Creating an Integrated System of Data and Statistics on Household Income, Consumption, and Wealth: Time to Build (2024)

3

Statistics for Policy and Public Understanding

While other chapters discuss the development of an integrated microdata system to support analyses with varying definitions of income, consumption, and wealth (ICW), this chapter discusses the necessary statistics. That is, it discusses the established estimates of household expenditures (household consumption statistics are under development), income, and wealth produced by federal agencies for general public consumption (see Box 3-1).¹

Such statistics serve a useful purpose, as estimates that can be tracked over time and across population groups and provide a common basis for understanding. These statistics are critical in understanding the changes in economic wellbeing and inequality in the United States, as discussed in Chapter 1. Researchers, meanwhile, can construct alternative definitions and household groupings from microdata available to them in secure ways or focus on specific variables or subjects in the underlying data, and statistical agencies can learn from their own and others’ research when it may be desirable to modify the established definition(s) or produce additional estimates for context. With the new statistics presented in this chapter, research

___________________

¹ The United States designates principal statistical agencies and recognizes statistical “units” (the term in the Foundations for Evidenced-based Policymaking Act of 2018) in other agencies (e.g., the Federal Reserve Board has a recognized statistical unit with responsibility for the Survey of Consumer Finances). The U.S. government does not use the term “official” or “established” statistics. The panel’s use of “established” is to identify statistical series that share characteristics that would lead policy makers and the public to turn to them first for reliable, useful information.

and policy makers will be better informed about the relationship between ICW and households’ respective responses to policy changes.

This chapter first summarizes material in Chapter 2 on the subset of definitions of household expenditures, consumption, income, and wealth that would qualify for use in new and improved established statistics on household distributions to better understand inequality and other aspects of household economic wellbeing. It then identifies desirable geographic, demographic, and other types of breakdowns for which established statistics should be produced for maximum utility. These breakdowns include underlying characteristics, such as the unit of analysis, which may include households, families, and other types of units. The chapter thereby identifies characteristics that need to be present in the microdata records in the panel’s recommended integrated data system on household ICW. The chapter’s recommendations identify (a) short-term priorities for statistical agencies to improve their current estimates, considering household consumption, expenditures, income, and wealth separately and (b) steps for meeting a longer-term goal—namely, to produce annual estimates of the joint distribution of these elements of household economic wellbeing.

BACKGROUND

The United States’ federal statistical agencies have a long history of producing estimates of household expenditures, income, and wealth. Interest in measuring the standard of living, both over time and across areas and population groups, dates to the late 19th century. BLS fielded the first survey of consumer expenditures in 1888, just 4 years after it was established as a federal statistical agency. Expenditure surveys were conducted at irregular intervals (eight in all) until 1980, when the current Consumer Expenditure Survey (CE) became continuous. BLS currently publishes 12-month consumer unit expenditure estimates every 6 months from the CE survey, and recently it has produced estimates for a new measure of consumer unit consumption (see Garner et al., 2023).²

Interest in household income, and the effects of government income support programs beginning with Social Security and unemployment insurance, dates to the Great Depression. The 1940 census was the first to ask about income, specifically wages and salaries and other sources of more than $50, and the Census Bureau began regular annual publication of household money income estimates in 1947. BEA (and its predecessor agency) first published “size distributions” of Personal Income (PI) in 1953, 6 years after it began publication of aggregate PI and other components in

___________________

² Chapter 2 describes the difference between expenditures and consumption.

the National Income and Product Accounts. The distributional estimates were discontinued after 1963, except for estimates produced for a few years in the early 1970s. The reason was lack of resources, which left the nation until recently without distributional estimates for households of its most comprehensive income measure. The Statistics of Income of the Internal Revenue Service (SOI/IRS) has published annual estimates of adjusted gross income (AGI) for tax filing units since 1999.

Interest in household wealth (assets and liabilities), together with consumption and income, led FRB to launch the Survey of Consumer Finances (SCF) in 1983. Due to the oversample of high-income households, the SCF has been the premier source of data on household wealth in the United States and in the examination of wealth disparities. The SCF takes care, through its sample design and editing and imputation procedures, to represent the totality of the income and wealth distribution (except for the wealthiest 400 people—see Table 4-1 in Chapter 4). However, its ability to produce annual estimates is limited since it is fielded only once every 3 years, and its small sample size limits its utility for examining differences across demographic groups.³

The Great Recession and its aftermath saw a revived interest in distributional estimates of household consumption, income, and wealth to understand levels and trends in inequality and for other purposes. The lack of household distributional estimates obscured the erosion of wages for less-skilled workers and the increased debt burden for middle-class America even as measures of real consumption expenditure, income, and wealth were climbing in the aggregate.⁴

In response, statistical agencies embarked on research to develop and improve distributional estimates (see Section 3.2 below). In 2022, BLS in collaboration with BEA produced a prototype set of household estimates of personal consumption expenditures (PCE), and BLS produced an initial set of estimates for household consumption (both sets of estimates use the CE as the underlying dataset). BEA produced prototype annual household estimates of PI beginning in 2020, and the Census Bureau published the first estimates from its National Experimental Wellbeing Statistics (NEWS) in February 2023 (both sets of estimates use the Current Population Survey Annual Social and Economic Supplement (CPS-ASEC) as the underlying dataset). NEWS is intended to improve the accuracy of the Census Bureau’s

___________________

³ The decennial census, beginning in 1850, for many decades typically included questions on home and farm ownership.

⁴ The Census Bureau has long published estimates of household money income, but their usefulness was undercut by high and increasing levels of nonresponse and underreporting of income in the CPS-ASEC, the lack of coverage of very-high-income households, and the fact that the money income definition omits many components of the comprehensive BEA PI measure.

household income statistics by using administrative records to adjust for substantial income reporting errors in the CPS-ASEC. Finally, FRB began publishing quarterly estimates of household wealth in 2019 using data from the SCF and the FRB Financial Accounts.

Internationally, OECD, to which the United States belongs, began similar efforts in 2011 to define and encourage member countries to develop household estimates of ICW (the part of income not consumed) based on the System of National Accounts (SNA).⁵ In 2023, OECD also began to develop guidance for estimates of household wealth in line with the SNA totals.⁶ In the United States, BEA made important strides in the past 15 years to achieve greater concordance with OECD definitions that correspond to PCE and PI, and BLS used the OECD definition in conceptualizing its proposed household consumption series.

ESTABLISHED ESTIMATES AND RECOMMENDATIONS FOR IMPROVEMENTS

Established Series on Household Expenditures, Income, and Wealth

Given the involvement of multiple statistical agencies in the United States in producing household economic statistics, it is not surprising that definitions of household expenditures, income, and wealth have varied among agencies and over time within agencies for at least two reasons. (As noted above, there has never been a household consumption series.) As discussed in Chapters 1 and 2, many established measures are constructed for specific purposes. As such, different agencies have decided whether to recognize new elements, such as in-kind benefits like Medicare and Medicaid, which began in the 1960s, as part of household income. Agencies have also decided whether to achieve greater comparability with an international standard, such as the SNA. The decentralization of the U.S. statistical system has made it harder for agencies to collaborate on such efforts as aligning definitions, although collaborative activities among BEA, BLS, and the Census Bureau have increased significantly in the past decade.

As shown in Table Tables 2-1A, 2-1B, and 2-1C in Chapter 2, alternative definitions of ICW are used by federal agencies. Table 3-1 summarizes

___________________

⁵ See Zwijneburg et al. (2021). Results in the paper are from 13 countries using the same framework and methods. See also the OECD (2021).

⁶ The expert group on distribution of household wealth was launched in view of the new G20 Data Gaps Initiative, see www.imf.org/en/News/Seminars/Conferences/DGI/g20-dgi-recommendations#dgi3. At the same time, OECD developed recommendations to further improve consistency and cross-country comparability of data on household ICW at the micro level. See OECD (2013a,b) and Balestra and Oehler (2023).

TABLE 3-1 Major Features of Established and Experimental U.S. Statistical Series on Household Consumption, Expenditures, Income, and Wealth

NOTE: AGI = adjusted gross income; BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; CE = Consumer Expenditure Survey; CPS-ASEC = Current Population Survey Annual Social and Economic Supplement; CU = consumer units; DB = defined benefits; DC = defined contributions; DFA = Distributional Financial Accounts; FRB = Board of Governors of the Federal Reserve Board; HI = household income; NEWS = National Experimental Wellbeing Statistics; NPISH = nonprofit institutions serving households; OECD = Organisation for Economic Co-operation and Development; PCE = personal consumption expenditures; PEU = Primary Economic Unit; PI = personal income; SCF = Survey of Consumer Finances; SNA = System of National Accounts; SNAP = Supplemental Nutrition Assistance Program; SOI/IRS = Statistics of Income Program of the Internal Revenue Service; SSI = Supplemental Security Income.

SOURCES: Table 2-1 in Chapter 2; this volume; Batty et al. (2019); Bee et al. (2023); Bureau of Economic Analysis (2023); Garner et al. (2023); Internal Revenue Service (n.d.).

the definitions and other major features of currently established household statistics and newly introduced experimental statistics. The panel assessed new and improved measures in terms of their adequacy to inform policy makers and the public that are mostly consistent with international estimates, are consistent across the three dimensions (ICW), satisfy the budget identity,⁷ and are consistent with national accounts. The panel also

___________________

⁷ As discussed in Chapter 2, one of the primary goals for data and estimates is to produce estimates that are consistent with the budget identity: I = C + S. Hence, the definitions used by agencies in their estimates for income, consumption, and wealth need to align with the identity. For example, if in-kind benefits are excluded from the income definition, then they should be excluded from the consumption measure. Hence, some definitions may not work well in presenting coherent information on ICW.

preferred statistics that have broad acceptance in the professional community and can easily be explained to policy makers.

Drawing on Table 3-1, the panel’s conclusions about established definitions and recommendations for improvements in the short term (next 3–5 years) are:

Conclusion 3-1: The initiatives by the Bureau of Economic Analysis, the Bureau of Labor Statistics, the Census Bureau, the Congressional Budget Office, and Division of Research and Statistics of the Federal Reserve Board to develop new and improved estimates of the distribution of household income, consumption, expenditures, and wealth, are laudable. An example is the Census Bureau’s National Experimental Well-being Statistics initiative to improve its household income estimates. These initiatives deserve support to become established series as soon as practicable.

Conclusion 3-2: Established definitions of household income, consumption, and wealth often match well with definitions in the System of National Accounts (as used by the Organisation for Economic Cooperation and Development). Such definitions include those by the Bureau of Economic Analysis for household personal consumption expenditures and personal income, the Bureau of Labor Statistics for consumer unit consumption, and the Federal Reserve Board for household wealth.

Suggested Changes in Measures

While the established measures are useful to evaluate the distribution of ICW, some changes may improve their usefulness. For example, the BEA measure of household income better represents the distribution of the household sector. Hence, excluding nonprofit institutions serving households (NPISH) in both PI and PCE makes sense. The distribution of NPISH may sometimes be combined with those of the household sector in NIPA for practical purposes. However, they refer to other units and, because their results cannot be attributed to individual households it is better, and more comprehensible for the public, to exclude them from the distributional results. (The PI and PCE series in Gindelsky & Martin, 2023, exclude NPISH.)

Other definitions of consumption/expenditures (as in Table 2-1) differ from PCE. The definition of AGI for tax filing units necessarily reflects the provisions of the federal income tax code; it is useful for comparisons with other definitions. The Census Bureau’s household “regular money income” definition does not accord with international practice or other established

income definitions and is overdue for revision as the basis for an established household income statistics series.

Instead of using money income or the AGI, a better household income measure more consistent with the budget identity would be an after-tax- and-transfer measure similar to that produced by the Congressional Budget Office (CBO; except without including capital gains).⁸ (See Box 3-2 for a discussion of accommodating programs that use money income.) Because a substantial component of income support in the United States comes through in-kind (as opposed to cash) benefits, and also from tax credits (which are captured in after-tax money income), money income does not give an adequate picture of the resources of people benefiting from income support programs. For example, in 2021, adding Supplemental Nutrition Assistance Program (SNAP) benefits to money income reduced the measured rate of child poverty by 2 percentage points—almost twice the reduction seen for older people, whose main sources of government support, Social Security and Supplemental Security Income (SSI), are already included in money income.⁹ On the other hand, by excluding employer retirement contributions for workers and including distributions for retirees, money income arguably provides a more useful picture of retirement income available for consumption or savings than the PI definition.

Recommendation 3-1: In the next 3 to 5 years, statistical agencies should build on their current initiatives to publish improved statistics of household income, consumption, expenditures, and wealth. Specifically:

• The Bureau of Economic Analysis (BEA), the Bureau of Labor Statistics (BLS), and the Census Bureau should move as quickly as possible, consistent with time and resources needed to assess quality, to compile distributional estimates for household personal income (PI), personal consumption expenditures, BLS comprehensive consumption, and income based on the Census Bureau’s National Experimental Wellbeing Statistics (NEWS). These statistics should be published at least annually and within 1 year of the time the underlying data are collected.
• BEA should make its distributional household PI estimates more useful by adding estimates for (1) disposable (after-tax) PI that excludes nonprofit institutions serving households (NPISH); (2) disposable PI (excluding NPISH) that adds retirement distributions and excludes retirement contributions (as in the System of National

___________________

⁸ This would be similar to the disposable household income concept recommended by the Canberra report (2011). The Census Bureau could also produce a pre-tax and transfer income series to show the effects of market income separately.

⁹ Calculated from Creamer et al. (2022). The poverty measure used is the official poverty measure, which is based on money income.

• Accounts definition), and (3) adjusted disposable income (similar to the Organisation for Economic Co-operation and Development).
• The Census Bureau’s household income estimates should include pre- and post-tax-and-transfer series, where transfers include in-kind benefits. NEWS should follow suit. The Census Bureau should publish money income estimates in appendixes to its reports for use by relevant programs.
• BLS, in cooperation with other economic statistical agencies, should expand and accelerate its program to publish annual estimates of household consumption that are definitionally consistent with previous recommendations.

Even with the changes in Recommendation 3-1, there will remain differences among definitions that are too difficult, for one reason or another, or not appropriate to eliminate. Consequently, it is important for the relevant statistical agencies to develop a collaborative program to regularly document definitional differences and their impact on distributional statistics.

The panel envisions a major report every 3 or so years jointly produced by BEA, BLS, FRB, SOI/IRS, and the Census Bureau. The document would describe the various definitions for household consumption, expenditures, income, and wealth and the differences in definitions for each among agencies and with SNA definitions. The document would also highlight empirical consequences of these and other differences (e.g., different estimation

methods, data sources). See Chapter 6 for how this report could fit under the governance structure that the panel envisions for an integrated system of household economic data and statistics. In addition, with the release of their established statistics (possibly annually), each agency would include in the explanatory material for its estimates and comparisons with estimates from the other agencies and any other relevant estimates (e.g., those for household income from CBO).

Recommendation 3-2: Relevant statistical agencies (Bureau of Economic Analysis, Bureau of Labor Statistics, Census Bureau, Statistics of Income Division of the IRS, and the Federal Reserve Board) should collaboratively publish a major report every 3 years that compares levels and trends among the household income, consumption, expenditure, and wealth statistical series using the improved measures as specified in Recommendation 3-1. In addition, their annual reports should compare their estimates to those from other agencies and, to the extent possible, identify the factors, including differences in definitions, that contribute to differences in estimates.

Other Statistical Estimates of Expenditures, Income, and Wealth

In addition to the established measures discussed in the previous subsection, there exists a wide variety of alternative measures on ICW, often serving different policy needs. BLS collects consumer unit income and assets and debts in the CE to cross-tabulate with expenditures. The Census Bureau collects household income information in surveys other than the CPS-ASEC, including the ACS and Survey of Income and Program Participation (SIPP); and it collects assets and debts in SIPP. The FRB Division of Research and Statistics collects primary economic unit household income, selected expenditures, and assets and debts in the SCF to produce a comprehensive assessment of consumer finances every three years. CBO has published approximately annual estimates of its definition of income and disposable (after-tax-and-transfer) income for households beginning in 2001 using the CPS-ASEC as the base.¹⁰

Different policy users with different needs require a variety of estimates on ICW. The ACS, SIPP, SCF, and BLS definitions of income are similar to the definition of money income, except that the SCF and BLS definitions include SNAP, BLS published income includes food and rent as pay, and SCF income includes realized capital gains. The CBO definition differs the most from money income but is closer to the after-tax-and-transfer income

___________________

¹⁰ For previous reports, see www.cbo.gov/taxonomy/term/1590/recurring-reports

measure recommended above for the Census Bureau to publish in place of money income. Thus, the CBO definition includes employer-provided health insurance, SNAP, Medicare and Medicaid, and other in-kind benefits (e.g., school lunch, Low Income Home Energy Assistance Program). The CBO definition also includes realized capital gains (see Table 2-1A in Chapter 2). Similarly, BLS, SIPP, and SCF definitions of assets and debts differ among themselves and differ from the distributional financial accounts (DFA) definition, and the SCF definition of expenditures differs from the BEA and BLS definitions.

Some of the current established estimates of ICW may have specific purposes and agencies may continue to produce these statistics. However, it would be very useful for agencies that produce these estimates of household consumption, income, and wealth to document differences in definitions and other features with the established and new and improved series and take steps to bring their definitions into alignment to the extent possible.

Joint Distributions of Household Income, Consumption, and Wealth

Having separate statistics on household consumption or expenditures, income, and wealth is useful for policy and public understanding. Expenditure data tell us how spending has changed over time and among types of households. For example, households today allocate less than half as much of their budgets on food and beverages at home as they did 60 years ago, but lower-spending households are constrained to devote more of their budget to food and beverages than higher-spending households (Bureau of Labor Statistics, 2023b; Marchesi & McLaughlin, 2024). Income data tell us how resources available for consumption or saving are keeping pace with inflation or changing in composition; for example, SNAP expenditures adjusted for inflation increased 147% from 1980 to 2019 (Tiehen, 2020). Wealth data (assets and liabilities) tell us how many households have adequate resources for retirement or for helping their children or grandchildren with education, housing, or starting a business.

Separate statistics, however, do not tell the entire story that policy makers and the public require. The joint distributions of household ICW are needed for full information and to answer research questions discussed in Chapter 1, such as: How do households that report lower income than consumption support their spending? Are they using loans or liquid assets? How do different households respond to government transfers? Which households are income-poor but asset-rich (and vice versa), and how do they weather recessions? That positions in the ICW distribution may be

largely driven by the life-cycle is another reason it would be important to look at these dimensions simultaneously.

The CE is the only federal survey that obtains ICW. Using the CE, one can create an augmented SCF that fully supports assessing the joint distribution of consumption (expenditures), income, and wealth for the full range of the distribution, including high-income households (see Fisher et al., 2022).¹¹ As noted previously, the SCF, which obtains income, wealth, and limited information on consumption, is only conducted every 3 years and has a small sample size (about 6,000 households in recent surveys). Recently, the SCF expanded the sample of people of color (see Moore & Pence, 2021); however, this would not provide a large enough sample of households to permit disaggregation by detailed race and ethnicity.

The panel puts forward as a critically important longer-term goal for the federal statistical system to develop an integrated dataset that can support estimates of joint distributions for characteristics of interest. These joint distributions are essential for researchers evaluating households’ response to cash transfers, such as the Economic Impact Payments, changes in their asset and debt portfolio, or their responses to changes in their retirement benefits. There are various ways in which this goal could be accomplished: for example, using statistical matching to add variables to the CPS-ASEC or the IRS’s tax return data or both, or adding a few questions to the CPS-ASEC or CE or both. The panel assesses these methods in Chapter 5. (See Chapter 6 for the panel’s suggested governance structure for the development of an integrated system of household ICW data that will support new and improved research and the established statistics of the future.)

Conclusion 3-3: A priority goal, requiring longer-term research and development, for relevant statistical agencies and coordinating mechanisms is to develop an integrated system of household income, consumption, and wealth data. The system needs to support improved estimates—to be the established estimates of the future—for each of these three dimensions of household economic wellbeing and joint distributions of all three dimensions for the same households. Differences in economic resources and living standards among households cannot be fully understood without considering at the same time and conjointly their income, consumption, or wealth.

___________________

¹¹ Two longitudinal surveys, the Panel Study of Income Dynamics and the Health and Retirement Study, include measures of ICW (see Fisher & Johnson, 2021) and the tables in Chapter 4.

FEATURES OF NEW AND IMPROVED ESTABLISHED STATISTICS

This section reviews dimensions of existing statistical series on household expenditures, income, and wealth that are key to their utility: the unit of analysis (e.g., household, family, consumer unit), the inflation adjustment method, the components provided (e.g., types of assets and debts), the economic levels and measures of inequality provided (e.g., deciles of households by percentage share of income, Gini index), the geographic detail provided, and the demographic detail provided. Each dimension is discussed below, in turn.

Existing series provide much useful information on these dimensions, but there are differences among series that make it difficult for users to compare them. Some series provide very little detail, while others provide extensive detail on some characteristics but not others. For example, the Census Bureau provides detailed breakdowns of money income levels for households and families by family/household type, educational attainment of the reference person, and other characteristics, cross-tabulated by race and ethnicity, but with no detail on components of money income except for earnings.¹² In contrast, FRB provides considerable detail on types of assets and debts but little detail on characteristics. Collaboration among agencies to develop a minimum agreed-on set of categories and measures—for example, race categories and inequality measures—and on additional measures where feasible would be helpful to produce new and improved versions of established statistics.

Detail for joint distributions of ICW will be most challenging to produce given the underlying data requirements. In practice, joint distributions (whether two-way as in consumption and wealth or three-way) will likely be less rich than one-way distributions. For example, it may be feasible to publish estimates of households in the top (bottom) decile of income who are also in the top (bottom) decile of consumption and wealth. It may not be feasible to publish similar estimates for the top or bottom percentiles nor to publish estimates of the full cross-distributions (e.g., households in each decile of income by their decile of consumption by their decile of wealth). Studies of joint distributions, however, have produced important results. Inequality (looking at quintiles and the top 5%) has increased more in the United States when looking at the two dimensional joint distribution of income and wealth two (or even or three dimensions) (Fisher et al., 2022). In addition. the distributions of consumption and income are more closely linked than those of consumption and wealth and the three distributions overlap more strongly at the top and bottom tails than in the

___________________

¹² The Census Bureau provides breakdowns of components in tables of money income for people but not for households or families. For more information, see www.census.gov/topics/income-poverty/income/data/tables/cps.html

middle (Balestra & Oehler, 2023, p. 13, comparing results from a number of OECD countries). Recently, BEA and BLS have collaborated to produce joint distributions of PI and PCE (see Gindelsky & Martin, 2023).

Unit of Analysis

Households, Families, and Other Units

People live in a variety of arrangements—alone, with roommates who may or may not share living expenses, and in families of people related to one another of varying complexity. The question is, considering this variety, what would constitute the best unit of analysis for improved established statistics on ICW? International guidance in both Canberra (2011) and the 2008 SNA recommends using the household as the unit of analysis. The 2008 SNA (4.149) states that “[a] household is a group of persons who share the same living accommodation, pool some, or all, of their income and wealth, and consume certain types of goods (mainly housing and food) and services collectively.” In general, each member of a household has some claim upon the collective resources of the household and some influence on the decisions affecting consumption or other economic activities. For these reasons, the household is regarded as an institutional unit in the SNA (Eurostat Commission (2008, p. 82).

The Census Bureau defines a household as all people living at the same address in a separate housing unit (e.g., an apartment, townhome, free-standing home), whether or not they are related and regardless of the extent to which they constitute an economic unit. This definition is straightforward conceptually and easy to identify in the field for data gathering. Moreover, empirically (see below), household members typically pool at least some of their income and wealth and consume housing, food, and some other goods and services collectively, which makes the household a natural unit of analysis for statistics on many topics, including ICW. The household, as defined by the Census Bureau, is used in money income statistics, in BEA’s distributional PI statistics, and in CBO’s income series, and it is also widely used internationally.

There is also policy and public interest in how families are faring, excluding household members who are not part of the family unit such as a boarder or a caretaker. For example, there is interest in comparing married and cohabiting couples by gender with and without children, single-parent families, multigenerational families, and the like. In its income and official poverty statistics, the Census Bureau does not treat cohabiting couples as

family units, but its SPM does.¹³ With a family definition, there is the issue of how to treat unrelated individuals. Census Bureau practice is to treat them separately even when they share income for expenditures on housing, utilities, and food. This treatment is not necessary—assuming the necessary data were collected, unrelated individuals could be grouped into economic units based on the extent of sharing, as is done in some other unit definitions such as BLS consumer units.

Other units of analysis beside the household or family are defined in some established statistics. BLS and BEA distributional PCE statistics and BLS’s prototype consumption statistics are based on consumer units (CUs), defined as people in a household who are either in a family of related members or who otherwise share at least two of these three types of expenditures: housing, food, and “all other” expenditures.¹⁴ DFA and SCF statistics are based on SCF primary economic units (PEUs), defined as the “economically dominant” single person or couple in a household and others economically dependent on them; limited information is collected about other people in the household. SOI/IRS statistics use the tax filing unit definition in the federal individual income tax code. Table 3-2 illustrates how households, families, CUs, PEUs, and tax filing units could differ for the same living situations.

Overlap Among Unit Definitions

In practice, there is considerable overlap among the various types of units. For example, BLS estimates that 97% of CUs include the entire household (see National Academies, 2023). In 2020, from the CPS-ASEC, the nation’s 128.5 million households were distributed as follows: 65% were two-or-more-person families (of which 5% included nonrelatives, treated as unrelated individuals), 28% were single-person households, 6% were households with two unrelated persons (who could be cohabitors), and 1% had three or more unrelated persons. Research preparatory to the adoption of the SPM (Provencher, 2011) found that in 2010 the OPM, based on families and unrelated individuals, estimated 131.9 million units of analysis compared with 124.2 million units for the SPM. The CE estimated

___________________

¹³ National Academies (2023, p. 39), recommends that the SPM move from SPM units (families, including cohabitors, and other unrelated individuals) to households as the unit of analysis.

¹⁴ The full specifications for a CU are (a) all members of a household related by blood, marriage, adoption, or other legal arrangement; (b) financially independent persons, including people living alone, sharing a household with others, rooming in a private home or lodging house or in permanent quarters in a hotel or motel; or (c) two or more people living together who use their income to make joint expenditure decisions (on two of the three major expense categories of food, housing, and all other).

TABLE 3-2 Examples of Units of Analysis Applied to Living Situations

NOTE: The current SPM unit definition would treat cohabiting couples and their children as 1 unit; National Academies (2023, p. 39) recommends that the SPM move to a household definition, which would treat all three examples as 1 unit each.

SOURCES: Panel generated using National Academies (2023) and survey information (see Tables 4-1).

121.1 million CUs for 2010 compared with 117.5 million households that year from the CPS-ASEC. Most of the decrease for the SPM compared to the OPM was due to treating cohabiting partners as parts of family units instead of treating them as unrelated individuals. The CU definition also treats cohabiting partners as one CU assuming they report sharing expenses.

The DFA and SCF primary economic units differ from households, families, and CUs in ways that are not altogether clear, except that many PEUs are smaller than the other unit types, on average. A three-generational

family, for example, is not recognized as such unless the grandparent(s) is financially interdependent with the parents (who may be cohabitors) and children. In 2020, around 8% of all children lived with their grandparents (Washington et al., 2023). On the other hand, about 2 million children live in “grand-families” where only grandparents and related grandchildren reside without the parental generation (Michelmore & Pilkouskas, 2022).

The SCF documentation states that PEUs are weighted to CPS-ASEC households, and published statistics use the term “families” in which 1-per-son PEUs are treated as 1-person families. The SCF, however, collects little information about unrelated individuals except those living alone—the documentation is not clear about how households with roommates are treated. Tax filing units outnumber households, but assuming older dependent children who file their own returns and married couples filing separately can be linked together, tax filing units are similar in concept to families and unrelated individuals.

Given the widespread use of the household as the unit of analysis and the close correspondence of consumer units to households, the panel concludes that having households (using the Census Bureau definition) as the main unit of analysis for new and improved household statistics on ICW makes sense. Families are also of considerable policy interest in their own right; but, when a family definition is used, it seems preferable to adopt something along the lines of the SPM approach, which defines family units as including cohabiting couples and foster children. The Census Bureau, then, might replace its family and unrelated individual tables with family households (treating cohabitors as families and singles as families) and then produce tables of family households without nonrelatives, of family households with nonrelatives, and of nonfamily households. (Further breakdowns of families by singles and couples with and without children would also be useful as noted above.) Family statistics would then aggregate to household statistics. BLS and FRB might want to consider adopting this breakdown as well. In any case, to facilitate the appropriate aggregation for the policy or research purpose, the underlying microdata need to tag each individual child and adult as to their membership in a household, a family, a tax filing unit, a CU, and so on (see Chapter 4).

Expanding our horizons to include inter-household or intergenerational transfers requires additional research and data sources that permit following people across time and place. Older and younger generations who are related may transfer resources across households. In some cases, such transfers are legally required, such as with child support; but in other cases they are voluntary to help support older units in need of caregiving or younger units forming their own households or needing support to pay their

college expenses. Reparations sent to relatives in foreign nations from U.S. immigrants are another type of transfer that is becoming pervasive (Darity et al., 2022). Finally, there is the issue of family complexity where children reside with only one biological parent with half-siblings and a step-parent. These cases of multi-partner fertility are often not recognized by policies or programs, which treat children and parents differently (Carlson et al., 2022; Michelmore & Pilkouskas, 2022). Clearly, while households make the most sense as a sharing unit at any point in time, additional research and more systematic evaluation and research on improving the measurement of family and living arrangements over time is called for.

People Who Are Homeless or Live in Institutions

Using household-based surveys excludes homeless people and people living in institutions, such as nursing homes.¹⁵ This implies that these groups are also excluded from the established statistics on the distribution of ICW. One exception is the distributional income statistics from the ACS, which include the institutionalized population. These groups total about 0.6 million people (homeless) and 3.6 million people (institutionalized), respectively, of a total of 331 million people in 2020.

Although relatively small in numbers, people who are homeless, people in prisons, and many people in nursing homes and other institutions are poor whether measured by income or wealth, and their consumption may represent a much lower living standard than most people would want. Moreover, the distribution of income and wealth for the institutionalized population is likely to be different than for the household population (see Wicks-Lim, 2023). It is difficult to measure these groups and their economic situation, but it could be useful at a minimum for new and improved statistics reports to include footnotes, text boxes, and the like to document the extent of the omissions and the likely effect on estimates for the lower end of the distribution.

The main reason to exclude these groups is that their results are not comparable to private households—for example, it does not make sense to add up the income or wealth of all people in a given prison or nursing home and to reflect that as the income or wealth of one household. On the other hand, they may consume specific goods and services jointly (such as accommodation costs and food), which may make it difficult to treat them all as

___________________

¹⁵ Household surveys typically include group homes, but not institutions. The CPS-ASEC and SCF exclude college students living in university- or college-sponsored housing (college students are to be included in the parental home), whereas the CE counts them in their own housing (see Chapter 4).

individual households. For some questions, it is important to include these populations, who are likely to be impoverished. Hence, it is important for reports to provide results for institutional households separately and to note the absence of data for people who are homeless or living in institutions.

Equivalizing Units for Differences in Size and Composition

Because households and families differ in size and composition, it is often desirable to develop statistics that account for these differences by using an equivalence scale to convert households or families to equivalent units in terms of consumption and income (to date, there has been no research on equivalizing units for wealth). The Census Bureau uses the equivalence scale of the SPM (which is used to determine poverty thresholds) to produce “equivalence-adjusted” estimates of income—for example, equating a single-person household with an income of $30,000 to a household consisting of a married couple with two children with an income of $65,000.¹⁶ A simple and commonly used equivalence scale is the square root of family size.

Conclusion 3-4: Because household members typically pool some, or all, of their income and wealth, and consume certain types of goods (mainly housing and food) and services collectively, it is sensible to use the household—defined as all people living in the same housing unit—as the main unit of analysis for new and improved statistics on income, consumption, and wealth. It is also useful to compile results for families, treating cohabitors and their children as family units, and for other groups of people who pool income for basic expenses, and to assess how to make units equivalent by applying appropriate equivalence scales.

For some research, it might be appropriate to use a household measure of resources, or a per-capita or per-adult measure may be more appropriate. BEA and OECD rank households by their equivalized income; however, both report the mean income by decile at the household level so that the sum of deciles equals the aggregate value in NIPA; World Inequality Database and Piketty et al. (2018) use a per-adult measure of income for similar reasons. To use individuals as the unit of analysis requires complete data on

___________________

¹⁶ The SPM uses a 3-parameter equivalence scale to adjust the 2-adult/2-child poverty threshold for different numbers of adults and children and economies of scale:

- One and 2 adults: scale = (adults)0.5

- Single parents: scale = (adults + 0.8 × first child + 0.5 × other children)0.7

- All other families: scale = (adults + 0.5 × children)0.7

each individual’s ICW. In contrast, using the household as the unit of analysis affords the option of obtaining data on ICW for the entire household, on the assumption of equal sharing of resources. Typically, surveys obtain considerable data for individual members (e.g., adult earnings or property holdings), but they also obtain some data for the household as a whole (e.g., housing expenses)—see Chapter 4.

Inflation Adjustment Methods

Statistical agencies vary in the index of inflation they use to produce established household ICW statistical series, and this variation can influence the trend picture that different series convey. For example, using the PCE deflator, household median money income would have increased by 28% in the period 1970–1999, compared to 22% using the Consumer Price Index for All Urban Consumers (CPI-U).¹⁷ BEA uses the PCE deflator for its household PCE and PI series, while the Census Bureau now uses a combination of the C-CPI-U and the CPI-U-RS.¹⁸ Neither BLS nor SOI adjust their time series of household expenditures or median AGI by any inflation measure, which is not helpful to policy makers or the public.

There is considerable debate about the merits of one or another deflator for various purposes (see, e.g., Meyer & Sullivan, 2023, in the context of poverty measurement). The deflator may need to differ depending on the specific ICW definition. For example, using an income measure that includes in-kind benefits, the deflator needs to include the price changes for these in-kind transfers. As shown in Figure 1-1, the choice of inflation measure significantly impacts the trend in income.

The PCE deflator, but not the CPI-U, includes spending by third parties on behalf of consumers, for example, payments of health insurance providers for prescription drugs. The CPI-U has been criticized as a fixed-weight index that is not updated rapidly enough for changes in consumer behavior; the chained CPI-U (C-CPI-U) changes weights each month but is subject to revisions to incorporate additional months of data.

Not expecting these debates to be resolved any time soon, it would be helpful for statistical agencies to collaborate on periodic reports that compare household tables for consumption, expenditures, income, and wealth on key statistics, such as means and medians broken out by decile,

___________________

¹⁷ Calculated from Unrath et al. (2022). The years 1970–1999 were chosen because Table 1 compared the chained CPI-U to the CPI-U for subsequent years.

¹⁸ The Census Bureau is switching to the chained CPI-U index developed by BLS for years beginning in 2000 to inflation-adjust household median income estimates (see Guzman & Kollar, 2023).

stated in nominal terms, and stated in inflation-adjusted terms using both the PCE deflator and the C-CPI-U (given the Census Bureau’s decision to move to that index).

Conclusion 3-5: Statistical agencies’ estimates of household income, consumption, expenditures, and wealth would be more helpful to users if presented in both inflation-adjusted and nominal terms. Agencies can usefully collaborate on periodic reports that compare series adjusted for inflation using the personal consumption expenditure deflator, the chained Consumer Price Index, and the Consumer Price Index Research Series or other inflation series.

In addition, deflators may be different for households at different income levels as discussed in National Academies (2022, Ch. 6) and the Interagency Technical Working Group on inflation measures (see Census Bureau, 2021), and deflators may be different for different geographic areas (as in BEA’s regional price parities).

Timeliness and Frequency

Principle 1 in the Committee on National Statistics volume on Principles and Practices for a Federal Statistical Agency (P&P) (National Academies, 2021, p. 2) highlights the importance of timely statistics, stating: “Federal statistical agencies must provide objective, accurate, and timely information that is relevant to important public policy issues.” This principle suggests that timeliness is a dimension of accuracy whose importance stems from policy and research information needs. For example, a statistic with short time lags between data collection and publication more accurately reflects the current poverty situation than does a statistic that uses outdated information.

Never has the need for timeliness been more apparent for economic statistics than during the market and nonmarket disruptions created by the COVID-19 pandemic. Obtaining timely estimates on the distributions of ICW will provide policy makers with the information needed to examine business cycle changes and even help in targeting transfer programs. Job losses resulted in lost income, consumers changed expenditure patterns in response to public health measures, and incomes changed due to the increased government transfers in the Economic Impact Program and Child Tax Credit expansion. Federal agencies responded by attempting to collect more timely data, such as the Census Bureau’s Household Pulse Survey.¹⁹

___________________

¹⁹ See www.census.gov/programs-surveys/household-pulse-survey.html

Related to timeliness is the frequency of an economic statistic—both in terms of release and reference period covered. Most of the current principal federal economic indicators are released every month with only a few weeks’ lag between data collection and release. Current income and poverty measures, which pertain to the preceding calendar year, are released annually, about 6 months after data collection ends. Recently, attempts to create a monthly poverty measure in close to real time have been undertaken. Parolin et al. (2022) developed a framework for producing monthly estimates of the SPM and OPM. Han et al. (2020) used high-frequency (monthly) CPS data to produce short-lagged, previous 12-month estimates of income and poverty (the monthly CPS obtains estimates of total family money income for the preceding 12 months). Finally, Realtime Inequality has produced monthly estimates of the distribution of national income.²⁰

At the panel’s open meetings and planning meeting, presenters stressed the importance of having distributional estimates on a “real-time” basis. Even OECD (see Levy, 2023) is looking into “nowcasting”—using modeling and forecasting techniques to produce a more timely estimate (e.g., as is done for FRB’s Distributional Financial Accounts). There is often a tradeoff between timeliness and accuracy, however, as recognized in Principles and Practices for a Federal Statistical Agency (National Academies, 2021, p. 96):

The panel acknowledges that there are downsides to releasing preliminary estimates for such statistics as annual poverty rates and the distribution of income-based on modeling techniques. Revisions once full data are available could undercut the credibility of the estimates. The hope of the panel is that blended data from surveys, administrative records, and other sources could be used to improve the timeliness of estimates in a 21st century data infrastructure: “Combining diverse data sources also provides the opportunity to produce timelier, more granular, and higher-frequency statistics, as needed” (National Academies, 2023, p. 95). As with the DFA, the modeling provides useful timely data on the distribution of wealth, and the benchmarking process confirms the results and leads to improved modeling (or nowcasting) estimates.

___________________

²⁰ See realtimeinequality.org

DETAILED ESTIMATES FOR COMPONENTS, GROUPS, AND GEOGRAPHY

The OECD (2021) guidance note for developing the new SNA for release in 2025 suggests a few criteria for selecting the categories, estimates, and specific groups for producing household distributions of ICW. These criteria include (OECD, 2024, Section 3.2, p. 13) selections that provide the “[…] most insight in differences in consumption, income, and wealth patterns between groups; [are of] most interest for economic analysis and government policy purposes; enable users to easily identify themselves with one of the groups; and meet specific user demands.”

As discussed in Chapter 4, the quality of the results plays an important role in selecting the breakdowns and targeted levels of detail. Breakdowns into household groups have to rely on information at the individual and household levels as available from the microdata sources. Specific household groups may be of much interest to users, but if their results have to rely on a very limited number of observations or on a large number of assumptions, it may not be opportune to target or publish statistics for them at this level. In this regard, the list presented in this subsection provides an overview of possible breakdowns.

Component Detail

Overall levels of household consumption, expenditures, income, and wealth are, of course, the headline measures in established statistical series. Components of each are also important for policy and public understanding. For example, it is important to know the percentages of household income that accrue from wages, from self-employment income, from property income, and from private and public benefits and assistance. Likewise, it is important to know the wealth composition of specific household groups to assess how they may cope with a temporary income loss. On the consumption side, more detailed information on specific consumption categories may provide insight into how different groups may be affected by increasing inflation.

Given the importance of the estimates for health insurance benefits, capital gains, retirement, housing, and pension wealth, it is important for statistics to show the various impacts of detailed components. For example, the Census Bureau shows the impacts on poverty for the various resource components of the SPM (e.g., see Creamer et al., 2022, Figure 11). As with every dimension examined in this chapter, the statistical agencies vary in the detail they provide for components.

More detailed information on components may also be helpful to rank households, for example, according to their main source of income. A common breakdown in this regard is into “wages and salaries,” “income from self-employment,” “net property income,” and “net current transfers received.” Each household is classified under the category that shows the highest contribution to its income. When applying this classification, it is preferable to look at income cleaned from cyclical effects, as these may otherwise lead to undesirable temporary reclassifications, such as when a self-employed adult suffers a temporary loss.

With regard to detailed categories, for households and families the Census Bureau only provides total money income and earnings and no other component. In contrast, for people ages 15 and older the Census Bureau provides the number of such people in various income categories for more than 25 components, including not only wages and salary but also veterans’ benefits, dividends, alimony, and others. Moreover, it cross-tabulates these details by sex, race and ethnicity, and broad age group. While this is useful for such purposes as calculating ratios of men’s to women’s earnings at different ages and for different race and ethnicity groups, what may be needed even more is a breakdown of income components for households and families so that one can understand the sources of income supporting their consumption and saving.

BLS provides detailed breakdowns for consumer units of mean annual expenditures and expenditure shares by income category for more than 80 expenditure categories, and it also provides seven components of income and four components of taxes. BEA provides decile shares for 15 categories of consumer unit PCE and eight categories of household PI. For example, the top decile receives only 8% of government social benefits but 83% of self-employment income and 71% of dividend income. SOI/AGI provides tax filing unit information for virtually every line on the 1040 form, including income types, deductions, and taxes owed.

Finally, FRB provides aggregate wealth by economic level (e.g., top 0.1%, bottom 50%—see next section) for seven asset components (real estate, consumer durable goods, corporate equities and mutual fund shares, defined pension benefit entitlements, defined contribution pension entitlements, private businesses, and other assets) and three liability components (home mortgages, consumer credit, other liabilities). Given public and policy concerns about medical and student loan debt, it could be useful to see those liability components separately, as is done in tables from the SIPP,²¹ which distinguish secured from unsecured debt. It could also be useful to classify assets into more and less liquid assets and within each category

___________________

²¹ See www.census.gov/data/tables/2021/demo/wealth/wealth-asset-ownership.html tables for wealth and asset ownership and for debt.

into financial and nonfinancial assets. Examples of relatively liquid financial assets are bank accounts, certificate of deposits, stocks and bonds; of relatively illiquid financial assets are pension entitlements; and of relatively illiquid nonfinancial assets are business and nonresidential property equity. Separately showing primary residence equity is important given that owned homes (along with vehicles and other consumer durables) are the major asset of most households. The importance of distinguishing liquid and illiquid wealth is that some population groups (e.g., the elderly) are more likely to have accrued and to need access to their illiquid wealth to support their consumption (see Volz, 2024).

Classification by Economic Levels and Distributional Measures

A key reason for producing statistics on household (or other unit) ICW is to understand the shape of the distribution—for example, the share of aggregate wealth for households grouped into lower and higher levels of wealth. This can usefully be accomplished by arranging households in order of their total ICW and then slicing them into decile groups (tenths). Households can also be sorted into quintile groups, recognizing that a quintile breakdown may conceal large inequalities within the quintile groups. This may be particularly relevant for the top income and wealth quintile groups, calling for more granular breakdowns for these groups if possible. As a minimum, one should target income decile groups, a median, and, if possible, results for the top 5%, top 1%, and even top 0.1%, as well as for the bottom 5%.²² Estimates for decile groups with additional detail at both the top and bottom of the distribution will shed light, not only on outsize gains at the very top but also stagnation at the bottom (e.g., in earnings). With regard to very granular breakdowns (e.g., percentile groups), as a general rule it should be assessed whether the quality of the results can be assured at such levels, also bearing in mind the complexities in deriving consistent results across income, consumption, and wealth.

Figure 3-1 shows household decile groups (and the top 90–95%, 95–99%, and 1%) by their percentage share of PI.²³ The figure shows the similar distributions of household income and disposable (after-tax) PI; lower tenths have a larger share of disposable income, while the share of disposable income is smaller for the top tenths. In particular, Figure 3-1 shows the importance of examining the top of the distribution.

___________________

²² See also the G20 Data Gaps Initiative, www.imf.org/en/News/Seminars/Conferences/g20-data-gaps-initiative

²³ See tables from BEA Distribution of Personal Income, www.bea.gov/data/special-topics/distribution-of-personal-income

The Census Bureau currently slices households into quintile groups (fifths), not decile groups (tenths), for measures of mean money income and percentage shares of aggregate money income, which obscures the key impacts of the top of the distribution on the relationships between pre-tax and after-tax income. The Census Bureau provides estimates by decile group of the top dollar amount, plus estimates for the top 5% of households of means, shares, and top dollar amounts. Given the results in Figure 3-1, producing estimates for the top 1% would be useful in assessing the differences in the distributions.

BLS publishes mean expenditures for decile groups of CU income, but not for decile groups of expenditures. It has produced CU estimates of PCE for deciles and the top 1% of CUs. FRB publishes Distributional Financial Accounts estimates of wealth for households below the 50th percentile, in the 50th–90th percentile, the 90th–99th percentile, the 99.0th–99.9th percentile, and the top 0.1 percentile. These cutoffs make sense because wealth is so concentrated among the very wealthy, but for comparative purposes it would be helpful to have wealth estimates by decile. FRB also publishes wealth estimates by income quintile groups and the richest 1%.

BEA, FRB, SOI/IRS, and BLS (for its CE-based estimates of PCE) all publish estimates for the top 1% of households. The Census Bureau and BLS (for its expenditure estimates) do not provide detail within the top 1%, even though, given the pronounced rise in inequality in recent decades, such estimates are crucially important for policy makers and public understanding. Indeed, estimates for the top 0.1% (which FRB and SOI/IRS do publish) would be useful.²⁴ Because most surveys (except for the SCF, which includes a high-income sample from tax records) do not adequately cover the upper tail of households, agencies may need to adjust measures or use models to produce estimates for the top 1%. Measures of uncertainty need to accompany such estimates to convey that they typically rest on strong assumptions and employ different models (see Chapter 4). In that regard, BEA uses tax data to adjust the top of the distribution, BLS uses a Pareto adjustment to the top of the expenditure distribution for its CE-based PCE distributional estimates, and CBO statistically matches CPS and tax data (see Chapters 4 and 5).

Distributional estimates may be presented in various ways (see also Box 1-1). For example, they can be produced in the form of absolute monetary values showing the totals for each of the household groups. They can also be produced based on results per household, per consumption unit, or per capita. From the per consumption unit (or per household or per capita) estimates, indicators can be derived to show the presence and degree of inequality between household groups. The following indicators can be used for that purpose:²⁵

• The ratio of the highest to lowest, which shows the value of equivalized ICW for the household group with the highest value to that of the group with the lowest value. This ratio is often used to make cross-country comparisons and to monitor changes over time within a country.
• The coefficient of variation, which shows the variation from the average. For a given classification of households the coefficient of variation is the ratio of the standard deviation to the mean. It is

___________________

²⁴ The World Inequality Database (WID.world, see Chapter 4), publishes results for the 0.1% and 0.01% levels. The SNA guidance recommends that statistical agencies try to publish at the top 1% level. Finally, the G20 Data Gaps Initiative recommends that G20 economies (including the United States) at least publish at the decile (income and wealth) level by the end of 2026.

²⁵ For some of these indicators, a problem may arise in case of negative values for specific households or household groups. This may for example be the case when some households report negative income or wealth. This may need to be corrected for in some way to arrive at sensible results.

• less relevant when estimates are broken down into relatively large groups of households (as these conceal a lot of inequality within the group), but becomes much more relevant when focusing on more granular breakdowns into relatively homogeneous groups.
• The Gini index and other summary measures of inequality (see Chapter 1).

Furthermore, specific indicators may be derived based on the estimates per household group, such as these:

• Share of each household group of total ICW;
• Composition of ICW for each household group;
• Specific financial indicators, such as the debt to income ratio, debt to financial assets ratio, and debt to residential assets ratio per household group; and
• Impact of redistribution measures by government for each household group.

Estimates may also be used to analyze trends over time. However, these trends may be influenced by dynamics between household groups over time, that is, by households moving from one household group to another.

BEA, BLS, and the Census Bureau provide measures that summarize the inequality in a distribution. Thus, Gini indexes and ratios of the 90th to the 10th percentiles are available for household PI, PCE, and money income.²⁶ In addition, BEA and BLS produce Theil indexes, coefficients of variation, and the ratio of the 80th to the 20th percentiles for household PI and PCE. The Census Bureau additionally produces ratios of the 90th to the 50th and the 50th to the 10th percentiles for household money income. No summary inequality measures are produced for household expenditures (BLS), wealth (FRB), or AGI (SOI/IRS).

It is also important to have consistent measures of the distributions. For example, the Census Bureau uses arbitrary income cutoffs, from households under $15,000 to those with $200,000 or more money income. Tables for families start with under $5,000 in money income up to $200,000 or more, and tables for people ages 15 and older start with under $2,500 money income up to $250,000 or more. SOI/IRS uses arbitrary income cutoffs as well, but they are finer-grained than the money income categories, ranging from under $5,000 to $10 million or more in AGI.

___________________

²⁶ Whether the Gini index is appropriate for household money income, which is based on a sample that underrepresents very-high-income households, is a question. For a view that it is appropriate, see Cubillos (2021). See also discussion in Chapter 1.

Finally, for maximum utility, it is crucial to provide distributional estimates of household ICW by geographic area to the extent feasible and by a variety of socio-demographic characteristics. For each such characteristic, shares and conditional estimates are useful—that is, publishing estimates of the share of households in each category (e.g., owner or renter) and then the conditional distribution of, say, income for households in each category. With these estimates, users can calculate unconditional estimates of interest to them (e.g., median interest income for all households).

Conclusion 3-6: To summarize levels and trends in household economic wellbeing, agencies that publish improved estimates of household income, consumption, expenditures, and wealth need to display them with categories that illuminate the entire distribution. At a minimum, it would be useful to provide levels for decile groups (tenths), as well as for the top 5% and 1% and the bottom 5% of households. The relevant statistical agencies could also collaborate on publishing a comparable set of inequality measures along with distributional breakdowns by socio-demographic groups.

The next subsection discusses the benefits and challenges of providing geographic detail, including for states, metropolitan areas, and smaller areas. The subsection following that discusses the importance of providing an array of socio-demographic detail (e.g., age, gender, race/ethnicity, family type, owner/renter, and other characteristics), while the final subsection considers issues in measuring race and ethnicity and phenotype.

Geographic Detail

A growing body of research across the social sciences has shown substantial geographic variation in numerous outcomes that are likely tied to wealth and wealth accumulation, such as economic mobility, educational attainment, and exposure to crime (Chetty et al., 2014). Despite this, there are currently no publicly available data with which to measure small-area levels and trends in household consumption, expenditures, or wealth. Scholars interested in understanding the spatial distribution of affluence tend to rely on measures such as income and self-reported home values, which are poor proxies for wealth—especially among the wealthiest households. Considering the fact that wealth inequality in the United States is more severe than income inequality, recent scholarship documenting increases in income segregation in recent decades (Reardon & Bischoff, 2016) may suggest a corresponding—and perhaps more dramatic—increase in the segregation

of wealth. Small-area estimates of wealth would provide scholars with much-needed insight into a wide range of related, spatially disparate phenomena, such as racial disparities in the returns to homeownership (Howell & Korver-Glenn, 2020), the wealth effects of local neighborhood-school structures (Rich & Owens, 2023), and the intra/inter-generational impacts of geographic wealth concentration (Sharkey & Faber, 2014). To assess differences across geographic areas, it would be useful to use geographic-specific price adjustments (as in the BEA regional price parities).

The surveys that support established statistics do not have sufficient sample size to support fine geographic detail. BLS publishes CU expenditures by region, metropolitan statistical areas for the nation’s highest level (four) regions, nine census regional divisions, and rural areas and urban areas by population size categories, as well as for five states (CA, FL, NJ, NY, TX). The Census Bureau publishes median money income for households by region and metropolitan status, that is, inside principal cities, outside principal cities, and outside metropolitan statistical areas. Even at these levels of aggregation, there is substantial variation—for example, in 2021 household median money income in the South was only $63,000 compared with $79,000 in the West, and only $54,000 in all non-metropolitan areas compared with $80,000 in metropolitan areas outside principal cities (Semega & Kollar, 2022, Table A-1). The Census Bureau also publishes the number of households and families by money income category by race and ethnicity by a number of characteristics, including metropolitan status, region, and division (the latter for households only).

FRB does not provide geographic breakdowns for its household wealth estimates, and it would be difficult to do so given the small sample size of the SCF, which provides the basis for distributing aggregate wealth data to households. Using 3 years of CPS-ASEC data, BEA produces state-level estimates for its household distributions of PI. Estimates by urban-rural and metropolitan status would also be useful, like those produced by BLS and the Census Bureau, given that aggregate PCE and PI are allocated to households in the CE and CPS-ASEC, respectively. SOI/IRS, on the other hand, publishes numerators and denominators to calculate per-return values of AGI and virtually every item on the 1040 form for all states and zip codes.²⁷

Ideally, there would be estimates of household consumption, expenditures, income, and wealth for census tracts or zip codes. However, even state and substate estimates would be of great value—for example, for metropolitan and micropolitan areas and remainders of states and/or for counties. Neither the CPS-ASEC nor the CE can reliably support such estimates, but

___________________

²⁷ See view.officeapps.live.com/op/view.aspx?src=https%3A%2F%2Fwww.irs.gov%2Fpub%2Firs-soi%2F20zpdoc.docx&wdOrigin=BROWSELINK

it could be possible to use modeling with the ACS to produce special reports at periodic intervals (see Chapter 4).

Recommendation 3-3: To provide additional geographic detail for new and improved estimates of household income, consumption, and wealth, relevant statistical agencies should conduct research and development and move to implementation, as appropriate, as follows:

• The Census Bureau should conduct research to produce subnational estimates (e.g., states, counties, cities) for the recommended household income definitions (e.g., using the American Community Survey in conjunction with the Current Population Survey and National Experimental Wellbeing Statistics data).
• The Bureau of Economic Analysis should expand its research producing state-level distributions for personal income and extend these estimates to metropolitan areas and counties for the recommended household income definitions.
• Other agencies should conduct research on subnational estimates for consumption and wealth (e.g., using alternative data or modeling techniques).

Demographic Detail

Having socio-demographic information on various household groups helps in identifying and monitoring policy-relevant target groups and, therefore, facilitates effective policy interventions. The United Nations Economic Commission for Europe guide to data disaggregation for poverty measurement (United Nations Economic Commission for Europe, 2017) recommends data disaggregation by specific socio-demographic variables, which would allow for a more complete analysis on the distributional side. As noted earlier, OECD guidance suggests that groupings of people should be targeted that (a) provide most insight in differences in consumption, income, and wealth patterns between groups; (b) are of most interest for economic analysis and government policy purposes; (c) enable users to easily identify themselves with one of the groups; and (d) meet specific user demands.

Some of these breakdowns may focus on specific socio-demographic characteristics of households, whereas others may focus on characteristics of the individuals belonging to the households in the various household groups. Which types of socio-demographic information can be published will depend on the information available from microdata sources. This section provides a description of the additional socio-demographic information that may be published together with the distributional estimates. Some of these breakdowns have already been discussed in the previous section, as they may also be used to break down the household sector into more granular household groups.

For comparison purposes, it is important to obtain consistent demographic characteristics both across surveys and internationally. From the UNECE Guide on Poverty Measurement (UNECE, 2017) and the Canberra Group Handbook on Household Income Statistics (Canberra Group, 2011), one can assemble a list of commonly used socio-demographic breakdowns that focus on individuals. These breakdowns, which focus on the reference person for household estimates, include sex, age group, education level, employment status, race/ethnicity, and household type. Other candidate breakdowns include housing tenure (owner/renter), household or family size, and disability status of the reference person. Yet another useful distinction could be estimates for households with owned businesses vs. other households and households that rent property vs. other households. For analytic purposes, combinations of two or more of these variables (especially age and sex) are often helpful to identify drivers and specific vulnerabilities across the life cycle.

Levels and trends in household consumption, expenditures, income, and wealth vary substantially by such characteristics as race and ethnicity, educational attainment, and family type. For example, race breakdowns from BLS reports show that, in 2021, (a) mean after-tax income and expenditures of Black CUs were only 71% and 74%, respectively, of the corresponding means for all other CUs; (b) only 43% of Black CUs owned their home compared with 64% of all other CUs; and (c) Black CUs (also Hispanic CUs) spent 38% of their (smaller) budgets on housing, compared with 33% for other race and ethnicity groups.

Distributional measures (discussed above) would be more useful if produced for various demographic groups, for example the percentage of Black people in the poorest fifth, as is done in OECD tables and the EUROSTAT/OECD report. There may also be research and policy questions that would require the distribution for one demographic characteristic, such as the top fifth of people ages 65 and older compared to the top fifth of people ages 25–34.

Conclusion 3-7: Agencies that publish improved estimates of household income, consumption, expenditures, and wealth need to display them for socio-demographic groups that are of policy interest and that provide insights into differences in levels and trends. Groups of interest include those defined by such characteristics (for the household reference person or individuals in households) as age, gender, race/ethnicity, education level, disability status, employment status, family type and size, and housing tenure (own/rent). The relevant statistical agencies could collaborate on publishing comparable categories for these and other characteristics of interest.

As with the other attributes of established statistical series discussed above, statistical agencies vary in the characteristics for which they produce estimates—see Table 3-3. There are also very few cross-tabulations of characteristics or of characteristics by geography, because the underlying datasets lack sufficient sample size (the exception is that the Census Bureau provides race, sex, and broad age group cross-tabulations for money income of people ages 15 and older). It would be highly useful for statistical agencies to collaborate to identify common attributes that they can all publish and to investigate ways to develop cross-tabulations.

Collaboration requires addressing such issues as appropriate measures of consumption, expenditures, income, and wealth to display for demographic groups. For example, DFA estimates of wealth by race and ethnicity provide only totals for various assets and liabilities. White households own the overwhelming majority of all asset types, but White households are also the majority of households; the same pattern applies to household debt. Without denominators for computing per-household assets, it is not possible to determine wealth disparities directly from the DFA published tables.

Race and Ethnicity

Race and ethnicity categories in federal surveys and publications are governed by Statistical Policy Directive No. 15, first promulgated in 1977 and updated in 1997. The current race categories are White, Black, American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander, and Some Other. Publications often include a “Two or More Races” category because people may check off more than one race category. The current ethnicity categories are Hispanic and not Hispanic.

Understanding disparities by race and ethnic categories depends on the ability to examine ICW by detailed categories. These disparities have been increasing over time (e.g., for wealth disparities; see Addo & Darity, 2021; Sabelhaus & Thompson, 2022). Figure 3-2A shows that the disparities between White people and Black and Hispanic people increase over the life cycle, while Figure 3-2B highlights the disparities increasing over time.

There is no doubt that the United States is becoming more diverse, which increases the need for race and ethnicity data on household ICW. In that regard, the 2020 census saw a modest decrease in the White Alone category among non-Hispanic people and a marked decrease among Hispanic people. Conversely, the Two or More Races category increased somewhat among non-Hispanic people and substantially among Hispanic people. A similar phenomenon was evident in the 2020 ACS compared with 2019. These changes were due to some combination of demographic change (a greater proportion of children who are not White), changes in how people chose to self-identify, and changes in how data on race and ethnicity were

TABLE 3-3 Demographic Characteristics Published by Statistical Agencies for Estimates of Household Expenditures, Income, and Wealth (characteristics are for the householder/reference person)

NOTE: AGI = adjusted gross income; BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; CU = consumer unit; DFA = Distributional Financial Accounts; FRB = Board of Governors of the Federal Reserve Board; HI = household income; PCE = personal consumption expenditures; PI = personal income; SOI/IRS = Statistics of Income Division of the Internal Revenue Service.

^a BEA tables do not include socio-demographic tables, although some are provided in working papers (see, e.g., Gindelsky, 2022).

SOURCES: BLS (2023a); Board of Governors of the Federal Reserve System (2024); Internal Revenue System (n.d.); Semega and Kollar (2022, Tables A-1 and A-2).

captured by the Census Bureau beginning in 2020, although the extent to which each factor contributed to a picture of a more diverse society is not known.²⁸

At present, under the leadership of the chief statistician, the OMB released a revised version of Statistical Policy Directive No. 15 in 2024. The two big changes proposed are to combine the race and ethnicity questions into a single question with Hispanic as a category and to add a Middle Eastern and North African category. Statistical agencies will need to collaborate on needed changes for collecting race and ethnicity information in relevant surveys and on ways to publish the new combined categorization in reports of household ICW (see Box 3-3 on issues with measuring race).

___________________

²⁸ The Census Bureau for the 2020 census and the ACS beginning in 2020 introduced new write-in spaces for people who checked White and/or Black, captured more write-in characters for coding than previously, and introduced a more expansive coding list. For example, people who checked White and wrote in “Hispanic origin” were classified as White and Some Other Race or Two or More Races, instead of White as in previous censuses. See National Academies (2023, Ch. 10).

Conclusion 3-8: Expanding on the recent efforts by the U.S. Office of Management and Budget on Revisions to the Standards for the Classification of Federal Data on Race and Ethnicity (Statistical Policy Directive No. 15) and the efforts of the Census Bureau to capture more information on people’s race and ethnic identities, critical analysis of these delineations of race and ethnicity, the definitions of the categories, and the data sources that could be used to measure race and ethnicity would have great value.

SUMMARY

Established statistics on household consumption, expenditures, income, and wealth play an important role in framing public debates on many issues of economic wellbeing in the United States, overall, over time, and for geographic areas and population groups. BEA, BLS, the Census Bureau, the FRB, and SOI/IRS publish many useful series that policy makers and the public look to for understanding. In recent years, BEA, BLS, and FRB have undertaken programs specifically to develop distributional statistics for households, families, and CUs, and the Census Bureau has invested in a program to improve the quality of its household income statistics (the NEWS program). The panel applauds these efforts. The panel also calls on the agencies to develop new statistics on the joint distribution of household ICW.

A drawback of currently available household consumption, expenditures, income, and wealth statistics is that they differ in so many ways—in definition, unit of analysis, inflation adjustment method, and detail provided for components, economic levels, inequality measures, geographic areas, and demographic groups.

The panel calls on the agencies to work collaboratively to develop comparable definitions, measures, and other features of household estimates of ICW to the extent feasible—for example, sample size in the underlying data combined with privacy considerations may limit the detail that can be provided. In that case, it would be useful to determine categories, such as race and ethnicity or educational attainment, for which additional detail that some agencies can provide will add up to the more limited categories that other agencies can provide.