The Impact of Precise Geographic Adjustments on the Supplemental Poverty Measure

Abstract

The official poverty measure of the United States remains unequipped to appropriately capture poverty across America. As a result, the Supplemental Poverty Measure (SPM) has increasingly supplanted the official measure in policy analysis and statistics. A primary point of conflict among poverty-focused scholars regarding the SPM is its current geographic adjustment, which adjusts poverty thresholds at three spatial scales: identified metropolitan areas, unidentified metropolitan areas by state, and nonmetropolitan areas by state. Pooling all nonmetropolitan counties within each state into a single adjustment is believed to be responsible for the ‘flip’ in the rural-urban poverty differential between the official measure and the SPM. Using federally restricted data, we address this conflict and generate novel estimates of the SPM using county-specific, hybrid, and commuting-zone geographic adjustments. Our estimates illustrate the role of the current adjustment in our understanding of rural-urban poverty, while also demonstrating the utility of our preferred commuting-zone-level adjustment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11113-026-09995-1.

Introduction

The overwhelming consensus among poverty scholars is that the Official Poverty Measure (OPM) of the United States is severely lacking in its ability to appropriately measure poverty (Brady, 2009, 2023; Hutto et al., 2011). The measure has faced criticism for its use of gross income as opposed to net, inattention to non-cash transfers, artificially low-income thresholds (i.e. the income needed to not be in poverty), and lack of a geographic adjustment for differences in cost of living, among others (Brady, 2023; Hutto et al., 2011). The lack of a geographic adjustment is largely thought to explain part of the large differences in poverty rates across U.S. regions and between rural and urban areas. Due to the significant shortcomings of the OPM, the Census Bureau—following a series of panels and workshops hosted by the National Academy of Science—developed what is now called the Supplemental Poverty Measure (SPM) (Hutto et al., 2011). This measure, which is now released alongside the OPM in the annual Census Bureau poverty reports, addresses many of the shortcomings of the OPM.

The improvements within the SPM include: (1) a more robust accounting of income, (2) a more holistic definition of families, (3) a regularly updated poverty threshold based on recent expenditure surveys, and (4) a geographic adjustment to the poverty threshold to account for differences in cost-of-living across space (Shrider & Creamer, 2023). As a result, the SPM has quickly become the preferred poverty measure among researchers for policy analysis and documenting poverty at a variety of geographic scales and between sociodemographic groups. Although the SPM can easily be regarded as an improvement over the OPM, the geographic adjustment, as currently employed, has come under scrutiny (Jensen & Ely, 2017; Mueller et al., 2022; Pacas & Rothwell, 2020). The source of these critiques is largely due to how the adjustment impacts differences in poverty between rural and urban areas. The OPM suggests that rural poverty is higher than urban poverty, whereas the SPM does the opposite. This rural-urban poverty flip has caused consternation among scholars and is solely due to the geographic adjustment (Pacas & Rothwell, 2020). As a result, researchers have called for investigation into the validity of the current approach, as well as the testing of alternatives. In this paper, we answer this call using restricted Census Bureau data to produce new estimates of the SPM relying on alternative geographic adjustments with greater precision than those currently used in the SPM.

In what follows, we detail the current geographic adjustment in the SPM and then discuss the conceptual strengths and weaknesses of three alternative adjustments: (1) county-level, (2) a hybrid approach aligned with the recommendations of the recent National Academies (NASEM) report on improving the SPM (NASEM, 2023), and (3) commuting-zone level. Following this conceptual discussion where we advocate for the use of commuting zones, we leverage restricted-use data from the Census Bureau to compare actual estimates of the SPM under each approach at various scales. We use this restricted data to make multiple novel poverty estimates along the dimensions of metropolitan status, region, race and ethnicity, and degree of rurality (i.e. Rural-Urban Continuum Codes, RUCC). Understanding how technical decisions regarding poverty measurement impact our understanding of geographic and racial poverty disparities is much needed in a time of pronounced spatial and racial inequality.

Background

The Current Geographic Adjustment in the SPM

Starting in 2009, the U.S. Census Bureau began reporting the SPM in order to supply estimates of poverty in the nation, with this new measure being widely considered a technical and substantive advancement compared to earlier measures of poverty in the United States (i.e., the Official Poverty Measure) (Shrider & Creamer, 2023). Broadly speaking, under the SPM, individuals are considered to be in poverty if their family’s (technically called an SPM-unit) resources are below their family’s size-specific geographically-adjusted poverty threshold (Shrider & Creamer, 2023). SPM-units consist of all related individuals within a household, including cohabitating partners and any children of cohabitating partners. Households can contain multiple SPM units, such as the case with households that contain multiple unrelated individuals (e.g., roommates, borders). The resources—the term used in the SPM to capture various sources of income—of an SPM unit are as follows: cash resources (e.g., wage income, social security income, investment income), non-cash resources (e.g., SNAP, WIC) and tax credits (e.g., EITC). Additionally, the SPM subtracts select expenses from these resources such as medical costs, childcare, and taxes (e.g., payroll tax)—with the logic being that SPM-units must devote income to these expenses before considering covering other expenses such as groceries or mortgage payments. Once SPM-units’ resources are properly calculated, they are compared with their relevant SPM threshold. This threshold has changed slightly over time and is currently based on 1.2 times 83% of median expenditures of two-parent, two-child households on food, clothing, utilities (e.g. water bill, internet), and housing costs as indicated by data from the previous five years’ Consumer Expenditure Surveys (Schild, 2023). This threshold is further adjusted based on the size of the SPM-unit and whether it contains older adults and children (i.e. equivalence scale), and, most central to our purposes, their geographic location (i.e. the geographic adjustment) (Shrider & Creamer, 2023).

The geographic adjustment is based on housing costs—an approach that Mueller et al. (2022) deemed reasonable in their analysis comparing rent and commuting time, and one that the NASEM report continues to broadly support (NASEM, 2023). The specific metric used is the median gross rent for a two-bedroom unit with complete kitchen and bathroom facilities. Gross rent captures both contract rent and utilities such as electricity and gas. Median gross rent is determined using microdata from the American Community Survey and facilitates the geographic adjustment presented in Eqs. 1 and 2 (Pacas & Rothwell, 2020; Renwick, 2020). Equation 1 shows that the first step is to create a median rent index (MRI) for each geographic unit being used for adjustment. In short, the MRI is calculated as the ratio of the “local” median gross rent to the national median gross rent.

1

Following calculation, the MRI is then used in Eq. 2 to adjust each local poverty threshold. In this equation, t denotes housing tenure (i.e. owner with mortgage, owner without mortgage, renter) and i denotes the local area. HousingShare is the portion of the poverty threshold that is made up of housing for each tenure group and is determined using the Consumer Expenditure Survey. Thus, the adjustment for each tenure group is scaled depending on the relevant portion of their household resources spent on housing. For the year of 2022, this was 35.3% for owners with mortgages, 22.5% for owners without mortgages, and 37.8% for renters (Bureau of Labor Statistics, 2025). This scaling term is then multiplied by the relevant Threshold for the poverty unit/household size and tenure status to determine their specific SPM threshold.

2

These two formulas, and the practice of dividing households based on housing tenure, has received some critiques. For example, Pacas and Rothwell (2020) demonstrated that MRI, not housing tenure, was the dimension largely responsible for the rural-urban poverty flip. Further, the recent NASEM report argues for the use of Fair Market Rent’s from the U.S. Department of Housing and Urban Development instead of MRI, while also arguing for the removal of tenure-specific adjustments to the threshold. Despite these critiques, our focus in this study is to evaluate the “local” in the SPM’s local geographic adjustment. In other words, any subsequent changes to the indicator of cost of living within the SPM should still rely on a conceptually grounded scale of geographic adjustment.

To understand the “local” geographic units used in the geographic adjustment employed in the SPM, we first must discuss the public-use files of the Current Population Survey (CPS), the underlying data source for SPM estimates. The CPS is a national monthly survey of households conducted by the U.S. Census Bureau and Bureau of Labor Statistics to inform the monthly employment situation report (colloquially known as the monthly jobs report). This survey includes a supplement every March known as the Annual Social and Economic Supplement (ASEC). The ASEC includes questions on household social and economic characteristics that are then used to calculate the various metrics required for the SPM. Within the public-use file, there are three mutually exclusive sub-state geographies for which estimates can be produced: (1) identified metropolitan areas, (2) unidentified metro areas, (3) nonmetropolitan areas (Pacas & Rothwell, 2020). To be identified, a metropolitan area must be greater than 100,000—although it is worth noting that the Bureau has recently introduced synthetic procedures to avoid disclosure risks in areas between 100,000 and 249,999 (U.S Census Bureau, 2022). Thus, within the CPS we can identify the specific metropolitan area of many residents of large cities and their suburbs, but only state and metro/nonmetro status for those residing in smaller cities, towns, and rural areas. According to the Census Bureau, only 260 of the 387 metropolitan areas are identified within the public data (U.S. Census Bureau, 2022).

Importantly, the official version of the SPM relies upon these geographies as well as state of residence for its geographic adjustment. This means that SPM thresholds (i.e. the amount of income that an individual needs to be considered not in poverty) are uniquely adjusted for those living in large metropolitan areas and state-level adjustments are applied to those living in either unidentified metropolitan or nonmetropolitan counties (Creamer et al., 2022). Effectively, all nonmetropolitan residents of a state receive the same adjustment regardless of where in the state they live, with residents of unidentified metropolitan areas receiving their own state-level adjustment. We argue these state-level adjustments for all unidentified metropolitan areas and nonmetropolitan areas pose serious issues in terms of construct validity.

The validity concerns related to the geographic adjustment stem from several factors. First, those living in larger cities are afforded greater precision in their poverty threshold than those living in smaller cities or rural areas. From a conceptual standpoint, there is no clear reason why the scale of adjustment for those who happen to live in bigger areas should be more precise. While it certainly makes sense to create a different adjustment for cities like Kansas City, KS and Wichita, KS, it is unclear why someone living in a nonmetropolitan county near the Kansas City metro area should have the same adjustment as someone living in remote western Kansas.

Second, the inclusion of any geographic adjustment is a recognition that the amount of resources required for living above the poverty line varies between labor markets in the United States. Thus, in some areas, one’s income can be lower because things cost less, and one’s income needs to be higher in areas where the same set of goods presumably cost more. This is thought to be particularly true for housing. Unfortunately for the official SPM, it seems highly unlikely that the state is the ideal scale for this adjustment. Whereas some very small states on the East Coast may contain a single labor market, this is an unrealistic notion for most states. Those living in the Texas Panhandle are not in the same labor market as those in the state’s Rio Grande Valley, and those living in the Florida Panhandle are in a very different labor market than those living in the Florida Keys.

The third issue is tied to the second. That is median rent, the basis of the SPM geographic adjustment, simply varies dramatically within states. In their 2022 paper, Mueller and colleagues demonstrated this high level of variability, showing that median rent can vary by more than $500 between nonmetropolitan counties within the same state. As such, the current practice of large-scale pooling presents both conceptual and methodological issues for properly measuring poverty in the United States. Given these issues, we believe changes to the scale of geographic adjustment should be considered.

Alternative Approaches

It is important to note that any decision regarding the scale of a geographic adjustment is conceptual. There is no clear test to determine which adjustment ‘most accurately’ reflects poverty across space. Poverty is a concept that rarely has an agreed upon definition and is endlessly debated and scrutinized (Brady, 2019). Some scholars advocate for absolute measures of poverty that only capture subsistence (or lack thereof) while others argue for fully relative measures that express an individual’s ability to live a full life; within and between either of these options are many branching decisions one can make. It is certainly true that one can evaluate poverty measures by comparing them to other measures of hardship. For example, one could plot different poverty measures alongside measures of food insecurity, housing insecurity, deaths of despair, or others to see which best aligns with these measures. For example, O’Neill’s (2025) recent work compares a variety of poverty measures to different indicators of well-being. However, given the considerable debate surrounding what exactly a poverty measure is meant to reflect, these comparisons would still fundamentally rely on a specific set of assumptions. As such, here we focus squarely on the conceptual rationale for selecting one geographic adjustment over others.

While there is a consensus among scholars that some form of geographic adjustment is necessary for measuring poverty in the U.S. context—as evidenced by the current practices used in the SPM and the recommendations in the recent NASEM report (NASEM, 2023)—the scale of that adjustment does not have a clear a priori answer. From our perspective, it appears that a more precise adjustment is necessary due to the validity concerns presented above. Within this paper, we focus on three alternatives to address these concerns: county-sensitive SPM, NASEM-SPM, and commuting-zone SPM. We visualize the geographic variation of these alternative approaches in Fig. 1. This figure shows a collection of counties that under at least one approach would be grouped together with counties that make up the Atlanta-Sandy Springs-Roswell MSA (outlined by a thick black border). Panel A of this figure shows the official SPM’s geographic adjustment, which provides unique adjustments for large identified MSAs, but separate statewide adjustments all unidentified metropolitan and nonmetropolitan counties within a state. Here, nonmetropolitan residents of Alabama receive a different geographic adjustment from their nonmetropolitan peers who live in Georgia, the Atlanta-Sandy Springs-Roswell MSA gets its own adjustment, and those who live in the “unidentified” Rome, GA MSA receive the same statewide adjustment of other unidentified MSAs in the state (e.g., Hinesville-Ft. Steward MSA in coastal Georgia).

Fig. 1.

Comparison of metropolitan statistical areas and commuting zones for Kansas City and Atlanta

Panel B shows an alternative adjustment, which we call county-sensitive SPM. County-sensitive SPM is as it sounds—each county receives a unique adjustment that is applied to the base SPM poverty threshold. Figure 1 shows 70 counties, which would be subject to 70 distinct geographic adjustments. From this, SPM resources are compared to this threshold and the number of individuals in poverty are determined. Importantly, a county-specific adjustment would not lead to county-level estimates of SPM poverty. Due to the sampling structure of the CPS, reliable geographic estimates of the SPM can still only be made at the state level and higher. While the exact number of individuals surveyed and the counties they live in vary year to year, the CPS is not designed to survey individuals from all 3,244 counties in the United States. Further, although a county-sensitive adjustment is incredibly precise, its validity is debatable due to over-fitting. Employing a county-specific adjustment risks understating connectivity in labor or housing markets between counties in the United States. Within Fig. 1 specifically, residents of the city of Atlanta who live in Fulton County would be subject to a different adjustment than Atlantans that live in Dekalb county (with the context here being that the city of Atlanta sprawls across county boundaries). As a further example, residents of Fulton and Dekalb counties would have a different poverty threshold than those who live in the MSA’s other 22 suburban counties—a decision that lacks a clear a priori rationale. Obviously, cost of living varies within metropolitan areas, but when adjustments become too precise, the generalizability of the measure, as well as its ability to capture the necessary resources for living in a broader labor market, disappears. Additionally, low-income individuals are highly mobile and generally move to places with a concentration of other low-income individuals (Lichter et al. 2022). Presumably, these are areas with a relatively lower cost of living. Thus, county-specific adjustments may not make sense if a given county within a metropolitan area is disproportionately home to a larger share of the metropolitan area’s lower income residents. Simply put, spatial sorting based on income within a labor market complicates county-specific adjustments.

The recent NASEM report appears to be aware of the limitations inherent to county-sensitive SPM. As a result, their recent report recommends that thresholds be adjusted by individual metropolitan areas and individual nonmetropolitan counties (NASEM, 2023). This approach—shown in Panel C—retains unique adjustments for identified metropolitan areas, but then allows for county-specific geographic adjustments for all nonmetropolitan counties and unidentified metropolitan counties. In the case of the state of Georgia, larger MSAs like the Atlanta area would receive a single MSA-level adjustment and the counties within Georgia’s unidentified metropolitan areas (e.g., Floyd county in the singular county Rome, MSA) would each receive their own county-specific adjustments. Georgia’s 82 nonmetropolitan counties would also receive their own unique adjustment. While a hybrid approach such that of NASEM’s may be preferable to current practice, it still relies on different conceptual scales for metropolitan and nonmetropolitan adjustments. Although likely an improvement, there are two immediate weaknesses with this approach: (1) many metropolitan areas today are far larger than a single labor market, which is likely to introduce overly coarse adjustments for metropolitan residents; (2) this approach still relies upon conceptually different approaches to metropolitan and nonmetropolitan areas, with metropolitan adjustments being informed by inter-county commuting patterns via the Office of Management and Budget, but nonmetropolitan areas simply receiving a unique county-level adjustment based on an assumed perfect overlap between a nonmetropolitan individual’s county and their labor market. This assumption is particularly dubious when we consider the many micropolitan statistical areas within nonmetropolitan America.

An alternative to county-sensitive SPM or NASEM-SPM are adjustments based on commuting zones. This alternative is shown in Panel D. Commuting zones, also known as labor market areas, are geographic clusters of counties that, “…are geographically defined areas where individuals can both live and work. While the ‘ideal’ labor market may mean different things to different people, it is generically a self-contained economic unit with dense economic activity within its border and little economic activity crossing its borders (Fowler 2024a, para. 3).” Commuting zones have historically been set by a variety of federal agencies using a number of different approaches. Recently, Fowler (2024b) provided an updated classification of commuting zones in the United States based primarily on data from the 2016–2020 American Community Survey and 2020 Decennial Census. In establishing this updated classification scheme, Fowler (2024b) built upon the work of Tolbert and Sizer (1996) and Fowler et al. (2016) and used proportional flows between counties as the underlying delineator of commuting zones. These proportional flows are used to generate a dissimilarity index that is subjected to hierarchical cluster analysis to determine the number and structure of commuting zones. All told, there are 593 distinct commuting zones in the 2020 vintage (Fowler 2024b). These zones combine nonmetropolitan counties into commuting zones with their metropolitan and nonmetropolitan economic neighbors, while also breaking up very large metropolitan areas into a more realistic and grounded representation of living and working patterns. Within Panel D, counties in the Atlanta Metropolitan area are split across seven commuting zones, with a nonmetropolitan county like Upson County, GA becoming part of the same commuting zone as Atlanta-housed Fulton and Dekalb counties. Indeed, every commuting zone shown in Panel D contains both metropolitan and nonmetropolitan counties.

Fowler (2024b) evaluated the validity of these 2020 commuting zones based on three criteria, all of which are relevant for the application of commuting zones to the SPM’s geographic adjustment. First, empirically valid commuting zones should have a clear economic core which the commuting flows themselves are largely based around. For most metropolitan areas, this core is one of the principal cities of a metropolitan area, but even commuting zones that consist entirely of nonmetropolitan counties have identifiable cores. Following notions of central place theory, the characteristics of this core (e.g. population size, housing availability) in remote rural regions likely impact the cost of living for the entire commuting zone. Second, commuting zones are evaluated on their connections, meaning that they should be economically connected. On the ground, this means that individuals who live in a commuting zone should, in theory, have a high probability of looking for work or housing within the counties of a commuting zone. Changes to rental prices or average incomes in one county of the commuting zone should logically spill-over into all other counties within the commuting zone.

Finally, commuting zones should have high levels of containment, meaning that the economic and housing borders between commuting zones are relatively strong. In a perfectly contained commuting zone, residents should only be looking for housing or employment within their commuting zone and should not be relocating across commuting zones unless there is a larger intentional decision to migrate. The median rents of one commuting zone, and any relevant shocks to it, should not spill over to neighboring commuting zones. Fowler believes that containment is particularly critical for more rural regions, since containment validates that residents truly conduct most of their economic activity within their multi-county commuting zone, and don’t live, work, or look for housing across multiple distinct zones. Importantly, Fowler (2024b) not only advocates for the importance of these criteria but also provides evidence demonstrating that the most recent vintage not only achieves these criteria, but does so at a similar level to the vintage based on 2010 data. Given the relevance of these three criteria, and Fowler’s (2024b) success in achieving them, we believe commuting zone SPM is the most valid option available and a prudent path forward for an improved SPM.

Data and Methods

We used restricted-access CPS and American Community Survey (ACS) data within the Federal Research Data Centers to calculate the six poverty measures: official poverty measure, SPM with the official geographic adjustments, county-sensitive SPM, NASEM-SPM, commuting zone SPM, and SPM with no geographic adjustment. We include the first and last measures as reference points. We calculate our estimates for the years of 2018 to 2022, using the 2019 to 2023 CPS-ASEC data. MRI was calculated using ACS five-year microdata, with end year representing the year of the estimate (e.g. 2018–2022 = 2022). We calculate estimates for these five years because 2019 was the first year of CPS-ASEC data where the Census Bureau-constructed SPM variables were available within the restricted-access dataset. The Census Bureau calculates many of the SPM variables using internal techniques that would be difficult to replicate within the RDC environment because non-Bureau researchers are not allowed to bring external microdata (e.g. the previously published SPM variables for years prior to 2019) into the RDC. Focusing on these years allowed us to generate SPM estimates while only adjusting the thresholds already defined by the Bureau and using SPM resources as previously calculated for the official estimates. Although we calculated estimates for all five of these years that we have made publicly available, in the main text, we only present the most recent data for 2022. Estimates for the other years, including population count estimates and unreported estimates for Census Divisions and Census Regions by Race, follow a similar pattern and are available in the supplemental materials (Tables S1-S6).

We first calculated MRI for each county and each commuting zone using ACS restricted microdata and then brought those estimates into CPS-ASEC. From there, we adjusted the thresholds as shown in Eq. 2, replicating the threshold adjustments within official SPM as initial validation. From there, we collapsed the data into weighted poverty headcounts using the CPS-provided person weights at multiple scales. In all cases, our rates were generated by collapsing individual-level data. Although the SPM is calculated at the scale of the SPM-unit, in order to properly capture racial and ethnic variation, aggregating up from the individual level was mandatory to ensure that mixed-race SPM-units were properly divided into their respective poverty rates. Although our initial plan included generating confidence intervals around our population estimates using the same method as the annual Census Bureau poverty reports (e.g. Shrider & Creamer, 2023), we were unable to do so due to replicate weights not being available within the FSRDC environment. As such, we focus on point estimates of poverty and present descriptive comparisons of different geographic adjustments within the SPM.

Here we present national metropolitan and nonmetropolitan estimates, followed by regional estimates, national estimates across racial and ethnic groups, combinations of race and region (all disaggregated by metropolitan and nonmetropolitan), and estimates disaggregated by Rural-Urban Continuum Codes (RUCC). Within our racial and ethnic estimates, we focus on four groups: non-Hispanic American Indian, non-Hispanic Black, non-Hispanic White, and Hispanic of any race. Due to the smaller sample size of the CPS relative to surveys like the ACS, our regional by race and divisional data are pooled into weighted three-year estimates (e.g. 2020–2023). Given the metropolitan-nonmetropolitan poverty flip was the motivating factor for this research, we compare estimates across metropolitan and nonmetropolitan areas using the 2023 release of the RUCC. The RUCC is a county-level classification scheme developed by the USDA Economic Research Service wherein levels 1 through 3 are metropolitan counties of decreasing size and levels 4–9 are nonmetropolitan counties of decreasing urban population, with even numbered counties being adjacent to metropolitan areas and odd numbered counties being non-adjacent. Although not a primary goal of this research, we believe our estimates of SPM by RUCC are the first ever produced.

Results

National estimates of each measure are presented in Table 1. Upon reviewing the table, one might be surprised to see such a limited impact of the different adjustments, with different specifications of the SPM only yielding a range of estimates 0.3 points apart. Consequently, these findings demonstrate the necessity of disaggregation for a comprehensive value assessment of each option. Given the strong influence of dense urban populations on national rates, a spatial analysis is necessary to fully capture the impact of each adjustment. Figure 2 illustrates this variation.

Table 1.

National poverty rate by poverty measure

Measure	National poverty rate
Official Poverty Measure	12.5
Official SPM	12.4
County-Sensitive SPM	12.3
NASEM-SPM	12.5
Commuting Zone SPM	12.4
SPM with no geographic adjustment	12.2

Source: 2023 CPS-ASEC via FSRDC Project Number 2972. Point estimates for 2018–2022 available in Table S1

Fig. 2.

National poverty rates.

Source: 2023 CPS-ASEC via FSRDC Project Number 2972. Point estimates for 2018-2022 available in Table S1

Figure 2 highlights the ‘flip’ between the OPM and SPM that has been a topic of debate among poverty scholars for years. Using the OPM, nonmetropolitan poverty is over 3% points higher than metropolitan (15.6% vs. 12.4%). However, using the official SPM, the two poverty rates are essentially equal, with nonmetropolitan now being 0.1% points lower. These disparities begin to shift as we consider our three alternatives and SPM without geographic adjustments. In all four cases, nonmetropolitan poverty is higher than metropolitan. Using our conceptually-preferred adjustment, commuting-zone SPM, the difference between metropolitan and nonmetropolitan poverty rates is 1.3% points less than the OPM but larger than what the disparity is under the official SPM, county-sensitive SPM, or NASEM-SPM. The largest gulf emerges when using the unadjusted SPM, where metropolitan poverty falls, and nonmetropolitan poverty rises (11.6% vs. 16.1%).

We caution that the quantitative values of these estimates should not be the primary determinant for arguments concerning a given geographic adjustment. Rather, the choice of which adjustment to use is a conceptual decision, and we assert that commuting zone adjustments are the most valid in this regard. Based on these considerations, we present the following comparisons without including county-sensitive SPM—which we consider inferior to both the official SPM and commuting zone SPM due to concerns of over-fitting—nor NASEM-SPM—which has significant validity concerns due to the coarse approach to metropolitan areas versus the precise county-level nonmetropolitan adjustments.

Figure 3 displays national estimates by four ethnoracial groups. We retain the OPM and unadjusted SPM for comparison but focus our discussion on official SPM vs. commuting zone SPM. When comparing these two estimates by race, a consistent trend begins to emerge. When switching to commuting zone SPM, metropolitan poverty for the four ethnoracial groups tends to drop, and nonmetropolitan poverty for those same groups tends to rise. This results in a narrowing disparity between a single ethnoracial group if metropolitan poverty was higher in the official SPM, or a widening if metropolitan poverty was lower. Thus, commuting zone SPM suggests poverty rates between metropolitan and nonmetropolitan populations are more similar among American Indian and Hispanic populations, but less similar for Black and White populations, than estimates based off the official SPM would have us believe.

Fig. 3.

National poverty estimates by race/ethnicity.

Source: 2023 CPS-ASEC via FSRDC Project Number 2972. Point estimates for 2018-2022 available in Table S5

Turning to regions (Fig. 4), we find a similar story. Across all four regions, switching to commuting zone SPM lowered metropolitan poverty and raised nonmetropolitan poverty. However, it is not as if nonmetropolitan poverty is always higher than metropolitan poverty under the commuting zone adjustment. Under this alternative approach, metropolitan poverty is higher in the Northeast and Midwest but lower in the South and West. In fact, across the racial and regional estimates, commuting zone SPM only resulted in a change in direction for the metropolitan-nonmetropolitan poverty differential in one region—the West. In all other cases, the differential simply narrowed and/or widened.

Fig. 4.

Regional estimates of poverty.

Source: 2023 CPS-ASEC via FSRDC Project Number 2972. Point estimates for 2018-2022 available in Table S3

Finally, Fig. 5 presents SPM estimates by RUCC. The OPM estimates look as we would expect under that measure, with a clear gradient from rural to urban. However, the SPM takes a distinctly different pattern. When looking at official SPM by RUCC, we see a more nuanced story with higher poverty in the most urban counties (RUCC 1), lower poverty in the less dense metropolitan counties (RUCC 2–3), higher poverty in the more urbanized nonmetropolitan counties (RUCC 4–6), and then the lower poverty again in the smallest counties (RUCC 7, 9). The exception to this is RUCC 8—counties with an urban population of less than 5,000 that are adjacent to a metropolitan area—where poverty was the highest of all RUCCs across all three versions of SPM. For example, under the commuting zone SPM, the poverty rate in these counties is 16.6%. This high level of poverty in RUCC 8 is previously unreported in the wider literature and warrants further investigation. Future research should examine the components of the SPM to determine what specific factors are driving this spike in poverty. To be clear, elevated poverty in RUCC 8 is not a product of the geographic adjustment, as it holds across all three SPM formulations. Rather, it appears that these small outlying counties of metropolitan areas serve as zones of elevated, yet potentially under-appreciated, poverty nationwide.

Fig. 5.

Estimates of poverty by Rural-Urban Continuum Code.

Source: 2023 CPS-ASEC via FSRDC Project Number 2972. Point estimates for 2018–2022 available in Table S2

When comparing estimates across RUCCs between official SPM and commuting zone SPM, we see a slightly different pattern than what we saw previously. While all nonmetropolitan poverty estimates did rise, poverty in the smaller two levels of metropolitan counties (RUCC 2, 3) also rose. This suggests that the decreases observed for metropolitan poverty under the commuting zone adjustments in the race and regional estimates are likely a reflection of what is occurring in the largest metropolitan counties.

Discussion and Conclusion

In this paper we have presented the conceptual case for a revised geographic adjustment in the SPM, arguing in favor of implementing adjustments at the commuting zone level. Further, we have used federally restricted data to demonstrate the feasibility and utility of our approach. At present, we conclude that the current geographic adjustment in the SPM is imprecise for accurately capturing variation in cost of living across labor markets in the United States. While a fully county-sensitive approach overfits to the data and the NASEM-recommended approach does not satisfy concerns of conceptual validity, the community zone-level adjustment strikes an appropriate balance between precision and generalizability. This is reflected in our estimates, which present metro-nonmetro poverty differentials more in line with conventional expectations regarding urban and rural poverty—while still acknowledging that the OPM likely overstates the extent of rural poverty. The use of these more precise and conceptually appropriate adjustments means we are more accurately capturing whether someone should be considered poor relative to the labor market they live in.

Although the overall national poverty estimates in Table 1 do not appreciably change under our preferred geographic adjustment, we would like to stress the real-world implications of these adjustments, particularly as they relate to sub-national estimates and rural areas. Our estimates suggest that the official SPM is likely understating nonmetropolitan poverty, at least to some extent. Given the increasing importance of the SPM for informing policy, inaccurately characterizing hardship in rural areas risks a misallocation of both resources and attention from state and federal governments. While we do not want to overstate rural poverty, as would happen if we used no geographic adjustment, ensuring accurate representation of hardship across space is essential for the development and administration of effective policy. Across our estimates, metropolitan poverty generally changed less than nonmetropolitan when switching to commuting zone SPM. This is relatively unsurprising for several reasons. First, many MSAs likely have strong overlap with their commuting zone, leading to small changes between estimates. Within the commuting zones produced by Fowler (2024a), only 36 out of 387 MSAs are split across multiple commuting zones. Second, the official SPM approach allowed for a more targeted adjustment for many metropolitan counties, meaning they were already being more accurately reflected within estimates. Third, the use of commuting zones, as seen in Fig. 1, generally brings nonmetropolitan counties into clusters of metropolitan counties that were already together in the older approach. In brief, the official SPM’s strategy of pooling all nonmetropolitan counties within a state allows very remote rural counties to bring down the geographic adjustment of those nonmetropolitan counties more integrated into metropolitan labor markets.

The importance of these adjustments for our understanding of rural poverty is also made visible by the changes observed in the regional and racial estimates, where we observe notable changes between official SPM and commuting zone SPM. In both cases, population clustering of poor and non-White populations is likely driving sensitivity of these alternative approaches. Poor and non-White populations tend to be geographically clustered in the United States (Weber & Miller, 2017). While the majority of non-White populations live in urban areas, urban estimates for racial and ethnic groups remain less variable due to the reasons discussed in the prior paragraph. However, this clustering is likely playing a significant role in differences between these estimates in rural areas. For example, rural Black populations are found predominately in the South and are generally highly clustered within counties hallmarked by persistent poverty (Weber & Miller, 2017). As such, rural estimates for NH-Black and the South change more between official SPM and commuting zone SPM than is seen for other groups due to the fact that the statewide adjustment used by official SPM washes over pockets of hardship found throughout the region.

In terms of implementation, there are several considerations requiring further scholarly attention. First, the necessity of using restricted data potentially poses difficulties for adopting this approach. It remains unrealistic to expect any and all researchers to work within the Federally Statistical Research Data Center environment. That said, this concern seems minor when considering the SPM, specifically, due to restricted data only being needed when determining the geographic adjustment. Thus, the Census Bureau could easily generate SPM values with restricted data and then release those estimates in a similar manner as they have previously. Even if the Census Bureau only released the ultimate SPM-unit-specific threshold—and not the relevant MRI—the fact that the restricted-use ACS is also required for calculating MRI’s would mean that back-solving other details would be difficult. The use of commuting zones also offers an element of disclosure risk reduction because states contain many commuting zones and commuting zones, for the most part, are comprised of more than a single county. That said, if the Census Bureau were to adopt this approach but only release aggregate estimates of poverty headcounts, it would be a significant blow to poverty research in the United States. It is vital that poverty researchers are able to assess specific elements of the measure and how those thresholds impact aggregate estimates. Given the increasingly cautious approach to disclosure avoidance we have seen from the Census Bureau in recent years, it is vital that any implementation of this approach does not further set back poverty research in the United States.

Beyond the considerations of restricted data, it is also important to note that transitioning to our proposed measure would represent a series break from prior SPM reports. This means that any broader revision will likely need to be projected backwards in order to yield estimates for prior years. Although the existence of decanally-updated commuting zone boundaries from 1980 onwards makes this possible, future research will need to evaluate the tradeoffs between different vintages of commuting zone classifications.

Work by the panel behind the NASEM report suggests that there is an interest among the research and policy community to revise and improve the SPM. As such, our hope is that those making revisions consider and potentially adopt the geographic adjustment we have presented. The decisions surrounding geographic adjustments remain conceptual and there is no truly objective statistical test we can perform to demonstrate one is more correct than another. Thus, we have made the case for commuting zones as the appropriate scale of adjustment due to their ability to balance precision and generalizability across labor and housing markets. Although we are confident in our conceptual argument, future research should attempt to evaluate how well different versions of the SPM do, or do not align with other measures of hardship. Although these types of comparisons also rely on conceptual assumptions, longitudinal comparisons with other statistics like food or housing insecurity would provide valuable descriptive information on how well these measures map onto other conventional measures of material hardship. That said, it is important to note that using a geographic adjustment at all is a conceptual decision. While income and housing do vary nationwide and within states, our increasingly global economy means that localized price differentials have potentially become less relevant in many people’s lives. Future research should seek to understand how changes in our global markets are impacting the relevance and importance of geographic variation in cost of living.

Finally, the decision to reduce poverty into a simple yes/no binary is a conceptual one, reflecting the fact that a robust measure of poverty is essential for understanding well-being nationwide and developing, implementing, and evaluating policy. Although we often accept this reduction as a matter of fact, other approaches, such as the work on capabilities by Amartya Sen (2000), exist and it is important to remember that any measure reducing poverty to a simple rate will lose nuance across space and place. While it is essential to properly document hardship across the United States, no measure will ever be perfect. Capturing poverty via a conceptually robust and methodologically sound measure is vital for the well-being of Americans and the development of effective policy and programs. Realigning the SPM to a more valid conceptualization of economic geography is essential for that effort.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

Any views expressed are those of the authors and not those of the U.S. Census Bureau. The Census Bureau has reviewed this data product to ensure appropriate access, use, and disclosure avoidance protection of the confidential source data used to produce this product. This research was performed at a Federal Statistical Research Data Center under FSRDC Project Number 2972. (CBDRB-FY24-P2972-R11654). This study was supported by the National Institute for Minority Health and Health Disparities of the National Institutes of Health under Award Number U01MD017700. This content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Declarations

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.