Contents

• Introduction
• Motivation for Targeting Funds to Persistent Poverty Counties
• Defining Persistent Poverty Counties
• Computing the Poverty Rate for an Area
• Data Sources Used in Identifying Persistent Poverty Counties
• Considerations When Identifying and Targeting Persistent Poverty Counties
• Selecting the Data Source: Strengths and Limitations of ACS and SAIPE Poverty Data
• Characteristics of Interest: SAIPE for Poverty Alone; ACS for Other Topics in Addition to Poverty
• Geographic Area of Interest: SAIPE for Counties and School Districts Only; ACS for Other Small Areas
• Reference Period of Estimate: SAIPE for One Year, ACS for a Five-Year Span
• Other Considerations
• Treatment of Special Populations in the Official Poverty Definition
• Persistence Versus Flexibility to Recent Situations
• Effects of Rounding and Data Source Selection on Lists of Counties
• Example List of Persistent Poverty Counties

Summary

Research has suggested that areas with a poverty rate 20% or greater experience more acute systemic problems than do lower-poverty areas. The poverty rate is the percentage of the population that is below poverty, or economic hardship as measured by comparing income against a dollar amount that represents needs for a low level of material well-being. Recent Congresses have enacted antipoverty policy interventions that target resources on local communities based on the characteristics of those communities, rather than solely on those of individuals or families. One such policy, dubbed the 10-20-30 provision, was first implemented in the American Recovery and Reinvestment Act of 2009 (ARRA, P.L. 111-5). Title I, Section 105 of ARRA required the Secretary of Agriculture to allocate at least 10% of funds from three rural development program accounts to persistent poverty counties—counties that maintained poverty rates of 20% or more for "the past 30 years, as measured by the 1980, 1990, and 2000 decennial censuses."

One notable characteristic of this provision is that it did not increase spending for the rural development programs addressed in ARRA, but rather targeted existing funds differently. Since ARRA, Congress has applied the 10-20-30 provision for other programs in addition to rural development programs, and may continue to do so, using more recent estimates of poverty rates. Doing this, however, requires updating the list of counties with persistent poverty, and that requires making certain decisions about the data that will be used to compile the list.

Poverty rates are computed using data from household surveys fielded by the U.S. Census Bureau. The list of counties identified as persistently poor may differ by roughly 60 to 100 counties in a particular year, depending on the surveys selected to compile the list and the rounding method used for the poverty rate estimates. In the past, the decennial census was the only source of county poverty estimates across the entire country (there are 3,144 counties or county-equivalent areas, nationwide). After 2000, however, the decennial census is no longer used to collect income data. There are two newer data sources that may be used to provide poverty estimates for all U.S. counties: the American Community Survey (ACS) and the Small Area Income and Poverty Estimates (SAIPE) program. The Census Bureau implemented both the ACS and SAIPE in the mid-1990s. Therefore, to determine whether an area is persistently poor in a time span that ends after the year 2000, policymakers and researchers must first decide whether ACS or SAIPE poverty estimates will be used to demarcate that time span. Which of these surveys is the best data source to use for compiling an updated list of counties with persistent poverty may differ based on the specific area or policy for which the antipoverty intervention is intended.

When defining persistent poverty counties in order to target funds for programs or services using the surveys above, the following factors may be relevant:

• Characteristics of interest: SAIPE is suited for analysis focused solely on poverty or median income; ACS for poverty and income and other topics (e.g., housing characteristics, disability, education level, occupation, veteran status).
• Geographic areas of interest: SAIPE is recommended for counties and school districts only; ACS may be used to produce estimates for other small geographic areas as well (such as cities, towns, and census tracts).
• Reference period of estimate: Both data sources produce annual estimates. The SAIPE estimate is based on one prior year of data while ACS estimates draw on data from the prior five years.
• Rounding method for poverty rates: Rounding to one decimal place (e.g., not including a county with a poverty rate of 19.9% because it is less than 20.0%) yields a shorter list of counties with persistent poverty than rounding to a whole number (e.g., including a county with a poverty rate of 19.9% because it rounds up to 20%).
• Special populations:
• Poverty status is not defined for all persons, such as unrelated household members under age 15 (e.g., children in foster care), institutionalized persons, and residents of college dormitories.
• Persons without housing are not included in household surveys.
• Areas with large numbers of college students living off-campus may have higher poverty rates than might be expected, because poverty is measured using cash income and does not include student loans.

---

Introduction

Antipoverty interventions that provide resources to local communities, based on the characteristics of those communities, have been of interest to Congress. One such policy, dubbed the 10-20-30 provision, was implemented in the American Recovery and Reinvestment Act of 2009 (ARRA, P.L. 111-5). Title I, Section 105 of ARRA required the Secretary of Agriculture to allocate at least 10% of funds provided in that act from three rural development program accounts to persistent poverty counties; that is, to counties that have had poverty rates of 20% or more for "the past 30 years, as measured by the 1980, 1990, and 2000 decennial censuses."¹

One notable characteristic of this provision is that it did not increase spending for the rural development programs addressed in ARRA, but rather targeted existing funds differently. Given Congress's interest both in addressing poverty (economic hardship as measured by comparing income against a dollar amount that represents needs for a low level of material well-being)² and being mindful about levels of federal spending, the 112^th through the 119^th Congresses included 10-20-30 language in multiple bills, some of which were enacted into law.³ However, the original language used in ARRA could not be re-used verbatim, because the decennial census—the data source used by ARRA to define persistent poverty—stopped collecting income information. As a consequence, the appropriations bills varied slightly in their definitions of persistent poverty counties as applied to various programs and departments. This variation has occurred even within different sections of the same bill if the bill included language relating to different programs. In turn, because the definitions of persistent poverty differed, so did the lists of counties identified as persistently poor and subject to the 10-20-30 provision. The bills included legislation for rural development, public works and economic development, technological innovation, and brownfields site assessment and remediation.

More recently, into the 119^th Congress much of the language used in these previous bills was included in three appropriations acts (see Table 1), as well as bills that were not enacted. In some instances, the references to persistent poverty may have targeted resources to counties other than by requiring a 10% set-aside specifically, or by using other geographic areas in addition to counties.⁴

This report discusses how data source selection and the rounding of poverty estimates can affect the list of counties identified as persistently poor. After briefly explaining why targeting funds to persistent poverty counties might be of interest, this report explores how persistent poverty is defined and measured, and how different interpretations of the definition and different data source selections could yield different lists of counties identified as persistently poor. This report does not compare the 10-20-30 provision's advantages and disadvantages against other policy options for addressing poverty, nor does it examine the range of programs or policy goals for which the 10-20-30 provision might be an appropriate policy tool.

Motivation for Targeting Funds to Persistent Poverty Counties

Research has suggested that areas for which the poverty rate (the percentage of the population that is below poverty) reaches 20% experience systemic problems that are more acute than in lower-poverty areas.⁵ The poverty rate of 20% as a critical point has been discussed in academic literature as relevant for examining social characteristics of high-poverty versus low-poverty areas.⁶ For instance, property located in high-poverty areas does not yield as high a return on investment as does property located in low-poverty areas, and that low return provides a financial disincentive for property owners in high-poverty areas to spend money on maintaining and improving property.⁷ The ill effects of high poverty rates have been documented both for urban and rural areas.⁸ Depending on the years in which poverty is measured and the data sources used, between 300 and 500 counties have been identified as persistent poverty counties, out of a total of 3,144 counties or county-equivalent areas nationwide.⁹ Therefore, policy interventions at the community level, and not only at the individual or family level, have been and may continue to be of interest to Congress.¹⁰

Defining Persistent Poverty Counties

Persistent poverty counties are counties that have had poverty rates of 20% or greater for at least 30 years. The county poverty rates for 1999 and previous years have traditionally been measured using decennial census data. For more recent years, either the Small Area Income and Poverty Estimates (SAIPE) or the American Community Survey (ACS) are used. Both of these Census Bureau data sources were first implemented in the mid-1990s and both provide poverty estimates no longer available from the decennial census.¹¹ The data sources used, and the level of precision of rounding for the poverty rate, affects the list of counties identified as persistent poverty counties, as will be described below.

Computing the Poverty Rate for an Area

Poverty rates are computed by the Census Bureau for the nation, states, and smaller geographic areas such as counties.¹² The official definition of poverty in the United States is based on the money income of families and unrelated individuals. Income from each family member (if family members are present) is added together and compared against a dollar amount called a poverty threshold, which represents a level of economic hardship and varies according to the size and characteristics of the family (ranging from one person to nine persons or more). Families (or unrelated individuals) whose income is less than their respective poverty threshold are considered to be in poverty (sometimes also described as below poverty).¹³

Every person in a family has the same poverty status. Thus, it is possible to compute a poverty rate based on counts of persons. This is done by dividing the number of persons below poverty within a county by the county's total population,¹⁴ and multiplying by 100 to express the rate as a percentage.

Data Sources Used in Identifying Persistent Poverty Counties

Poverty rates are computed using data from household surveys. Currently, the only data sources that provide poverty estimates for all U.S. counties are the ACS and SAIPE. Before the mid-1990s, the only poverty data available at the county level came from the Decennial Census of Population and Housing, which is collected once every 10 years. In the past, these data were the only source of estimates that could determine whether a county had persistently high poverty rates (ARRA referred explicitly to decennial census poverty estimates for that purpose). However, after Census 2000, the decennial census has no longer collected income information in the 50 states, the District of Columbia, and Puerto Rico, and as a result cannot be used to compute poverty estimates.¹⁵ Therefore, to determine whether an area is persistently poor in a time span that ends after 2000, it must first be decided whether ACS or SAIPE poverty estimates will be used to demarcate the end of the time span or, depending on the availability of the preferred dataset, whether it may be used for the beginning or other points within it.¹⁶

The ACS and the SAIPE program serve different purposes. The ACS was developed to provide continuous measurement of a wide range of topics similar to that formerly provided by the decennial census long form, available down to the local community level. ACS data for all counties are available annually, but are based on responses over the previous five-year time span (e.g., 2020-2024). ACS five-year data are available beginning for the 2009 five-year estimates (covering 2005-2009) and onward. The SAIPE program was developed specifically for estimating poverty at the county level for school-age children and for the overall population, for use in funding allocations for the Improving America's Schools Act of 1994 (P.L. 103-382). SAIPE data are also available annually, and reflect one calendar year, not five. However, unlike the ACS, SAIPE does not provide estimates for a wide array of topics. SAIPE county data are available for 1989, 1993, 1995, 1997, and annually thereafter. For further details about the data sources for county poverty estimates, see the Appendix.

Considerations When Identifying and Targeting Persistent Poverty Counties

Selecting the Data Source: Strengths and Limitations of ACS and SAIPE Poverty Data

Because poverty estimates can be obtained from multiple data sources, the Census Bureau has provided guidance on the most suitable data source to use for various purposes.¹⁷

Characteristics of Interest: SAIPE for Poverty Alone; ACS for Other Topics in Addition to Poverty

The Census Bureau recommends using SAIPE poverty estimates when estimates are needed at the county level, especially for counties with small populations, and when additional demographic and economic detail is not needed at that level.¹⁸ When additional detail is required, such as for county-level poverty estimates by race and Hispanic origin, detailed age groups (aside from the elementary and secondary school-age population), housing characteristics, or education level, the ACS is the data source recommended by the Census Bureau.

Geographic Area of Interest: SAIPE for Counties and School Districts Only; ACS for Other Small Areas

For counties (and school districts) of small population size, SAIPE data have an advantage over ACS data in that the SAIPE model uses administrative data to help reduce the uncertainty of the estimates. However, ACS estimates are available for a wider array of geographic levels, such as ZIP code tabulation areas, census tracts (subcounty areas of roughly 1,200 to 8,000 people), cities and towns, and greater metropolitan areas.¹⁹

Reference Period of Estimate: SAIPE for One Year, ACS for a Five-Year Span

While the ACS has greater flexibility in the topics measured and the geographic areas provided, it can only provide estimates in five-year ranges for the smallest geographic areas. Five years of survey responses are needed to obtain a sample large enough to produce meaningful estimates for populations below 65,000 persons. In this sense the SAIPE data, because they are based on a single year, are more current than the data of the ACS. The distinction has to do with the reference period of the data—both data sources release data on an annual basis; the ACS estimates for small areas are based on the prior five years, not the prior year alone.

Other Considerations

Treatment of Special Populations in the Official Poverty Definition

Regardless of the data source used to measure it, poverty status is not defined for persons in institutions, such as nursing homes or prisons, nor for persons residing in military barracks. These populations are excluded from totals when computing poverty statistics. Furthermore, the homeless population is not counted explicitly in poverty statistics. The ACS is a household survey, thus homeless individuals who are not in shelters are not counted. SAIPE estimates are partially based on Supplemental Nutrition Assistance Program (SNAP) administrative data and tax data, so the part of the homeless population that either filed tax returns or received SNAP benefits might be reflected in the estimates, but only implicitly.

In the decennial census, ACS, and SAIPE, poverty status also is not defined for persons living in college dormitories.²⁰ However, students who live in off-campus housing are included. Because college students tend to have lower money income (which does not include school loans) than average, counties that have large populations of students living off-campus may exhibit higher poverty rates than one might expect given other economic measures for the area, such as the unemployment rate.²¹

Given the ways that the special populations above either are or are not reflected in poverty statistics, it may be worthwhile to consider whether counties that have large numbers of people in those populations would receive an equitable allocation of funds. Other economic measures may be of use, depending on the type of program for which funds are being targeted.

Persistence Versus Flexibility to Recent Situations

The 10-20-30 provision was developed to identify and target funding to counties that have had persistently high poverty rates over an extended period. Therefore, using that approach by itself would not allow flexibility to target funding at counties that have only recently experienced economic hardship, such as counties that had a large manufacturing plant close within the past three years. Other interventions besides the 10-20-30 provision may be more appropriate for identifying and targeting funding at counties that have had a recent spike in the poverty rate in contrast to having experienced persistent poverty.

Effects of Rounding and Data Source Selection on Lists of Counties

In ARRA, persistent poverty counties were defined as "any county that has had 20 percent or more of its population living in poverty over the past 30 years, as measured by the 1980, 1990, and 2000 decennial censuses."²³ Poverty rates published by the Census Bureau are typically reported to one decimal place. The numeral used in the ARRA statutory text was the whole number 20. Thus, for any collection of poverty data, two reasonable approaches to compiling a list of persistent poverty counties include using poverty rates of at least 20.0% in all three years, or using poverty rates that round up to the whole number 20% or greater in all three years (i.e., poverty rates of 19.5% or more in all three years). The former approach is more restrictive and results in a shorter list of counties; the latter approach is more inclusive.²⁴

Two tables in the Appendix illustrate the effects of rounding and data source selection. Table A-2 illustrates the number of counties identified as persistent poverty counties using the 1990 and 2000 decennial censuses, and various ACS and SAIPE datasets for the last data point, under both rounding schemes. The rounding method and data source selection can each have large impacts on the number of counties listed. In most years, using SAIPE for the latest year resulted in more counties being identified as persistently poor than were identified by using the ACS; the exceptions were 2019 and 2020. Compared to using 20.0% as the cutoff (rounded to one decimal place), rounding up to 20% from 19.5% adds approximately 40 to 60 counties to the list. Taking both the data source and the rounding method together (as shown in Table A-3), the list of persistent poverty counties could vary by roughly 60 to 100 counties in a given year depending on the method used.

During the 115^th Congress, when the 10-20-30 provision began to be used in annual appropriations acts for the first time since ARRA, the acts defined persistent poverty using decennial census data and the latest ACS or SAIPE data. Congress has used additional approaches since that time. Beginning with the Consolidated Appropriations Act, 2022, the definition of persistent poverty has been updated for the Agriculture, Commerce, and Environmental Protection Agency programs to use the 1993 SAIPE as the source of the first poverty rate, Census 2000 as the second, and the most recent SAIPE as the third (which taken together cover a 31-year span, using the most recent SAIPE estimates as of the cover date of this report). As indicated in Table 2, this definition continues to be used for these programs. To establish clear comparisons that isolate the effects of rounding and data source selection, Table A-2 and Table A-3 display older definitions using the 1990 and 2000 decennial census estimates.

Example List of Persistent Poverty Counties

The list of persistent poverty counties below (Table 3)²⁵ is based on data from the 1993 SAIPE, Census 2000, and the 2024 SAIPE estimates, and includes the 366 counties with poverty rates of 19.5% or greater (that is, counties with poverty rates that were at least 20% with rounding applied to the whole number). These same counties are mapped in Figure 1.

This definition of persistent poverty was first used in the Consolidated Appropriations Act, 2022 in reference to Public Works grants authorized by the Public Works and Economic Development Act of 1965 and grants authorized by Section 27 of the Stevenson-Wydler Technology Innovation Act of 1980 (P.L. 117-103, Division B, Title V, §533), as well as in Division G, Title II of the same act, for State and Tribal Assistance Grants administered by the Environmental Protection Agency for carrying out Section 104(k) of the Comprehensive Environmental Response, Compensation, and Liability Act of 1980.

This list of 366 counties (out of a total of 3,144 nationwide) is similar but not identical to a list that would be compiled if ACS data were to be used, because poverty estimates from different data sources almost always differ.

Figure 1. Persistent Poverty Counties Using Two Rounding Methods, Based on 1993 and 2024 Small Area Income and Poverty Estimates and Census 2000

Source: Created by the Congressional Research Service (CRS) using data from U.S. Census Bureau, 1993 and 2024 Small Area Income and Poverty Estimates, and Census 2000.

Appendix. Details on the Data Sources

Decennial Census of Population and Housing, Long Form

Poverty estimates are computed using data from household surveys, which are based on a sample of households. To obtain meaningful estimates for any geographic area, the sample has to include enough responses from that area so that selecting a different sample of households from that area would not likely result in a dramatically different estimate. If estimates for smaller geographic areas are desired, a larger sample size is needed. A national-level survey, for instance, could produce reliable estimates for the United States without obtaining any responses from many counties, particularly counties with small populations. To produce estimates for all 3,144 county areas in the nation, however, not only are responses needed from every county, but those responses have to be plentiful enough from each county so that the estimates are meaningful (i.e., their margins of error are not unhelpfully wide).

Before the mid-1990s, the only data source with a sample size large enough to provide meaningful estimates at the county level (and for other small geographic areas) was the decennial census. The other household surveys available prior to that time did not have a sample size large enough to produce meaningful estimates for small areas such as counties. Income questions were asked on the census long form, which was sent to one-sixth of all U.S. households; the rest received the census short form, which did not ask about income. While technically still a sample, one-sixth of all households was a large enough sample to provide poverty estimates for every county in the nation, and even for smaller areas such as small towns. The long form was discontinued after Census 2000, and therefore poverty data are no longer available from the decennial census for the 50 states, the District of Columbia, and Puerto Rico.²⁶ Beginning in the mid-1990s, however, two additional data sources were developed to ensure that poverty estimates for small areas such as counties would still be available: the American Community Survey (ACS), and the Small Area Income and Poverty Estimates (SAIPE) program.

American Community Survey (ACS)

The ACS replaced the decennial census long form. It was developed to accommodate the needs of local government officials and other stakeholders who needed detailed information on small communities on a more frequent basis than once every 10 years. To that end, the ACS questionnaire was designed to reflect the same topics asked in the census long form.

To produce meaningful estimates for small communities, the ACS needs to collect a number of responses comparable to what was collected in the decennial census.²⁷ To collect that many responses while providing information more frequently than once every 10 years, the ACS collects information from respondents continuously, in every month, as opposed to at one time of the year, and responses over time are pooled to provide estimates at varying geographic levels. To obtain estimates for geographic areas of 65,000 or more persons, one year's worth of responses are pooled—these are the ACS one-year estimates. For the smallest geographic levels, which include the complete set of U.S. counties, five years of monthly responses are needed: these are the ACS five-year estimates. Even though data collection is ongoing, the publication of the data takes place once every year, both for the one-year estimates and the estimates that represent the previous five-year span.

Small Area Income and Poverty Estimates (SAIPE)

The SAIPE program was developed in the 1990s in order to provide state and local government officials with poverty estimates for local areas in between the decennial census years. In the Improving America's Schools Act of 1994 (IASA, P.L. 103-382), which amended the Elementary and Secondary Education Act of 1965 (ESEA), Congress recognized that providing funding for children in disadvantaged communities created a need for poverty data for those communities that were more current than the once-a-decade census. In the IASA, Congress provided for the development and evaluation of the SAIPE program for its use in Title I-A funding allocations.²⁸

SAIPE estimates are model-based, meaning they use a mathematical procedure to compute estimates using both survey data (ACS one-year data) and administrative data (from tax returns and numbers of participants in the Supplemental Nutrition Assistance Program, or SNAP). The modeling procedure produces estimates with less variability than estimates computed from survey data alone, especially for counties with small populations.

Guidance from the U.S. Census Bureau,
"Which Data Source to Use for Poverty"²⁹

The CPS ASEC[³⁰] provides the most timely and accurate national data on income and is the source of official national poverty estimates, hence it is the preferred source for national analysis. Because of its large sample size, the ACS is preferred for subnational data on income and poverty by detailed demographic characteristics. The Census Bureau recommends using the ACS for 1-year estimates of income and poverty at the state level. Users looking for consistent, state-level trends should use CPS ASEC 2-year averages and CPS ASEC 3-year averages for state to state comparisons.

For substate areas, like counties, users should consider their specific needs when picking the appropriate data source. The SAIPE program produces overall poverty and household income 1-year estimates with standard errors usually smaller than direct survey estimates. Users looking to compare estimates of the number and percentage of people in poverty for counties or school districts or the median household income for counties should use SAIPE, especially if the population is less than 65,000. Users who need other characteristics such as poverty among Hispanics or median earnings, should use the ACS, where and when available.

The SIPP[³¹] is the only Census Bureau source of longitudinal poverty data. As SIPP collects monthly income over 2.5 to 5 year panels, it is also a source of poverty estimates for time periods more or less than one year, including monthly poverty rates.

Table A-1 below reproduces the Census Bureau's recommendations, summarized for various geographic levels.

Comparing Lists of Persistent Poverty Counties Using Different Datasets and Rounding Schemes

Table A-2 illustrates the number of counties identified as persistent poverty counties using the 1990 and 2000 decennial censuses, and various ACS and SAIPE datasets for the last data point, under two rounding schemes: rounded to one decimal point (which includes only poverty rates of 20.0% or greater) and rounded to the whole number (which includes poverty rates of 19.5% or greater). The rounding method and data source selection can each have large impacts on the number of counties listed. In most years, using SAIPE for the latest year resulted in more counties being identified as persistently poor than were identified by using the ACS; the exceptions were 2019 and 2020. Compared to using 20.0% as the cutoff (rounded to one decimal place), rounding up to 20% from 19.5% adds approximately 40 to 60 counties to the list. Taking both the data source and the rounding method together (Table A-3), the list of persistent poverty counties could vary by roughly 60 to 100 counties in a given year depending on the method used.

As of the cover date of this report, no program or funding stream uses a definition of persistent poverty counties as the combination of 1990 Census, Census 2000, and the most recent (2024) ACS five-year poverty estimates. The combination of 1990 Census, Census 2000, and the most recent (2024) SAIPE data is used in 49 U.S.C. §6702(a)(1), with regard to local and regional project assistance grant programs.