dollars each year to states and localities (U.S. General Accounting Office, 1999), and the estimates have other program uses as well. These uses place significant requirements on estimates, including requirements for geographic and population detail, timeliness of production, and accuracy of measurement. No estimates will perfectly meet all requirements. Users, including government agencies that administer programs by using estimates and policy makers that legislate program uses of estimates, need to be aware of the strengths and weaknesses of estimates for their purposes.

In this chapter we describe the growing demand for small-area income and poverty estimates, identify key requirements for estimates from the perspective of program uses, and assess in general terms the ability of alternative data sources to satisfy these requirements. The chapter is addressed primarily to users, but also to suppliers of estimates who should be knowledgeable of the need for their product and the implications for estimation methodology.

PROGRAM TRENDS

The use of small-area statistics for such purposes as allocation of federal funds to states and localities has a long history (see Anderson, 1988:178-179, 203-205). Financial grants-in-aid were developed in the late



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond 2 Needs for Small-Area Income and Poverty Estimates Regularly updated small-area estimates of income and poverty are increasingly in demand for federal and state programs. Formulas that include such estimates are used to allocate billions of dollars each year to states and localities (U.S. General Accounting Office, 1999), and the estimates have other program uses as well. These uses place significant requirements on estimates, including requirements for geographic and population detail, timeliness of production, and accuracy of measurement. No estimates will perfectly meet all requirements. Users, including government agencies that administer programs by using estimates and policy makers that legislate program uses of estimates, need to be aware of the strengths and weaknesses of estimates for their purposes. In this chapter we describe the growing demand for small-area income and poverty estimates, identify key requirements for estimates from the perspective of program uses, and assess in general terms the ability of alternative data sources to satisfy these requirements. The chapter is addressed primarily to users, but also to suppliers of estimates who should be knowledgeable of the need for their product and the implications for estimation methodology. PROGRAM TRENDS The use of small-area statistics for such purposes as allocation of federal funds to states and localities has a long history (see Anderson, 1988:178-179, 203-205). Financial grants-in-aid were developed in the late

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond nineteenth century as an alternative to land grants, which Congress had historically used to encourage state and even private development. The first grant-in-aid was adopted in 1887 in the Hatch Act, which provided a small amount of funds to each state for agricultural experiment stations. By the early 1920s, grants covered highways, vocational education, agricultural extension work, conservation programs, and public health. Generally, the formulas were simple, using such measures as total population, area, or road mileage. Beginning with the New Deal in the 1930s, grant-in-aid programs were more and more seen as a way to help equalize the national impact of programs by accounting for differences among states in their wealth and fiscal capacity. Also, an increasing number of programs specifically targeted low-income areas as a way to redistribute national wealth to address social problems. In response, allocation formulas were written to require estimates for such factors as per capita income, poverty rate, unemployment rate, or proportion of substandard housing; all of these are more difficult to measure than total population. In addition, some programs provided allocations directly to cities and other local areas, which necessitated estimates for areas smaller than states. Formulas also became more complex, not only by including multiple factors, but also by incorporating such provisions as thresholds for eligibility, minimum allocation amounts, and maintaining a percentage of prior year grant amounts for areas that would otherwise see a decrease in funding (“hold harmless”). These kinds of provisions generally required more accurate estimates. By the 1990s, an estimated $180 billion of federal funds were allocated each year to states and localities on the basis of formulas that included one or more factors requiring estimates for population groups (e.g., total population, elderly, children; see U.S. Census Bureau, 1999d; see also U.S. General Accounting Office, 1999). Although no precise dollar figure is available, more than $130 billion of these funds were allocated on the basis of formulas that specifically included small-area income or poverty estimates as a factor. Estimates were obtained from such sources as the decennial census long-form sample income information; administrative record counts of participants in particular social programs; the per capita income estimates developed from administrative records, censuses, and surveys by the Bureau of Economic Analysis (BEA); and, more recently, the estimates from the Census Bureau's Small Area Income and Poverty Estimates (SAIPE) Program developed by applying statistical methods to data from several sources. The largest federal program that uses small-area income or poverty estimates for fund allocation is Medicaid, which has a matching formula for reimbursing state expenditures, the Federal Medical Assistance Per-

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond centage (FMAP). Other programs, including Adoption Assistance and Foster Care, also use FMAP. To determine what percentage of state expenditures will be reimbursed (the percentage is constrained to be no less than 50% and no more than 83%), the FMAP formula takes account of the ratio of state per capita income to total U.S. per capita income–the higher the ratio, the smaller the amount reimbursed. The per capita income estimates for each state are 3-year averages from BEA. Some federal programs also use small-area income or poverty estimates to determine eligibility to apply for project grants or other benefits. For example, to receive tax benefits from the U.S. Department of Commerce Empowerment Zones Program, rural and urban areas that are designated for the program must demonstrate a poverty rate of not less than 20 percent in each census tract in the area and, for at least 90 percent of the census tracts in the area, the poverty rate must not be less than 25 percent.1 Looking to the future, it seems likely that an increasing number of federal programs will provide funding on the basis of allocation formulas that include small-area poverty or income estimates because of the trend in social welfare policy of replacing individual entitlement programs with block grants to states and localities. For example, the new block grant Welfare to Work Program, funded beginning in fiscal year 1998 at $1.5 billion, requires that 75 percent of the funds be allocated to states according to the state share of the national number of poor persons and the state share of the national number of adult recipients of Temporary Assistance to Needy Families (TANF), with each factor equally weighted. In turn, states must suballocate 85 percent of the federal funds they receive to service delivery areas, which can be a county, city, consortium of counties or cities, or, in some cases, part of a large city.2 Half of the suballocations must be made according to the number of persons in poverty in excess of 7.5 percent of the population in each service delivery area; the other half of the suballocations can be made according to the number of adult TANF recipients and the number of unemployed people in the area. The Welfare to Work Program was enacted after welfare reform, in which the 1996 Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) abolished the Aid to Families with Dependent Chil- 1   Poverty rate estimates for census tracts are from the census long-form sample. Empowerment zone areas must also meet several other criteria, such as how much of the area is zoned for commercial and industrial use. States and localities nominate empowerment zone areas; the final designation is made by the U.S. Secretary of Housing and Urban Development (for urban areas) and the U.S. Secretary of Agriculture (for rural areas). 2   States must match every $2 of federal funding with $1 of state funding.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond dren (AFDC) entitlement program and replaced it with the block grant TANF program. The TANF formula does not include poverty or income as a factor (it essentially allocates to states the amounts they had received in federal AFDC matching funds at an earlier point), but PRWORA requires state-level child poverty rates for administering TANF. States must measure year-to-year changes in their child poverty rates and develop an action plan if the child poverty rate can be determined to have increased by more than 5 percent due to the operation of the TANF program. For federal formula allocation programs in education, two recent trends have added to the requirements for small-area poverty estimates. First is a trend–evident in other programs as well–toward requiring estimates of child poverty more frequently than can be provided by the decennial census. Second is a trend toward requiring estimates for very small areas, namely, school districts, to permit direct allocation of federal funds to those areas. Developing estimates for school districts is particularly difficult because of the small population size of most districts and because school district boundaries in many instances cross county lines and can and often do change over time. The Improving America's School Act of 1994 reflected both of these trends in mandating changes for Title I of the Elementary and Secondary Education Act, which provides more than $7 billion each year for programs to help educationally disadvantaged children. Under the act's provisions, the U.S. Department of Education must allocate funds directly to school districts by using estimates of poor school-age children that are updated every 2 years, provided the estimates are found to be sufficiently reliable for this purpose. (Previously, the department allocated funds to counties on the basis of estimates from the most recent decennial census, and states then suballocated the county funds to school districts.) A new program in fiscal 1999 for class size reduction will allocate $1.2 billion to school districts in 2000 according to the same school district estimates of poor school-age children that are used for Title I allocations. These estimates are from the SAIPE Program. In addition to federal program uses of small-area income and poverty estimates, state governments increasingly use such estimates for allocating state funds to local areas (see Midwest Research Institute, 1999). These uses are in addition to the requirements in some federal programs for states to suballocate federal funds to localities by means of a formula that includes income or poverty estimates. Many state allocation formulas that include income or poverty are education programs that are primarily targeted at school districts. Some states also use income or poverty-based allocation formulas for allocating social service and health program funds to counties or other areas. Data sources that states use to provide estimates for state allocation formulas include the decennial census and state

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond administrative records (e.g., the number of children approved for free or reduced-price school lunches or state income tax data). Table 2-1 summarizes key features of selected federal and state allocation formulas that use small-area income or poverty estimates as a factor, chosen to illustrate the variety of program areas, funding levels, and provisions of current allocation formulas. As the table shows, a wide variety of programs allocate funds on the basis of small-area income and poverty estimates, including child care, community development, education, job training, nutrition, and public health. REQUIREMENTS FOR ESTIMATES The use of small-area income and poverty estimates for allocating funds or related program purposes imposes significant requirements for the estimates if they are to satisfy the intentions of the framers of program legislation.3 Ideally, requirements for estimates include the desired concept or definition of poverty or income measured, level of geographic detail, level of population or demographic detail, timeliness of production and updating, and accuracy of measurement. The cost of producing estimates is also a consideration. In practice, it is rare that any one data source or even a combination of data sources can provide estimates that satisfy all requirements: for example, one source may provide an outdated measure of the specified poverty concept, while another source may provide an updated measure for a concept that is only partly related to poverty. This situation does not mean that users should decline to target programs on the basis of small-area income or poverty estimates. Rather, they should select sources of estimates with as much knowledge as possible of the strengths and weaknesses of each source and the implications for the resulting fund allocations or other program uses of the estimates. Users should also consider interactions of estimates with formula features. It may be that altering a formula provision would result in more appropriate use of available estimates (see Chapter 6). In the section below we briefly discuss each type of requirement for estimates and indicate some of the tradeoffs involved. 3   For discussion purposes, we take as given that the elements of allocation formulas and their intended purposes are clear, although this may be far from true in many cases. For example, the concept underlying a formula factor may not be well specified.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond Concept of Poverty or Income Many programs that use small-area estimates of poverty for fund allocation or other purposes specifically target those in poverty according to the official poverty measure (i.e., those in families with before-tax money incomes below the official poverty level). Alternatively, some programs target those in families with incomes that are a multiple of the official poverty line (e.g., 125% or 185%). Other programs use a different standard of need, such as those with incomes below 70 percent of the lower living standard income level defined by the U.S. Department of Labor.4 Other programs have a less specific definition, targeting those with “low incomes” or those who are “needy.” Income concepts used for fund allocation often specify per capita income or, sometimes, median income. Given a specific definition of poverty (or income) in an allocation formula, the small-area estimates used for allocation should measure that concept. To the extent that the estimates measure a somewhat different concept, there may be a bias that results in a persistent misallocation of the funds.5 As an example, when Title I education funds were allocated by a two-stage process, many states obtained approval from the U.S. Department of Education to suballocate the county amounts to school districts on the basis of the number of children approved for free or free and reduced-price school lunches in each district. However, school lunch counts include children in families with incomes as high as 130 percent of poverty (free lunch) or as high as 185 percent of poverty (reduced-price lunch).6 Hence, some school districts that received Title I concentration grant funds because they had sufficient proportions (or numbers) of children approved for free or free and reduced-price school lunches would not have received funds on the basis of an estimate of children in families with incomes below 100 percent of poverty.7 Given a fixed total amount, allocating funds to these districts meant less money for other, possibly poorer districts. 4   Lower living standard income levels are published by the Employment and Training Administration for 25 metropolitan areas and for metropolitan and nonmetropolitan components of the four census regions, Alaska, and Hawaii. The levels represent the Bureau of Labor Statistics lower level family budgets, developed for 1967 on the basis of 1960-1961 Consumer Expenditure Survey data and last published for 1981, updated for price changes. Seventy percent of these levels for a family of four ranges from about 100 to 166 percent of the official poverty level (which does not vary by area). 5   See “Accuracy of Measurement” below for definitions of bias and other types of errors in estimates. 6   Almost twice as many children are in families with incomes below 185 percent of the poverty threshold (38%) as are in families with incomes below 100 percent of the poverty threshold (20%). About 26 percent of children are in families with incomes below 130 percent of the poverty threshold. (Data from panel tabulations of the March Current Population Survey [CPS] for income years 1994-1996.) 7   Title I concentration grants allocate funds only to school districts with large numbers or proportions of poor school-age children, in contrast to Title I basic grants, for which the thresholds for eligibility to receive funds are low (see Table 2-1).

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond This page in the original is blank.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond TABLE 2-1 Key Features of Selected Federal and State Allocation Formulas that Use Small-Area Income and Poverty Estimates FEDERAL FORMULAS Agency and Program Name (Amount Allocated per Year) Areas to Which Funds Allocated Income or Poverty Estimates Required (Other Factors if Known) Data Source for Estimates U.S. Department of Agriculture Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) ($3 billion in 1999) States States receive allotments equal to previous year; up to 80 percent of remaining funds are allocated as inflation adjustments; remaining funds are allocated based on state share of national number of children in families below 185 percent of poverty Model-based estimates using March CPS, decennial census, and administrative records (prior to fiscal 1995, used decennial census) Rural Housing Preservation Grants ($9 million in 1999) States Formula assigns one-third weight to state share of: (1) total rural population, (2) total rural occupied substandard housing units, (3) total rural families with income below poverty Decennial census U.S. Department of Education Title I of Elementary and Secondary Education Act ($7.6 billion for 1999-2000 school year)a School districts (prior to 1999, the Department of Education allocated to counties; states suballocated to school districts) Basic grants allocated to school districts with at least 10 formula-eligible children and more than 2% formula-eligible children; concentration grants allocated to school districts with more than 15% or more than 6,500 formula-eligible children (principally school-age children in poor families) Model-based SAIPE estimates (prior to 1997-1998 school year, used decennial census estimates) U.S. Department of Health and Human Services Child Care and Development Block Grant ($1 billion in 1999) States Formula considers number of children under age 5, number of children receiving assistance, through the National School Lunch Program, and state per capita income Population estimates, administrative records, BEA income estimates Title V Maternal and Child Health Services Block Grant ($576 million in 1999) States Formula includes number of poor children under age 18; some states also use poverty measures to suballocate funds to counties Decennial census Medicaid (reimbursement of state expenditures) ($200 billion, federal and state, in 1999) States Formula includes the ratio of state per capita income to U.S. per capita income (the higher the ratio, the smaller the amount reimbursed; between 50% and 83% of state expenditures) BEA income estimates

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond This page in the original is blank.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond U.S. Department of Housing and Urban Development Community Development Block Grants ($1.2 billion in 1999 to states to use for non-entitlement communities; $2.9 billion in 1999 directly to entitlement communities, which are metropolitan cities and counties) Cities with 50,000+ population, metropolitan counties with 200,000+ population, some nonmetropolitan areas States receive funds equaling the greater of two formulas. First formula includes factors for: (1) population, (2) poor households, (3) overcrowded housing units (1.01+ persons per room) in the balance of the state outside entitlement communities (factors weighted at 0.25, 0.50, 0.25, respectively). Second formula includes factors for: (1) population, (2) poor households, (3) number of housing units built before 1940 (factors weighted at 0.2, 0.3, and 0.5, respectively). The formulas for entitlement communities include poor households and several other factors. Decennial census HOME Investment Partnership Program ($1.5 billion in 1999)States, cities, urban counties, and consortia of local governments States, cities, urban counties, and consortia of local governments Formula includes six factors and weights: (1) 10% weight on vacancy-adjusted rental units with household head in poverty; (2) 20% weight on occupied rental units with defined housing problems; (3) 20% weight on rental units built before 1950 occupied by poor families; (5) 20% weight on number of poor families; (6) 10% weight on population adjusted by the ratio of net per capita income to U.S. net per capita income. Decennial census U.S. Department of Labor Job Training Partnership Act Title II-A (adult) ($955 million in 1999) Service delivery areas (one or more counties or cities of 200,000+ population) Formula includes number of persons aged 22-72 in families with income not more than higher of OMB poverty line or 70% of lower living standard income level. Decennial census (special tabulation) Job Training Partnership Act Title II-B (summer youth) and Title II-C (youth training) ($871 million and $130 million, respectively, in 1999) Service delivery areas (as defined above) Formula includes number of persons aged 16-21 in families with income not more than higher of OMB poverty line or 70% of lower living standard income level. Decennial census (special tabulation)

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond This page in the original is blank.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond estimated variance) into a single accuracy measure, mean square error. As its name indicates, mean square error measures the average (squared) difference between an estimated value and the true value. To understand the effects of estimates on fund allocation and other program uses, it is important to assess bias and variance separately to the extent possible. Not only are their effects likely to differ (discussed below), but so also are ways to reduce those effects. As an example, expanding sample size is a way to reduce sampling error in a survey, although at likely considerable expense. Another way to reduce survey sampling error is to average estimates for several years, but this procedure may introduce bias. Reduction of bias requires different strategies–for example, rewording of questions in a survey or adjustment of survey estimates by using other data sources. Persistent bias is of serious concern for small-area estimates of poverty or income. Such bias means that certain areas may consistently receive more or less funding than what they would receive with unbiased estimates. Bias can occur because of some of the characteristics of estimates discussed above, for example, if the estimates measure a concept of poverty or income that is not the same as the concept specified in the formula, or if the estimates are very out of date and do not reflect changes since the reference year for the data. But bias can also occur even if the data used for estimates are otherwise timely and measuring the appropriate concept. Regularly conducted household surveys (such as the March CPS) that are designed to collect information with which to estimate the official concept of poverty can result in biased estimates due to measurement problems (see Chapter 4). For example, two sources of downward bias (i.e., underestimation) in poverty rates from household surveys are that they tend not to cover lower income groups as completely as middle- and higher-income groups, and that they have disproportionately more nonrespondents among lower income people. A source of upward bias (i.e., overestimation) in survey poverty rates is that respondents tend to underreport their income. For administrative records that are otherwise appropriate to use for income and poverty estimates, there can be biases across areas due to differences in participation rates and other factors of program design and administration (see Chapter 5). If the resulting overall bias in estimates, up or down, is the same for all areas, then the effects on allocations may not be great if the formula does not also contain such provisions as thresholds for eligibility. Much more likely, however, is that biases will affect some types of areas more than others, affecting allocations even in formulas that simply distribute shares of a fixed amount to all areas. Variance is generally of less concern than bias because it is expected

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond that errors will balance out over time–for example, one state or county may have its poverty rate overestimated in one year and underestimated the next year and vice versa for another state or county. However, formula provisions, such as thresholds, can interact with the variability in estimates in ways that disproportionately favor or disfavor particular areas. Moreover, the greater the variability, the greater these kinds of effects (see Chapter 6). DATA SOURCES Current and planned sources of data for developing small-area income and poverty estimates include: the decennial census; household surveys, including the March CPS, the Survey of Income and Program Participation (SIPP), and the planned American Community Survey (ACS); administrative data, ranging from school lunch counts to federal and state income tax records; and programs for deriving estimates from multiple sources, including the BEA program for estimating total and per capita personal income for states and counties and SAIPE. The strengths and weaknesses of these data sources are briefly described below in terms of the requirements for income and poverty estimates, further illustrating the tradeoffs involved in using particular sources of estimates for program purposes. Chapter 4 and Chapter 5 provide more detailed discussion of household surveys and administrative records, respectively, and the role they can play in improving SAIPE estimates. Decennial Census The decennial census long form, which was sent to 15-18 million households in the 1990 and 2000 censuses, contains the small number of questions that are asked of all households on the short form and other questions that are unique to the long form. The additional information collected includes annual income amounts for about seven sources and other characteristics that permit estimating income and poverty for a wide range of population groups and geographic areas. Census long-form income and poverty estimates, classified by age and other characteristics, are routinely provided for the income reference year (1989 for the 1990 census and 1999 for the 2000 census) for states, counties, towns and townships, places, census tracts, and block groups. Estimates are also often prepared for other kinds of small areas, such as school districts, by aggregating the estimates for individual census blocks. As noted above, these estimates are usually released 2 to 3 years after the census is completed. The long-form census survey has major strengths as a source of income and poverty estimates: it measures official concepts of household

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond poverty and income;8 it collects a range of characteristics for developing estimates for specific population groups; and it provides estimates with low sampling error for many subnational areas. The long-form survey also has important drawbacks: it is conducted only once every 10 years; it is believed to measure income and poverty less well than the March CPS and SIPP (although more research is needed to compare measurement error in the census with household surveys–see Chapter 4); and long-form estimates for very small areas can have high sampling variability. Generally speaking, long-form estimates for areas with fewer than about 20,000 people have relatively large sampling errors, and there are many areas smaller than this size for which estimates are needed: 47 percent of counties and 82 percent of school districts are below 20,000 population (although these areas account for small proportions of people–see Chapter 4). If the American Community Survey is implemented as planned, there may be no long-form survey in the 2010 or subsequent censuses. Household Surveys March Current Population Survey The March income supplement is an annual addition to the monthly CPS labor force survey that currently has a sample size of about 50,000 households. (The sample size may increase in the future.) The March supplement obtains detailed responses on sources of income in the preceding calendar year (about 30 separate questions) and on many other characteristics that permit estimating a range of income and poverty statistics. The March supplement is the source of official income and poverty estimates for the preceding year that are published each fall for the United States as a whole and certain population groups. The March CPS has potential advantages for small-area income and poverty estimates because it is conducted annually and obtains extensive income data and other characteristics. However, at present, the survey 's relatively small sample size rules out its use to produce reliable direct estimates for subnational areas, except for the largest 10 or 12 states and a handful of very large counties (see Chapter 4). Indeed, the survey includes no households in the sample from which to develop direct esti 8   That is, the census uses the official poverty thresholds for different size and type families and compares them to the official income definition, which is before-tax money income. However, it does not provide precisely the same estimates as the March CPS, which is the official source of income and poverty statistics, because of differences in questionnaires, data collection procedures, and other features of the two surveys.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond mates for about two-thirds of the country's counties. The only way to use the March CPS data for small-area income and poverty estimates is in statistical models, such as the SAIPE state and county models discussed below, that combine information from multiple sources. Survey of Income and Program Participation The SIPP survey began in 1983 as a series of panels, each of which followed a sample of household members for about 32 months, with interviews every 4 months. The 1996 SIPP panel followed the members of about 37,000 originally sampled households over a 4-year period from 1996 to early 2000. A new SIPP panel of about the same size is expected to follow sample members for a 3-year period beginning in 2001. Design changes are being considered for SIPP that could make it possible for the survey to provide official income and poverty statistics in place of the March CPS. SIPP obtains even more detailed information about income and population characteristics than the March CPS, asking about 60 questions on sources of income. The survey obtains more complete reporting of many income sources than the March CPS, and two Committee on National Statistics reports have recommended that SIPP become the basis of official income and poverty statistics (National Research Council, 1993, 1995a). To date, however, such a role for SIPP has not proved practicable. One reason is the time to process the data, which has typically delayed release of data files for several years after the income reference period. Another reason relates to the longitudinal nature of the survey. Sizable proportions of households drop out of each panel before it is completed, and research shows that the dropouts differ importantly from full-panel respondents in their income and poverty characteristics. Funding is being sought for a design that would introduce a new 3-year SIPP panel every year, so that several panels would be in progress at the same time. This design would make it possible to develop annual poverty and income statistics that do not have an increasing level of error due to dropouts over the course of a single panel. Like the March CPS, SIPP cannot provide reliable direct income and poverty estimates for subnational areas because of its relatively small sample size (smaller than in the CPS), and, unlike the CPS, it is not currently designed to provide reliable estimates at the state level even for the largest states. There is a potential to use SIPP in models if these problems are resolved.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond American Community Survey The ACS is intended to be a large-scale, monthly sample survey of U.S. households, similar to the census long-form survey in content and administration but operating continuously. The ACS is now (1996-2002) in a design and testing phase. If funds are appropriated, it will become operational in 2003, sampling 250,000 households each month, spread across the nation, so that every county, school district, and other small area will have sample households. The annual sample will be about 3 million households; over a 5-year period, the ACS sample size will cumulate to about 15 million households. This sample size is only somewhat smaller than the expected 2000 census long-form sample size, although the ACS sample size will be reduced for analysis because only one-third of households that do not respond to a mail questionnaire or telephone follow-up will be followed up in person. 9 If it is implemented as planned, the ACS will have important advantages for small-area income and poverty estimates. It will measure current official concepts of income and poverty, collect a range of population characteristics permitting estimates for particular groups, provide data at frequent intervals, and have much larger sample sizes (when cumulated to 1 or more years) than any existing household survey. Also, its design will provide sample households in every state and county each year. However, the ACS may have important disadvantages as well. Although research will be needed to evaluate income measurements across surveys, it is likely that the ACS will prove to be a relatively crude instrument for measuring income and poverty in comparison with the March CPS and SIPP. One reason is that the ACS questionnaire, like the long form, contains a small number of questions on income. Also, the “rolling” nature of the ACS may create measurement problems. Thus, the questionnaire will ask about income in the past 12 months and not the more natural reference period of the past calendar year (see Chapter 4). In addition, although the ACS will have a much larger sample size than other household surveys, direct estimates of income and poverty will still not be reliable for many small areas, such as school districts, even if the data are cumulated for as many as 5 years. Moreover, cumulating data for multiple years could lead to biases that would affect program uses. For example, an allocation formula that targeted poor areas might, using 5-year poverty estimates, give the same allocation to an area that had experienced a pronounced rise in poverty over those 5 years as to an 9   For the census, the goal is 100 percent folllow-up for nonrespondents.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond area that had experienced a pronounced decline if the two areas had the same average poverty estimate over the 5-year period. Moving averages, in which 5-year estimates are produced annually that dropped the earliest year and added the most recent year, would gradually direct funds toward areas with increasing need and away from areas with declining need. However, the adjustment might be more gradual than intended for some programs, unless some other form of weighting were used (e.g., weighting recent years more heavily than earlier years). The potential of the ACS for small-area income and poverty estimates warrants careful consideration by the users and suppliers of estimates. Such assessments should include both its role for direct estimates and its use with other data for model-based estimates. Administrative Records Many federal and state programs include data from administrative records as factors in formulas for allocating funds to states and local areas. Examples include the number of children approved for free or reduced-price school lunches; participants in the TANF program; post-secondary students who receive Pell Grants or other assistance; people who receive food stamps; children enrolled in Head Start; and people in families with low income, based on their tax returns (see Midwest Research Institute, 1999). Administrative records vary in how much information they provide on the characteristics of people, and the information recorded may change over time in response to program administration needs. For fund allocation purposes, administrative data may be included in a formula because of an intent to target funds to particular groups of people receiving related benefits. Often, however, administrative data are used in formulas as a proxy for poverty estimates that are not available or that are perceived to have drawbacks in comparison with the administrative data (e.g., lack of timeliness). The use of administrative data as a proxy for poverty is particularly common in states for suballocating federal funds or allocating their own funds to localities. As a proxy for poverty, administrative counts of program beneficiaries (e.g., food stamp or school lunch recipients) have advantages, particularly for use by states: they are often readily available at little added expense for such areas as counties and school districts; they are regularly updated; and they are not subject to variability from sampling error, although they may have other sources of random error (e.g., errors in data entry and updating). They often have “face validity ”: comments from state agencies suggest that school lunch counts are viewed as good proxies for school district estimates of poor children and are preferred to out-

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond of-date census estimates and to SAIPE estimates, when those estimates do not match the school lunch counts (Midwest Research Institute, 1999). Administrative counts of program beneficiaries may not be a good proxy for differences in poverty across areas because of differences in program administration and participation. For food stamps, for example, eligibility requirements are similar to official poverty concepts–generally, eligible households must have gross income below 130 percent of the poverty level and net income after certain deductions below 100 percent of the poverty level. However, data for states and counties are counts of people actually receiving food stamps, not of people who are eligible to apply for them, and research has shown that the proportion of the eligible population enrolling in the program varies across areas. Reasons for such variations include differences in program outreach and other features of program administration, as well as differences in the willingness of eligible people to sign up for benefits. Whatever the reasons, differences in participation rates mean that food stamp recipient populations may not be a consistent indicator of poor populations across areas. Moreover, changes in program features may affect how consistently recipient populations relate to poor populations over time. For example, the 1996 welfare reform legislation restricted food stamp eligibility for certain groups, such as recent immigrants, who are distributed unevenly across geographic areas, and may have had other effects on both interarea and intertemporal comparability as well (see Chapter 5). Other administrative programs do not relate as closely in their eligibility requirements to official poverty concepts as the Food Stamp Program. For example, the eligibility standards for the National School Lunch Program are considerably higher than the poverty threshold (130% of the poverty threshold for free lunch and 185% of the poverty threshold for reduced-price lunch). Consequently, it is likely that using school lunch data will overestimate the number of poor children, and the extent of overestimation across areas will vary. Reasons for such variation include: differences across areas in the income distribution –one area may have fewer near-poor children relative to poor children than another area; in program administration–some school districts may be more aggressive in encouraging families to participate than other districts; and in participation–some families may not enroll their children because of perceived stigma. For allocation programs for poor children that have significant thresholds to receive funding, such as Title I concentration grants, the use of school lunch counts as a proxy measure would likely provide funds to districts that would not be eligible if a poverty measure were available. For programs that have no or very low thresholds for eligibility, the use of school lunch counts to apportion shares of the total amount to school

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond districts would not necessarily be problematic if there were no variations across areas in the extent to which school lunch counts overestimated poverty. However, such variations are likely. Analysis by the panel found no advantage of school lunch counts over SAIPE school district estimates of poor school-age children for Title I allocations in two states (see Chapter 3; see also National Research Council, 2000c). Generally, careful consideration needs to be given to the use of administrative data as proxy measures of poverty in an allocation formula. However, such data can be very useful in another role, namely, to provide predictor variables for developing small-area poverty (and income) estimates from models, as is done in the SAIPE program. For this use, it is not necessary that the administrative data measure the official poverty concept, but only that the data are a good predictor of poverty and be available at the required geographic level of detail. Yet this use still requires that the administrative data be consistently measured across areas, such as states and counties, so that biases favoring some areas over other areas are not introduced in the prediction models. BEA Income Estimates BEA produces state and county estimates of personal income and per capita personal income as part of the national income and product accounts. The data used to produce the income estimates are primarily from the decennial census and administrative records from federal and state government programs (e.g., records for unemployment insurance, Social Security, Medicare, Medicaid, other social welfare programs, and tax records); surveys also provide some data inputs. The per capita income estimates for state and counties are the total income estimates divided by population estimates, which are obtained from the Census Bureau's population estimates program. The BEA estimates are produced quarterly for states and annually for counties, with a 2-year lag between the time of release and the reference year for the income data. The advantages of the BEA income estimates for use in fund allocation formulas and other program purposes are that they are regularly updated and measure an income concept that distinguishes more well-off from less well-off areas. However, the BEA program does not provide estimates for subcounty areas or for population groups, and it does not provide estimates of poverty or other types of income measures, such as median or average family income. Also, while the BEA personal income measures are viewed as more complete than household income reports from surveys, the BEA personal income concept is not quite the same as the household income concept that is measured in surveys. The BEA concept is broader than household income, including income of quasi-

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond persons (e.g., nonprofit institutions that serve individuals and private trust funds) and treating some sources of income differently. The sampling variability of the BEA estimates for states and counties is not known. SAIPE Estimates The SAIPE Program is a new source of regularly updated small-area estimates of income and poverty. As noted in Chapter 1, SAIPE currently produces the following estimates for states and counties: all poor persons, poor children under the age of 5 (states only), poor children under the age of 18, poor related children aged 5-17, and median income of households. For school districts, SAIPE produces estimates of poor related school-age children. SAIPE state estimates are available for 1993, 1995, and 1996, and will be released annually hereafter. County and school district estimates are available for 1993 and 1995 and will continue to be released on a biennial schedule, with about a 3-year lag from the income reference year. The SAIPE estimates for states and counties are developed from regression models that predict poverty (or income) in the March CPS on the basis of data from administrative records and the previous census. Predictions from the regression model are then combined, when possible, with the direct estimates from the March CPS to form model-based estimates (see Chapter 3 for a description of the estimation procedure). The income tax and food stamp administrative data that are used in the state and county regression models are not currently available for school districts, so a simpler model is used to estimate poor school-age children at the district level. That model applies each school district's share or proportion of the county total of poor school-age children, as measured in the 1990 census, to the updated county estimate from the SAIPE county model. The school district model captures changes in poverty across counties, but necessarily assumes that, within each county, the poorer (less poor) districts at the time of the census remain just as poor (less poor) in later years. The SAIPE model-based estimates have several advantages for use in fund allocation formulas and other program purposes: they are updated annually or biennially; they reflect official concepts of income and poverty with the survey that is currently the source of official income statistics; and they are available for school districts as well as states and counties. Validation work conducted to date indicates that the SAIPE estimates are preferable to continuing to use outdated census estimates: the differences between SAIPE model-based estimates for income year 1989 and 1990 census estimates are substantially smaller than the differences between 1980 census and 1990 census estimates (see Chapter 3).

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond However, the SAIPE estimates have some disadvantages for program use. Although more up to date than census estimates, they lag the income reference year by 3 or more years. Also, they are currently limited in scope (e.g., no estimates are available of poor elderly or poor single-parent families). In the future, it would be possible to reduce the time lag somewhat and to develop estimates for other groups. Although considerable evaluation work has been done, more needs to be learned about the properties of the models and data inputs to assess whether any persistent biases are present in the estimates. Random error will always be present in model-based estimates (which is also true for estimates from any other source). Also, model estimates will generally be less accurate for areas that are at the extremes of the variable being predicted in comparison with areas that have less extreme values. In this regard, evaluation showed that the SAIPE county model overpredicted (underpredicted) the number of poor school-age children in 1989 in areas that experienced a marked decline (increase) in child poverty from 1979 to 1989, but the SAIPE estimates performed substantially better for these areas than the 1980 census estimates. Evaluation to date has not identified significant biases in the SAIPE estimates for other characteristics, but further work is needed on this issue. The production of model-based estimates, such as the SAIPE estimates, requires a significant, continuing investment in model and data validation, research and development, and related activities to ensure that estimates are as accurate as possible. A model-based estimates program should provide full documentation to inform users about the properties of the estimates and their advantages and drawbacks for program use (see Chapter 7).10 CONCLUSION Different data sources for estimates of poverty and income for small areas each have strengths and weaknesses for use in fund allocation formulas and program administration. For example, while the decennial census provides estimates of poverty and income with low sampling variability for many small areas, the estimates are only available once every 10 years. In contrast, administrative records may be available on a timely basis and have other advantages, but they do not always measure the 10   Users should also require documentation and evaluation for estimates that are not developed on the basis of an explicit model: for example, evaluation of the effects on fund allocations of using census estimates over a decade or more or of using administrative data as a proxy for poverty.

OCR for page 18
Small-Area Income and Poverty Estimates: Priorities for 2000 and Beyond concept of income or poverty targeted in the program for which they are needed or consistently relate to that concept across geographic areas or over time. In considering particular sources of small-area income and poverty estimates, it is important for agencies and policy makers to understand their properties and how bias and variability in the estimates may affect their intended program use. We were charged with reviewing the SAIPE Program of small-area income and poverty estimates. We believe that these estimates will be increasingly used for such purposes as fund allocation and program evaluation, as users come to understand their properties and as the SAIPE Program responds to user needs. For example, state estimates are now produced annually instead of every 2 years, and it may be possible to develop estimates for other population groups. In Chapter 3 we identify short-term priorities for research and development for the current SAIPE models. In Chapter 4 and Chapter 5 we consider the possible role of new or modified sources of survey data and administrative data to further improve the estimates, particularly for subcounty areas. These three chapters are aimed primarily at the Census Bureau and other researchers in the field, but they have overview sections that highlight key points of interest to users. In Chapter 6 we return to a user perspective, considering how errors in estimates, which always will be present, may affect formula fund allocations. We illustrate the possibly unintended consequences that can result from interactions between the properties of estimates and provisions of formulas. It is particularly important for users to consider such interactions when deciding to change from one source of estimates to another–as occurred when the Title I program shifted from using decennial census estimates for allocations to using SAIPE estimates –or when developing new funding formulas and deciding which source of estimates to use for them.