Disclaimer: Not all the authors of the research papers and conference presentations are affiliated with BLS. This information is provided for your convenience and does not necessarily reflect the views or policies of BLS.
Since 1995, the Bureau of Labor Statistics (BLS) has been conducting research on the development of expenditure-based poverty thresholds. The earliest work focused on the development and production of what are referred to as the National Academy of Science (NAS) Thresholds. The BLS production of these thresholds are based on recommendations of the 1995 NAS report Measuring Poverty: A New Approach (Citro and Michael 1995). In 2010, an Interagency Technical Working Group (ITWG) provided a framework for a second set of poverty thresholds as part of a new Supplemental Poverty Measure (SPM). The ITWG's recommendations are outlined in the document "Observations from the Interagency Technical Working Group on Developing a Supplemental Poverty Measure". This was followed by the creation of a second ITWG focused on the SPM in January 2016. One of the goals of this group was to consider improvements to the SPM. Based on research conducted at the BLS and U.S. Census Bureau since 2010, on September 30, 2020 the SPM ITWG voted on a series of changes to be made. For a summary of the changes to the thresholds and resources with impacts on poverty statistics visit the Census Bureau website. To examine the impact on SPM thresholds, included on this website is an additional set of 2019 thresholds (referred to as “revised”) that show the impact of each change relative to the 2019 previously published thresholds. All of the ITWG SPM approved changes are implemented with the production and release of the 2021 SPM poverty statistics in June 1st 2022.
A guiding principle included in the SPM documents is that resources and thresholds be consistently defined in the development of the SPM. Individuals are considered poor when the consumer unit or household in which they live do not have resources to meet their needs as defined by the SPM thresholds.
Both the NAS and ITWG documents refer to the U.S. Consumer Expenditure Survey (CE) as the basis for these expenditure-based poverty thresholds. The documents also note that the BLS is responsible for conducting research on expenditure-based poverty thresholds, and for providing these thresholds to the U.S. Census Bureau for use in producing research poverty statistics. The ITWG acknowledged that the BLS had produced the NAS thresholds in the past and expected that the BLS would continue to play this role for the SPM.
Within the BLS, the Division of Price and Index Number Research (DPINR) conducts all expenditure-based poverty threshold research. Support regarding the CE data is provided, as needed, by staff within the Division of Consumer Expenditure Surveys. DPINR research is conducted in consultation and cooperation with U.S. Census Bureau researchers.
Research Poverty Thresholds are presented with a caveat: what appears on this BLS web page does not reflect the rigors of production quality thresholds or related statistics. For such thresholds to be produced, a broader BLS endeavor would need to be created that coordinates the development of improvements in, and the production and dissemination of, expenditure-based SPM thresholds. This effort would include support for: research economists to devise and test suggested improvements in the thresholds and share this research with the economics and statistics profession at large, as well as the general public; IT staff to design, code, test, and provide diagnostic statistics; statistical methods staff to develop measures of data and statistical quality; and economists to analyze the data, produce the thresholds and related statistics, and disseminate the thresholds to the public. Currently, the BLS produces the SPM thresholds using CE Interview data as an experimental research product, since production quality thresholds cannot not be produced within existing resources.
On this web page, recently produced BLS-DPINR Research Poverty Thresholds are presented along with papers and presentations related to these. Much of the research was conducted by the BLS in cooperation with U.S. Census Bureau staff and other academic researchers. Again, as noted above, the thresholds developed and described in the research papers and conference presentations are not produced using standard BLS production procedures
Starting in May 2019, this BLS website is the host for the Research NAS Poverty Thresholds. Through July 2019 these thresholds also appeared on the U.S. Census Bureau website. Similarly, the U.S. Census Bureau website also hosts the official measure.
The official poverty measure of the United States was first developed in the early 1960s and adopted as "official" in 1969. The official poverty threshold was determined to be the dollar value of a minimally adequate diet times three. The multiplier of three was used because 1955 Food Consumption Survey data showed that food expenditures accounted for one-third of after-tax income for an average family with children. An annual threshold of about $3,100 for a family with two adults and two children was set as the standard of need for 1963, and has been fixed in inflation-adjusted terms since then. The U.S. Census Bureau is responsible for publishing official annual poverty thresholds, rates, and other statistics.
The NAS recommendations provide the framework for a definition of the SPM. However, research over the years has suggested modifications to the NAS recommendations; the modifications are discussed in detail in the ITWG recommendations document referenced earlier. The SPM is not intended to replace the official poverty measure, but is to be considered a work in progress, with the expectation that there will be improvements to it over time. Changes in the SPM are to be decided upon in a process led by research economists, survey methodologists, and statisticians within the U.S. Census Bureau in consultation with the BLS, other appropriate data agencies, and outside experts, and will be based on solid analytical evidence.
Following a NAS report recommendation, the Office of Management and Budget (OMB) in January 2016 convened a new SPM ITWG to provide advice on challenges and opportunities brought before it by the U.S. Census Bureau and the BLS concerning data sources, estimation, survey production, and processing activities for development, implementation, publication, and improvement of the SPM (Renwick and Fox 2020). For further information refer to Census’ website Research on resources and resulting poverty statistics can be found on the U.S. Census Bureau website. The new methodology is described below in the section, “New Methodology to Produce the SPM Thresholds.”
Presented in this section are the methodologies to produce the previously published SPM thresholds (for 2005-2019) and those that incorporate the changes approved by the SPM Interagency Technical Working Group (TWG) on September 30, 2020 (for 2019 revised – 2021). For both, five years of quarterly data from the U.S. Consumer Expenditure Survey (CE) Interview data are used. Both start with out-of-pocket expenditures for food, clothing, shelter, and utilities (FCSU) to which a multiplier is applied to account for the expenditures of other basic goods and services, like those for household supplies, personal care, and non-work-related transportation. For the earlier methodology, telephone services were included in utilities. However, with the newer methodology, telephone services are separated from other utilities, and internet services and the value of select in-kind transfers (FCSUti) are added. With the addition of in-kind benefits, SPM thresholds are more consistently defined with respect to resources than those published previously. The earlier methodology did not sufficiently account for the receipt of in-kind benefits. The SPM resource measure produced by the U.S. Census Bureau includes the value of in-kind benefits associated with the following programs: Low Income Housing Energy Assistance Program (LIHEAP), National School Lunch Program (NSLP), Supplemental Nutrition and Assistance Program (SNAP), Women, Infants, and Children Program (WIC), and rental assistance. For the CE, since SNAP benefits are delivered as electronic benefit transfers, they are assumed to be used in the same way as other forms of cash or credit to purchase food, and thus to be included in reported out-of-pocket food expenditures. Given this, in-kind benefits included in the 2005-2019 thresholds were limited to those associated with SNAP. However, with the publication of SPM thresholds in September 2021, benefits associated with LIHEAP, NSLP, WIC, and rental assistance are added to FCSU out-of-pocket expenditures at the consumer unit level before the thresholds are estimated. Thresholds for 2019 revised, 2020, and 2021 account for the addition of these in-kind benefits.
Thresholds are based on the distribution of the of FCSU or FCSUti, depending on the threshold year. Before the distribution is produced, the values of these are converted to threshold year dollars for consumer units composed of two adults and two children, the reference group. While the SPM estimation sample is composed of consumer units with children, their expenditures are converted to those of the reference group using a three-parameter equivalence scale. For each consumer unit, the values for FCSU and FCSUti are converted to annual threshold year dollars by multiplying each quarter of data by four and updating using a consumer price index. The All-Items CPI-U was used to update quarterly CE expenditure data for the production of the 2005-2019 thresholds. A new composite CPI-U index, based on FCSUti, is used to update FCSUti expenditures for the production of the 2019 revised-2021 thresholds. In the section are description of the three-parameter equivalence scale, price indexes, and threshold estimation for housing tenure groups defined as owners with mortgages, owners without mortgages, and renter.
The previously published Research SPM thresholds were based on a range of FCSU expenditures centered on the 33rd percentile using data from an estimation sample composed of consumer units with exactly two children and any number of adults. These expenditures were equivalized to the expenditures of a consumer unit with two adults and two children through the use of the three-parameter equivalence scale proposed by Betson (1996). The parameters allow for the differing needs of adults and children and for economies of scale of consumption within the consumer unit. A distinguishing feature of the three-parameter equivalence scale is the adjustment for single parents; no adjustment for single parents was included in the two-parameter scale proposed by the NAS Panel. The three-parameter equivalence scale has been used in the production of the NAS and SPM thresholds in the past (e.g., Garner 2010). Directly below, we present the three-parameter equivalence scale that is applied to the estimation sample for the production of the BLS-DPINR Research SPM Thresholds:
Single adults with children scale = (1 + a + ß(K-1))f
Multiple adults with children scale = (A + ßK)f
a = parameter to account for the needs of the first child,
ß= parameter to account for the needs of additional children,
f = parameter to account for economies of scale within the consumer unit,
A = number of adults within the consumer unit, and
K = number of children within the consumer unit
The parameters a, ß, and f were estimated by Betson to fit the literature on the cost of children, and when rounded, were 0.8, 0.5, and 0.7, respectively.
For poverty measurement, the three-parameter equivalence is also used to convert the two-adult two-child SPM thresholds to thresholds for consumer units with differing numbers of adults and children. For consumer units with children, the equivalence scales presented above are used. For one and two adult consumer units, the equivalence scale equals (A)0.5. To produce thresholds for two adults, the scale is set to 1.41.
To produce the reference unit thresholds, the most recent five years of CE Interview data were used, with the earlier years of data updated to threshold year dollars using the All-Items CPI-U. Two adult-two child FCSU expenditures were ranked to identify the point in the distribution upon which to derive the thresholds. The 2010 ITWG Observations document noted that the 33rd percentile of FCSU equivalized expenditures would serve as the basis of the thresholds. However, the 33rd percentile is one point in the distribution, and thus, represents a single consumer unit who could be a renter, owner with a mortgage, or owner without a mortgage. To allow for the production of the three thresholds, a range of expenditures around the 33rd percentile was used, this being the 30-36th percentile of two adult-two child FCSU expenditures. The previously published Research SPM Thresholds for renters, owners with mortgages, and owners without mortgages were produced using the equation below.
SPM ThresholdEh =1.2 * FCSUE - (S + U)E + (S + U)Eh
1.2 = multiplier used to account for expenditures for other basic goods and services, like those for household supplies, personal care, and non-work related transportation.
FCSU, S, and U refer to the means of the sum of expenditures for food, clothing, shelter and utilities, and the shelter and utilities portions of FCSU, respectively, for the estimation of sample CUs within the 30th to 36th percentile range of FCSU expenditures.
E refers to consumer units in the estimation sample within the 30th to 36th percentile range of FCSU equivalized expenditures.
h refers to one of three housing tenure groups:
Owners with mortgages
Owners without mortgages, or
A noted earlier, changes approved by the ITWG SPM on September 30, 2020 underlie the new methodology used to produce the SPM Thresholds published in September 2021. The decision to implement these changes was based on work from Fox and Garner (2018), Garner et al. (2019), and Garner (2020). The following changes go into effect with the release of the 2020 thresholds.
The base of thresholds has been moved from the FCSU and S+U averages within the 30th-36th FCSU expenditure percentile range to 83 percent of the averages inherent within the 47th-53rd FCSU percentile range.
The estimation sample has been expanded from consumer units with exactly two children to consumer units with any number of children.
The CE Interview data are lagged by one year.
Imputed in-kind benefits from LIHEAP, NSLP, WIC, and rental assistance from government sources are added to the thresholds.
Telephone service expenditures are no longer geographically adjusted; in other words, they are no longer included with other utilities but are instead included along with food and clothing.
Home internet service expenditures have been added to the commodities.
To adjust the five years of CE Interview data to threshold year dollars, the All Items, Consumer Price Index for All Urban Consumers (All Items CPI-U) has been replaced by a composite “Food, Clothing, Shelter, Utilities, and telephone and internet service” price index; this is referred to as the FCSUti CPI-U.
As with the previously published SPM thresholds, five years of data are used to produce the thresholds under the new revised methodology. However, a decision was made by the ITW SPM to lag the CE data by one year. The new thresholds for 2020 are based on data collected in the CE Interview from 2015 quarter two through 2020 quarter one to represent the years 2015 through 2019. The reference period for each quarter is the previous three months. Lagging the expenditure data is preferred since the methods to impute the in-kind benefits for LIHEAP, NSLP, and WIC are based on data from the Current Population Annual Social and Economic Supplement (CPS ASEC). CPS ASEC data for the threshold year, let’s say 2020, are not released until after the 2020 SPM thresholds are needed to produce the U.S. Census Bureau poverty report.
Incorporating the seven changes, the following formula is used to produce the thresholds:
SPMEh = 0.83 * (1.2 * FCSUtiE - SUE, w/o telephone + SUEh, w/o telephone)
with subscripts defined as earlier. With telephone separate from household utilities and with the addition of internet, we now refer to the base of the thresholds as FCSUti. Also, rather than the 30-36th percentile, the revised methodology uses the 47th to 53rd percentile range; this represents the median. The percentage of the median chosen, 83 percent, is based on two factors: (1) the NAS guideline that a percentage of the median be used to produce the poverty threshold; and (2) 83 percent of the median, estimated when all of the other changes are implemented to produce a revised 2019 set of thresholds, results in an overall poverty rate that is closest to the previously published poverty rate for 2019 (Burns and Fox, 2021).
In-kind benefits are imputed to consumer units in the CE for all but SNAP. For rental assistance, rent paid and housing units characteristics in the CE are available to support the imputation of benefits. However, for LIHEAP, NSLP, and WIC no data are collected in the CE, but consumer unit characteristics are available and thus can be used to impute values from another source. In this section, the imputation methods used to impute the values of rental assistance and the listed in-kind benefits are briefly described. Detailed technical documentation for the imputation methods is forthcoming in 2022.
Imputed values of rental assistance are produced for consumer units who report receiving government assistance in paying rent, those who live in public housing, and those who live in rent-controlled units; rental assistance for these types of housing can be referred to as public rental assistance. Due to a coding mistake, 2019 and 2020 thresholds also include in the imputed rental assistance portion renter consumer units who receive part of their pay as rent. Starting with 2021 thresholds, this inconsistency was corrected. The CE rental sample was divided into two subsamples: one including CUs identified as receiving rental assistance, and another as those not receiving this assistance. To produce imputed rent values for the rental assistance subsample, a limited dependent variable model is used. The rent dependent variable is assumed to be censored from above for the rental assistance sample.
CE imputed values for LIHEAP, NSLP, and WIC are based on reports from the Current Population Survey Annual Social and Economic Survey (CPS-ASEC). Multiple imputation models are used to assign LIHEAP benefits, and NSLP and WIC participation from CPS-ASEC households to CE consumer units. An underlying assumption for this imputation approach is that both the CPS-ASEC and CE samples are both representative of the sampled U.S. population, and thus data from the two surveys can be pooled. In the pooled sample, values for LIHEAP benefits, NSLP participation, and WIC participation exist for the CPS-ASEC but are missing for the CE. Using comparably defined household and consumer unit characteristics, regression-based models are used to assign CPS-ASEC recorded non-missing LIHEAP benefits, participation in the NSLP, and participation in WIC to the CE which, by design, has missing values for these programs. Benefit values for consumer units participating in NSLP and WIC are assigned using published data for these programs from the U.S. Department of Agriculture (USDA).
The price index used to adjust quarterly CE expenditures to threshold year dollars is referred to as the FCSUti CPI-U. The FCSUti index is processed as an annual average for the urban population. Upper-level index estimation for the FCSUti index is comparable to the CPI-U as a modified Laspeyres formula. See the Chart below for the FCSUti CPU-U indexes relative to the All-Items CPI-U indexes; the All-Items indexes were used in the production of previously published SPM thresholds.
To test for a significant change in the threshold from the previous year, or to make a comparison between thresholds within a year, one would conduct a Z-test. The test statistics are specified below, for each type of comparison. First, to test for the statistical difference in thresholds from one year to the next (e.g., SPM renter thresholds in 2020 as compared to 2019 revised), simply divide the difference in the thresholds for time t and t-1 by the standard error of the year-to-year difference.
Z Renters, t , t-1 = (SPM Renters, t - SPM Renters , t-1 ) / Standard error Renters, t, t-1
For a statistical comparison of thresholds within year t (e.g., renter thresholds compared to owner without mortgage thresholds), simply divide the difference in the two thresholds within the year by the standard error of the difference between the two housing tenure groups that is listed for the current threshold year.
Z Renters compared to Owners without mortgages , t =
(SPM Renters, t - SPM Owner without mortgages , t ) / Standard error Renters compared to Owners without mortgages, t
Research SPM Thresholds for 2009-2010 were first posted to the BLS website in table format in November 2011. Thresholds for 2005-2008 were added shortly thereafter. Ever since, the time series has been supplemented by an additional year's threshold each year after the release of CE public use data.
The BLS-DPINR Research SPM Thresholds and associated standard errors can be found through the links below. Also available are the expenditure shares of each of the components of the thresholds. Please note that use of all of the significant digits presented in the spreadsheets are necessary for inclusion in calculations; precision will affect the resulting dollar value.
The weighted share distribution of consumer units by housing tenure for the SPM thresholds for each year is included in the "BLS-DPINR Research SPM Thresholds with Housing Tenure Shares" file. This is included to facilitate the calculation of a weighted average of the three SPM thresholds for users who are interested in FCSU threshold that does not account for housing tenure. Note: The ITWG included the recommendation for the production of three housing tenure thresholds, not a single threshold. A weighted average of a particular component of the SPM thresholds (i.e., food, clothing, shelter, utilities) can be computed using the housing tenure distributional weights along with the housing tenure specific component from the "Expenditure Shares" file below. For example, a weighted average of Shelter across the three housing tenure groups would be calculated as follows:
S W = (P Owners with mortgages * S Owners with mortgages * SPM Owners with mortgages)
+(P Owners without mortgages * S Owners without mortgages * SPM Owners without mortgages)
+(P Renters * S Renters * SPM Renters)
S W = weighted average of shelter share in SPM threshold
P = weighted share distribution of consumer units by housing tenure
S = shelter component share of SPM threshold for housing group
SPM = SPM threshold for housing group
Last Modified Date: November 6, 2023