2

Definitions and Conceptual Issues

This report investigates data needs for the purposes of tracking and assessing the economic circumstances and behavior of American individuals and families.¹ More specifically, it provides guidance for the development of an improved national data system to measure the extent to which prosperity is shared across the U.S. population and to promote a better understanding of how government policy and economic events affect the distribution of resources. The study’s charge directs the panel to analyze and compare data sources and statistical estimates for income, consumption, and wealth for individuals and families, and then to consider how such data can be integrated to inform conclusions about their wellbeing. As suggested in the previous chapter, the current data infrastructure falls short in measuring the impact of economic events on families. Examples include the effect of recent recessions on household wealth, the effect of shifts to alternative work arrangements on workers’ income, and the effect of the COVID-19 pandemic on household’s income and consumption. The current system fails to record these impacts in a timely and coordinated manner.

The recommendations in this report aim to spur development of an internally consistent framework for collecting and using data on the income, consumption, saving, and wealth of families, a framework that can serve multiple analytical purposes and provide opportunities for regional

___________________

¹ For ease of exposition, in this chapter, the term “family” is used as shorthand in labeling the reference unit; in practice, however, data may be collected at the household, family, or individual level. When discussing specific data sources and income, consumption, and wealth (ICW) measures, more precision will be used in the labeling of the reference unit.

and international comparisons. To that end, this chapter discusses the following: (a) the measurement unit (individuals, families, households, and others), including some basic definitions; (b) the relationship between measures of income, consumption, and wealth (ICW) and the broader concept of economic wellbeing; (c) the linkages among these measures through a fundamental economic relationship; (d) the definitional complexities in the production of ICW data and statistics (which are dealt with in greater detail in subsequent chapters), as well as key sources of microdata on families; and (e), in the chapter’s annex, more details on a number of thorny measurement areas—such as the treatment of health insurance and retirement benefits in estimates of income and wealth—that arise when conceptualizing and generating ICW statistics.

BASIC NOTIONS

Household and Other Units

In developing a system of data collection and dissemination capable of tracking the distribution of economic wellbeing across the U.S. population, it is useful to first clarify what the basic micro units are that underpin such a system. Variables of interest may be associated with individuals, with families, or with households. For example, data on wealth and consumption are typically collected at the household level. But if individuals belong to different household units over time, as seen in panel data, individual wealth may need to be followed as well.

A key part of the conceptual strategy involves being able to work across these groupings in their hierarchical way. In some cases, for example, it is possible to sum individual wages to calculate total family or household wages; in others, it is possible to go in the other direction and decompose resources—for example, by assigning equal ownership of jointly held assets (such as homes or stock accounts) to the individuals in a married couple. Crucially, the distinction between individual and household variables matters for distribution analyses. As pointed out by Heathcote et al. (2023), income pooling and resource sharing within the family as well as redistribution by the government have significant impacts on the level and trend of household inequality. Analyses based on family and household units—which often pool resources in a way that can smooth shortfalls—typically show lower levels of inequality compared to individual-level analyses.

Several important surveys and administrative sources in the U.S. statistical system generate relevant measures from the household sector of the economy; these are reviewed in the last section of this chapter. The Census Bureau’s American Community Survey (ACS) is a prominent example of a survey, and income tax data compiled by the Internal Revenue Service in

Statistics of Income Program is a leading example of an administrative data source. For many data sources, the coverage objective (or target population) is all households residing in the country within a specific reference period. As is discussed in detail in Chapter 4, survey and administrative data each have comparative advantages in terms of coverage and detail.

Basic Definitions

Analyses of households’ economic wellbeing generally draw on information about one or more of the following categories: consumption, income, saving, and wealth. The first three of these measures are flows measured over a period of time. This period is often 1 year but may also be longer or shorter depending on the type of analysis and temporal availability of reliable data. The fourth of these measures, wealth, is a stock, measured at a specific point in time—often designated as the beginning of the year and the end of the year over which flows are measured.

These four dimensions of economic wellbeing may be defined, in a simplified manner, as follows:

1. Consumption: Expenditures and service flows from the use of goods that contribute to wellbeing during the year (or other period)—purchases of food, shelter, transportation, entertainment, and the like. It is important to point out that a family’s consumption can differ from its expenditures. For example, a family may buy a car every few years. Its consumption of the services of cars occurs every year, while its expenditure on cars is concentrated in occasional years.
2. Income: Funds available for personal use, received as employment earnings and benefits, from income-generating assets, including retirement accounts, from Social Security, and from returns on capital.
3. Saving: Flow of funds accumulated or decumulated during the year because consumption differs from income.
4. Wealth: The stock of funds at the beginning of the year in bank accounts and other holdings of financial accounts or assets, plus real estate, business wealth, and other stores of value (e.g., pension wealth), net of all debt.

These are basic definitions and hence omit important details. For example, interhousehold transfers need to be accounted for on both the donor’s and the recipient’s economic ledger.² The last section of this chapter includes more details on the range of specifications and definitions used by statistical

___________________

² See example 5 on page 67, for instance.

agencies such as the Census Bureau and the Bureau of Economic Analysis (BEA). However, at this point it is important to add an important distinction: wealth falls into two broad categories, human wealth and nonhuman wealth. Human wealth, which reflects people’s skills, health, and abilities, is the capitalized value of future wages and earnings. Although human wealth plays an important role in some economic research questions, it is not a measure found in any of the data sources central to the ICW framework as conceptualized in this report.³ The panel follows the standard approach of considering only capitalized values for assets that are traded in markets, such as securities and real estate. Thus, the focus in this report is on nonhuman wealth, which may also be called net worth—in other words, financial wealth, like bank accounts and securities, plus the value of houses and private businesses, minus debt, such as credit card balances and mortgages.⁴

THE RELATIONSHIPS OF ICW TO ECONOMIC WELLBEING

At the outset, it is important to acknowledge that “economic wellbeing,” as specified in the previous section, provides only a partial view of what constitutes overall wellbeing. In an effort to begin measuring broader constructs, economists and other social scientists have extended the determinants of wellbeing to include health, social connectedness, the environment, and other dimensions. The focus of this report is on economic wellbeing rather than overall wellbeing or happiness, some aspects of which may only be weakly related to economic determinants.⁵ Also, beyond the short discussion below, the report does not include a comprehensive analysis of the role of work effort and other uses of time affecting wellbeing. It

___________________

³ The 2021 System of National Accounts (SNA) Guidance Note (unstats.un.org/unsd/nationalaccount/RAdocs/ENDORSED_WS4_Labour_Human_Capital_Education.pdf) on Labour, Human Capital, and Education does address the issues, however, proposing to only include human capital estimates as experimental (supplementary to SNA data), given the extensive discussions still ongoing regarding its proper measurement.

⁴ Another term for all or part of nonhuman wealth is savings, that is, saving with an “s”. This term is avoided because it is confusingly close to saving (the flow measure defined in the text).

⁵ In the psychology and economics literature, “happiness” is sometimes shorthand for “subjective wellbeing,” which may be measured through people’s self evaluations of several related dimensions—life satisfaction (overall satisfaction with life), eudaimonia (feeling that things done in life are worthwhile), and affect (which has to do with day-to-day feelings such as happiness or anxiety). Economic wellbeing has to do with material standard of living (financial), whereas overall wellbeing is defined more broadly to include health and social contextual elements. It has been observed that that these dimensions of subjective wellbeing do not always track together; for example, evaluations based on Gallup-Healthways Well-Being data suggest that people’s average life satisfaction does not seem to depend on their income levels above a certain point (Kahneman & Deaton, 2010).

recognizes the benefit that a family receives in the form of wages but not the burden of the time spent earning those wages.

Users of ICW data are often motivated by an interest in understanding and comparing the experiences of families through a unified measure of economic wellbeing, as opposed to overall wellbeing. It is important to note that while the four variables defined above—consumption, income, saving, and wealth—play important roles in the measurement of economic wellbeing, none on its own provides a complete or ideal picture. Economic wellbeing is directly related to income and wealth because money buys goods and services, and families are assumed to feel better off if their consumption flow of these goods and services increases.

As noted above, one limitation of even comprehensive ICW data is that they are not sensitive to the volume of work effort a family or individual puts forth. For example, one may assign the same level of wellbeing to a family with two hard-working members and a given level of consumption as one assigns to a family with enough wealth to sustain the same level of consumption without the burden of work by any family member. In principle, and subject to some caveats, a comprehensive measure of economic wellbeing would increase with consumption and decline with work effort. Such a measure is conceptually based on a utility function that specifies the disutility associated with labor or, perhaps more usefully, the utility associated with leisure.⁶ Research has recently advanced in the area of measuring the economic burden of work effort based on techniques of marketing science, and government agencies are increasingly showing interest in its inclusion in national accounts statistics.⁷

Underlying the tradeoff between labor and leisure as it relates to wellbeing is something very fundamental—the value of time. Time is such an important part of consumption that failure to consider it can mask some of the inequality between those families where, for example, all parents work and those in which one of the parents is not in the labor force or working limited hours. While time accounting is beyond the scope of the ICW data base envisioned in this report, it is an essential dimension of wellbeing; and, as detailed in Box 2-1, one to which statistical agencies are rightly turning their attention. Future work by investigators interested in measuring wellbeing will no doubt continue to seek ways of blending ICW data with

___________________

⁶ As pointed out by Steinsson (2023), the relationship between work and leisure, and real-life decisions about tradeoffs made by individuals and households, are complex: “How much people work over their lifetime depends both on how much they want to work and on whether they are able to find the amount and type of work that they would like.”

⁷ See, for example, Jones and Klenow (2016), Kahneman et al. (2004), Mas and Pallais (2017), National Research Council (2005), and ongoing work in this area as part of the update of the SNA.

available data on work effort and the value of time, applying the guidance of the developments cited above along with other new research.

Income

Income is the most commonly used proxy for economic wellbeing, in part because it is the most typical component of ICW for which data are collected at the household (or individual) level. Income scores somewhat better than nonhuman wealth in approximating current period economic wellbeing. This is in part because, for lower- and middle-income families, current-period consumption tracks more closely with income than it does with wealth or change in wealth. For younger families and those with modest access to financial resources, defining disposable income is straightforward—it is after-tax earnings and transfers. Even for those who essentially live off paychecks, however, there are tricky issues having to do with workers’ accrual of Social Security benefits or funds in retirement plans. Are these accruals to be counted as income or wealth or, as in the SNA, both?

And what about the financial benefits of home ownership? In contrast to their younger counterparts, older families are more likely to own their homes and therefore benefit from implicit rental income—they could rent out their homes to generate income or live in the house to avoid having to pay rent to acquire shelter services. As discussed in the next section (Definitional Diversity) and itemized in Table 2-1A, imputed rent is included in estimates of income in the National Income and Product Accounts (NIPA) and SNA, among other sources, while it is left out in the Census Bureau money income concept, as well as in the Congressional Budget Office (CBO)’s “after tax and transfer” and the Internal Revenue Service (IRS)’s “adjusted gross income” constructs. While it is more likely to factor into wealth estimates, older families are more likely to have equity in their homes. Still, it is a legitimate question whether to consider the flow value of that equity as part of income. For some households, current-period income may portray an incomplete picture, particularly if their members are able to draw on previously earned income or nonwage sources to purchase goods and services.

Another potential complication arises as some financial flows, such as government-provided health insurance, may not be valued by individual recipients at face value, meaning that, offered the equivalent amount in cash, they would choose to spend the money on different consumption items. In-kind and cash interfamily transfers also raise the question of whose consumption or income is being measured, as mentioned above. And defining income for families whose historical earners are now retired involves a long list of tricky questions regarding the definition of income. And, to answer some questions, home production enters the discussion of what may

constitute income (and consumption). As summarized later in this chapter, in many cases the literature provides good answers to these questions.

Nonhuman Wealth

A family’s nonhuman wealth, which reflects its accumulation of saving⁸ (and return on that saving) over time, can be an informative measure of economic status and stability, in that it provides an indicator of the family’s long-term consumption prospects. Indeed, especially at the high end of the joint ICW distribution, families’ incomes often pale in comparison to their wealth holdings. However, for an important subset of the population, nonhuman wealth is a poor measure of economic wellbeing. Many lower-income families have essentially zero wealth but are not destitute, because of the receipt of earnings and, in some cases, government benefits. Similarly, younger families may not yet have accumulated significant levels of wealth but may be experiencing income flows that are adequate to meet month-to-month consumption needs. Conversely, some households with little or no income finance their consumption with their accumulated wealth or with borrowed funds. When looking at the joint distribution of economic wellbeing, it is not immediately obvious how to compare low-income, high-wealth families with high-income, low-wealth families.

Consumption

The third interacting element of economic wellbeing is consumption. Consumption includes (a) market spending on nondurable goods and services, such as food and utility bills; (b) the flow of services from the existing stock of durables, such as the service value of cars; and, in the most comprehensive definitions, (c) goods and services that the household obtains either in-kind, such as through Medicaid benefits, or through in-home (nonmarket) production.⁹ This definition of consumption differs from that of consumption expenditure, which counts purchases of durables (during the period in which they are bought) instead of the flow of services from them. Expenditure measures also typically exclude the monetary value of in-kind transfers and in-home produced goods and services (although these are, with the exception of in-home produced services, included in the national accounts’ consumption measure). The two concepts of consumption and expenditure are obviously closely related and sometimes are even used

___________________

⁸ A family’s wealth would also reflect inherited assets, which need not originate from inside the family.

⁹ Farmers provide one example of households whose consumption may rely more heavily on home production.

interchangeably but, for some households, consumption and expenditures may diverge significantly. Examples are provided below that illustrate how the differences between the two concepts may play a role when addressing distributional questions.

A view commonly expressed in the economics literature is that consumption has several advantages, relative to either wealth or income, as a measure of a family’s or household’s current material wellbeing.¹⁰ There are a number of reasons supporting this view. The life cycle and permanent-income theories (Friedman, 1957; Modigliani & Brumberg, 1954) maintain that consumption reflects the long-term (life cycle) or permanent-income prospects of a household, which may differ significantly from current income, especially in the initial and final stages of the life course. Consumption may thus be a more reliable metric for capturing the average living standards or wellbeing of a family. Individuals may be able to support positive consumption even when they have no disposable income available if they have past accumulated assets to rely on or access to sources of borrowing. On the other hand, consumption may be below income if individuals decide to save part of their income. Attanasio et al. (2012), Jorgensen and Slesnick (2014), and Meyer and Sullivan (2013) provide additional support for the use of consumption measures, all noting that consumption patterns can be smoothed because of nonwage- and nonwealth-financed (e.g., by government programs) spending.

Consumption and consumption expenditure are also typically more equally distributed across the population than either income or wealth (see Attanasio & Pistaferri, 2016; Fisher et al., 2022). Examining the United States using data from 2019, Garner et al. (2023) find that the bottom 50% (in the distribution of expenditures) accounts for about 30% of aggregate spending as captured in personal consumption expenditures (PCE)—a measure of how consumers spend their money produced by BEA—and the top decile accounts for about 25% of spending. Meanwhile, the parallel figures for income are 19.2% for the bottom half and 39% for the top decile. The wealth statistics are even more skewed—with 2% being held by the bottom half of the population and 69% by the top decile.

For many families, consumption and income measures may largely reflect two sides of the same coin. Where the family’s adults work or look for work full-time and consume about as much as they earn, consumption is close to earnings, so families can be ranked either by consumption or by income because there is little variation in the burden of time allocated to the

___________________

¹⁰ Cutler and Katz’s (1991) and Slesnick’s (1993) account of consumption recognizes the flow-of-services of durables to households. More recent studies such as those by Johnson et al. (2005) and Krueger and Perri (2006) also examine the benefits of consumption measures as distinct from expenditures.

labor market. The fact that net financial holdings are small further confirms this conclusion. As discussed later in this report, cash earnings are well measured, but valuing cash income from some benefit programs—such as Medicaid—poses important challenges. Consumption measures tend to give reasonable results for young families, for families with low-earned income but adequate benefits, and for families with retired former earners whose consumption is close to pension and Social Security receipts.

To summarize, standard expenditure measures—which emphasize purchases of goods and services provided in the market—miss some house activities that contribute to economic wellbeing, as well as goods and services provided by the government, by other households, or by nonprofit institutions. Nonetheless, expenditure (or consumption) measures may, under some circumstances, provide a better gauge of economic wellbeing than nonhuman wealth or income measures. For this reason, and because they are relatively less developed, consumption measures once improved could provide substantial benefits for distributional analyses, even above alternative measurements of income or wealth. However, because consumption measurement interacts directly with income and wealth measurement, the three-pronged approach to improved measurement of economic wellbeing described by this report has merit.

INCOME, CONSUMPTION, SAVING, AND WEALTH LINKAGES

Standard definitions of the value flow (with I = income, C = consumption, S = saving) and stock (W) measures satisfy clearly specified accounting relationships (Deaton, 1993; Hall, 1989; Jappelli & Pistaferri, 2017). During a given period, say a year, families divide their use of resources between current use—consumption expenditures—and provision for future needs, namely saving (or dissaving, when consumption exceeds income). The accounting identity linking resources available this year as income (often called Y, but which will continue to be labeled I) with the allocation of resources between consumption C and saving S, is:¹¹

I = C + S.

The second standard accounting identity is that the flow of saving, S, equals the change in nonhuman wealth holdings across two time periods, labeled here as ΔW, net of re-evaluations and other changes to wealth (e.g., the value of housing wealth deteriorating due to a fire). The identity expressed below captures the fact that there is the same amount of saving in the

___________________

¹¹ This is called the Haig-Simons definition of income as an individual’s consumption plus net changes in wealth.

spending plan I = C + S as in the wealth accumulation plan. This principle can be expressed as:

S = ΔW (net of re-evaluations and other changes to wealth).

Thus, since two measures, such as C and W, are interrelated, one can compute I and S from C and W (assuming there is access to observations on W at the beginning and end of the year). It is equally true that one could start with I and W (or more precisely ΔW) and then compute C and S. The relation among the variables can also be written as:

C = I – ΔW (net of re-evaluations and other changes to wealth).

The expression on the righthand side, I – ΔW (net of re-evaluations and other changes to wealth) (where ΔW = change in financial wealth + net revaluations + other changes to wealth) describes the way the family finances its consumption, as the sum of income I less saving S. In other words, funds for consumption are equal to income net of saving. These well-known accounting identities give researchers degrees of freedom on the information from which they draw—data on all four measures are not needed for most analyses.

Sources and Uses of Cash in the Household

For some applications, insights may be gleaned by considering the identities captured in the equations above within a framework often used in business accounting, called sources and uses of funds. Within this framework, the righthand side of the identity “C = I – ΔW” establishes an accounting of all the cash coming in and (nonconsumption expenditures) going out of the household. The residual—the lefthand side of the equation—is then the value of the amount consumed. Income, I, can be broken down into income-type flows, which include earnings less taxes, retirement benefits net of contributions, interest and dividends less interest paid, government transfers, interhousehold transfers, and similar flows. And the change in wealth, ΔW, is net purchases of assets and net repayments of liabilities, which include bank and other financial accounts, real estate less mortgage debt, and credit card debt and repayments. This approach breaks down a family’s observed consumption into a balance sheet—possibly fairly detailed—of how the family arranged to be able to pay for its consumption. The viability of the approach rests on availability of data for all the items on the righthand side of the equation.

If a researcher has access to data on all the important sources and uses of cash for a respondent, then it may be more constructive to arrange the

information accordingly, rather than trying to fit the data into predefined categories of income and saving. Or it may lead to a more satisfactory way to arrive at definitions for income and saving. For example, suppose that a researcher has obtained, for each respondent, W-2 forms showing after-tax earnings and contributions to retirement plans, bank records reflecting interest received, deposits, and withdrawals at ATM or otherwise; credit card purchases, cash benefits, interest payments, and balances; and down payments, mortgage payments, and receipts from house sales net of mortgage repayment. Such information could produce the following examples:

1. A family has $20,000 in annual after-tax wage receipts. The family has a bank account showing an increased balance over the year of $2,000, and no re-evaluations or other changes in wealth. The inference is that its consumption is $18,000, with an inflow from work of $20,000 and an outflow to the bank account of $2,000.
2. A family like the one in scenario 1 reports a $20,000 withdrawal from a 401K retirement account, no labor earnings, and a $2,000 outflow to a bank account. This family can also be inferred to have $18,000 in consumption. This example illustrates that a researcher can apply the sources-and-uses approach without worrying about whether a 401K withdrawal is income or a wealth transfer.
3. Another family has the same earnings as above, but also withdraws $500 at the ATM for purposes that are not entirely consumption—such as repaying a friend for a loan. The inflow to the bank account is still $2,000. If the $500 is used entirely for non-consumption purposes, the family’s consumption is $17,500, whereas if it is all used for consumption, the family’s overall consumption is again $18,000. Lack of data on the use of currency is a common obstacle to assembling a complete picture of consumption.
4. Consider instead a household with no annual after-tax wage receipts, but that benefits from a holding gain on its stock investments of $25,000. It sells off part of its stock to capitalize part of this holding gain of $20,000 and puts $2,000 in its bank account. Its consumption would be $18,000 with a change in wealth of $7,000. In reverse, consider a different household that, during the year, experiences a $10,000 loss in the value of its stocks. The household sells part of its stocks ($20,000) and puts $2,000 in its bank account. In this case, its consumption would again be $18,000 but the change in wealth would equal −$28,000.
5. Consider, finally, a household that receives an intergenerational transfer of $15,000 but reports no income receipts for the year and a negative change in wealth of $3,000. If the transfer appears on the bank records as a deposit (or a wire transfer), then the inference

1. is that income was actually $15,000 and consumption for the year was $18,000. If the transfer was partly in-kind (e.g., parents paid $1,000 per month for the child’s rent expenses directly to her landlord and deposited $250 per month in the child’s account for her other spending needs), then the child’s consumption would be substantially underestimated (at only $6,000). The transfer of $15,000 should also appear as a side outflow from the donor household.

Scenario 5 illustrates that the sources-and-uses-of-funds framework not only requires detailed information on account inflows and outflows, but also may miss the consumption value of in-kind transfers that appear as consumption in someone’s else account (as in scenario 3). In the aggregate, this would not matter; but distributionally, it does matter, as it tends to overstate inequality. A correct accounting of interhousehold transfers should record the transfer as an expenditure (consumption) by the donor household financed by an income or wealth reduction, and as income or consumption to the recipient household. Since these transactions are important for many households (e.g., paying tuition, room, and board for a daughter or a granddaughter in college; or helping subsidize a child or grandchild who is looking for a job or working as an apprentice at low wages in another city), correct distributional accounting needs to reflect these transfers to maintain the budget identity above.

In each case above, a researcher could construct an estimate of economic wellbeing for the household—suitable for use in a distributional analysis—without imposing formal definitions of income and wealth.

Discrepancies

Given the complexities inherent in the real-world economy, some transactions that would cause the identities above to hold exactly are not captured because of omitted or erroneous data. Assuming that C and ΔW are correct, then the indicated values of S (saving) and I (income) can be calculated from the formulas above to generate values of, say, S* and I*. If independent measures of S and I are available, then discrepancies S – S * and I – I * can be constructed to indicate how badly the data depart from the theoretical relations. Looking at all four empirical measures may therefore be useful even within the accounting identity framework illustrated above.

Checking the discrepancies is part of data evaluation, not model building. Some agencies check the identities for users and publish the discrepancies, and some correct the data where possible. Publication of the statistical discrepancy between the NIPA—the difference between gross domestic

income and GDP—is a leading example of this principle in action.¹² Balancing national accounts’ data involves checking supply and demand for each transaction and position, cross-checking data on ICW for each sector, and checking that the flows fully explain the change in the balance sheet positions. Indeed, this is one of the main benefits of building a data system around the identities discussed above.

Capital gains present another example of potential divergence because measures of the change in wealth generally include capital gains, while a measure of income (and thus saving) may not include them unless it is adjusted to do so. But even when these issues are resolved, room for debate remains about how to account for certain hard-to-measure components. The treatment of in-kind transfers from others and the value of shelter services from home ownership are examples affecting consumption and income measurement; the value of artwork is an example of a tricky aspect of estimating wealth. Essentially, whatever is on the consumption side of the equation—in-kind health benefits and imputed rent in these examples—needs to be accounted for in a conceptually consistent way on the income side (and vice versa). These examples and other challenging measurement issues are discussed in the annex to this chapter.

Conclusions About Definitions and Identities

The identities above linking income, consumption, saving, and the change in wealth are conceptually unambiguous. To avoid inconsistencies or double counting, the accounting system for a family should generally conform to them.

Conclusion 2-1: Conceptual definitions of household income, consumption, and wealth are most useful when they are constructed to satisfy the budget identity, C = I – S. The components of income (I) and saving (S) should be consistent with a sources-and-uses framework for consumption (C). Such a framework accounts for all the value coming in and going out of the household (via saving and transfers), with the residual being the amount consumed. A fundamental principle in specifying an integrated data system is that decisions about what to count as income have direct implications for what to count as consumption and wealth.

___________________

¹² For example, BEA estimated that gross domestic income was 3.4% higher than GDP in Quarter 1 of 2022 (the September 2022 revision closed this discrepancy to 1.1%) (www.clevelandfed.org/en/publications/economic-commentary/2023/ec-202301-discrepancy-between-expenditure-income-side-estimates-us-output).

The motivation for conceptualizing the components of economic wellbeing in a way that emphasizes the ability of households, families, and individuals to consume goods and services is based on the idea that, on their own, measures of income and wealth are incomplete. Consumption captures economic wellbeing regardless of the source of funds and, in so doing, better reflects household perceptions concerning longer-term resources (permanent income hypothesis). This is why economists have found, historically, that the consumption patterns of households are also more stable than the income patterns, which can fluctuate from period to period—that is, they reflect income smoothing (Campbell & Deaton, 1989). Of course, for unusual periods—such as the COVID-19 experience and associated government relief payments—consumption itself may deviate sharply from longer-term trend lines.

In the actual data, the identity described above often does not hold. Investigators should construct and study the discrepancies, just as the BEA prepares and publishes the discrepancy between national product and national income in the NIPA. In particular, cash and in-kind transfers between households will impact the distribution of income, consumption, and even wealth, but may not affect the aggregate measures of the NIPA.

Conclusion 2-2: For the purpose of addressing distributional questions, a complete description of the budget identity (and, in particular, an accurate portrayal of consumption) requires the calculation and investigation of discrepancies therein.

DEFINITIONAL DIVERSITY IN PRODUCTION OF ICW DATA AND STATISTICS—AN INTRODUCTION TO MICRODATA ON HOUSEHOLDS AND INDIVIDUALS

The primary aim of this report is to present a plan to help government agencies collect and use data on income, consumption, wealth, and saving that shed light on the economic wellbeing of individuals and families. As has been documented in numerous reports—notably in the Report of the Commission on Evidence-Based Policymaking (Commission on Evidence-Based Policy Making, 2017)—the past decade or so has been marked by considerable upheaval in the ways information is collected for the purpose of producing social and economic statistics. Traditional data collections, including those done through telephone interview, are struggling with declining response rates, while the collection of data from government agencies and from private sources, such as credit-card issuers, is advancing rapidly (Abraham et al., 2022; National Academies of Sciences, Engineering, and Medicine, 2017, 2022).

The core historical sources of information on family-level economics are surveys conducted by the Census Bureau, such as the Current Population Survey (CPS), and by the Federal Reserve Board, sponsor of the Survey of Consumer Finances (SCF). In general, a single survey does not seek detailed information on all four key measures (income, consumption, wealth, and saving). This is partly due to the need to minimize the burden on respondents and, in turn, minimize any resulting reporting errors, as monitored by the Office of Management and Budget. Hence, surveys are typically designed with specialized measurement objectives: the CPS is about employment status, the SCF about wealth, and the CE about consumption expenditure.¹³ Concepts of ICW represented in surveys, and the accuracy with which they can be captured, are further shaped by respondents’ capabilities to understand them and to recall the needed information. Many surveys—the CE and SCF are good examples—are also quite long and burdensome, which further impacts the quality of data collected. Surveys may ask about earnings from employment, income from self-employment, pension receipts, and other types of receipts that add up to income. They may (or may not) also include items such as health benefits and the service value of an owner-occupied house. Survey administrators compile and publish statistics on income, but users of the survey data may construct their own measures from the detailed responses. Similarly, agencies publish information about the probability distributions of key variables; but users, if provided with public-use samples, can construct their own distributions. The CPS, SCF, and CE provide access to public-use data, wherein careful attention has been paid to maintaining the anonymity of respondents.¹⁴

Over the past two decades, administrative microdata have been increasingly used to refine what is known about family-level economics. Government administrative data are generated by the tax system, Social Security, and other public transfer programs. Private administrative microdata are generated by entities such as the credit reporting agency Equifax, credit card companies, banks, and companies processing scanner data (such as Nielsen). Administrative data sources are often thought to achieve greater accuracy than surveys because they generally have far larger coverage than do surveys and they are less prone to respondent reporting errors. Even so, coverage of administrative data may not be representative of the desired population because the administrative concepts are directly tied to

___________________

¹³ The interview component of the CE collects detailed information on the make, model, and purchase price of vehicles owned by the household, from which it would be possible in principle to obtain a measure of the flow-of-services estimate.

¹⁴ A description of statistical safeguards in place at the Census Bureau can be found at www.census.gov/about/policies/privacy/statistical_safeguards.html.

the purpose of collecting the data. Also, the availability of administrative data sources typically does not simplify the conceptual problems associated with measuring the determinants of economic wellbeing. In many cases, the available administrative data are based on narrower concepts than those used in surveys and may not capture all relevant items.¹⁵ For example, the IRS does not collect data on assets until they are sold because capital gains are taxed only upon realization (Gale & Vignaux, 2023). This coverage problem underlines how, for most statistical purposes, a specific definition of the target population is needed, and it may be insufficient to simply refer to the households captured in surveys and administrative data sources.

The range of definitions employed by statistical agencies for data collections and resultant publications on ICW is broad. While consistency and integrability are valued characteristics in economic statistics, a degree of flexibility must be accepted in defining ICW because of the many purposes to which the resultant data are ultimately put—for research, for statistical use, and for general information. Tables 2-1A, 2-1B, and 2-1C illustrate the range of established definitions applied to ICW by different organizations. Given their different measurement objectives, the composition of elements embedded in the definitions likewise varies.

Conclusion 2-3: Multiple definitions of household income, consumption, and wealth (ICW) are needed to examine different policy and research questions ranging from the impacts of fiscal policies to distributional analyses. To aid users of ICW statistics in selecting appropriate measurement constructs, each definition requires specification of accompanying purpose(s) and a transparent guide to its construction.

While many established definitions of ICW currently usefully coexist, these can be modified to improve consistency across measures in a way that will ultimately be more useful for research and policy, and that may dissuade advocates from misusing estimates to support a particular position.

One area of economic statistics where standardization (including across countries) has been highly consolidated is in national accounting protocols using the specific definitions included in Tables 2-1A, 2-1B, and 2-1C. National accounts provide comprehensive ICW measures, capturing all elements that would normally contribute to economic wellbeing (and elements that may not be captured in surveys). Furthermore, they ensure consistency in a way that aligns with the accounting identity discussed above. Finally, the national accounts generate important aggregates statistics such as GDP, gross national income, and human development index, providing policy

___________________

¹⁵ Slemrod (2016) fully details the caveats of using administrative data for research purposes.

makers the opportunity to directly link these aggregates to distributional results.

Conclusion 2-4: Definitions embedded in systems of national accounts provide a well-established starting point for the variety of income, consumption, and wealth concepts envisioned for the dataset recommended in this report. These definitions include an income measure comparable to the national accounts, a measure of adjusted disposable income comparable to the Organisation for Economic Co-operation and Development measure (including retirement distributions and in-kind transfers), a measure of consumption similar to personal consumption expenditures, and a measure of wealth consistent with the financial accounts, which is similar to the definition used in the regular publication for the Survey of Consumer Finances but which also adds the value of defined benefit pension wealth.

One lesson to be drawn from the national accounting world is that standardization and refinement of terminology can improve the clarity of the resulting statistics, and this would apply to ICW measures used for distributional and other analyses as well. Definitions of income, for example, may refer to “labor income” or to “total income,” where the latter includes income generated from capital sources. Similarly, “labor income” is sometimes used in reference to the “wage and salary income” of employees, excluding the self-employed, while sometimes the term is used to refer to income from all sources of labor including self-employment. The clarification (and, where possible, standardization) of terminology laid out later in this report is important for ensuring that appropriate measures are applied to specific questions, while serving to help educate the public and researchers about what precisely each term means. As another example, it may be useful to use one term when pensions are included and another term, such as total compensation, when benefits are included. Ideally, the terminology would be developed through a careful study across agencies with input from researchers and then heavily publicized and consistently used in all official publications. Just as fairly standard definitions exist for GDP, gross national product, and current accounts in national accounting, it will be important to develop analogous standards for individual- and household-level variables and statistics. Definitions are already somewhat more standardized for wealth concepts—such as liquid wealth, financial wealth, and net worth—and are used fairly consistently.

Throughout this report, the emphasis is on the development of a data system for use in the production of ICW statistics that, as described in Chapter 1, support research on disparities in economic wellbeing and the

impact of government policies and programs. Examples of these types of analyses include:

• Cross-group comparisons, including international comparisons. For example, how does the United States compare with other Organisation for Economic Co-operation and Development (OECD) countries in terms of joint IWC distributions? What are the trends?
• Assessments of the impact of major events, such as financial crises, recessions, pandemics, or wars; or assessments of the relationship between business cycle swings and the distribution of IWC.
• Estimates of intergenerational economic mobility. For example, what factors contribute to disparate intergenerational economic mobility between race/ethnic groups or different geographic areas/countries? Why has the United States become less upwardly mobile than peer countries?
• Calculations of the distributional impact of taxes and government spending. For example, what is the effectiveness of different kinds of taxes and transfers on IWC distribution?
• Study of the role of homeownership on economic wellbeing. For example, how have discriminatory housing policies/practices affected long-term wealth creation for Black Americans?
• Assessments of retirement issues/security. For example, what percentage of Americans have adequate resources to retire comfortably?

The nature of these statistical needs, which are broader than economic accounting, and the research that the envisioned dataset is intended to support provide direction in how the key ICW constructs should be defined for this purpose. The subsections that follow provide a summary of what is currently known based on the main U.S. data sources and what would be ideal to have for reaching these goals.

Income

Income has historically been the starting point for conducting welfare analysis. This is partly due to practical considerations—many surveys collect information on income but not on consumption or wealth. However, none of the different survey or administrative data sources produced by the U.S. statistical system is exhaustive or complete enough to serve all income measurement needs. Moreover, the different measures that are produced tend to differ from one another. Given the diversity of data needs and purposes (indicated in Table 2-1A), it is unsurprising that some of the definitions of income used by researchers and statistical agencies are unable to meet the theoretical goal identified in Conclusion 2-1. However, these

definitions are still useful, even though they may not completely satisfy the budget identity or aggregate to NIPA.

Some differences in income definitions arise because of the functional nature of the databases. In macroeconomic accounting (BEA, SNA, for example), the income concept is quite comprehensive, including investment income, imputed rents, and, in the case of BEA’s disposable personal income concept, in-kind transfers. As itemized in Table 2-1A, in the NIPA, for example, Earned Income Tax Credit (EITC) payments and the value of in-kind government transfers received are included, but retirement income disbursements (e.g., from 401(k) accounts) are not, even though they constitute a key resource for some older individuals and are part of disposable income as defined in the SNA. On the other hand, in line with the SNA, NIPA includes imputed interest on investments and imputed rents from homeownership.¹⁶

Other definitions are less inclusive. The Census Bureau’s “money income” measure, with its intended application in establishing eligibility for safety net programs and for estimating poverty guidelines, is mainly limited to cash inflows. As itemized in Table 2A, the desired concept for the Census Bureau’s household surveys (CPS, ACS, and Survey of Income and Program Participation) covers money income received (exclusive of certain money receipts such as realized capital gains) before deductions for personal income taxes, Social Security taxes, union dues, Medicare deductions, and the like. Hence, it includes employee contributions to Social Security and cash transfers from the government or others, but it excludes noncash transfers (from the government, employers or others) as well as imputed returns on assets. In particular, Census Bureau money income is the sum of labor earnings, unemployment compensation benefits, worker’s compensation payments, Social Security payments received, and supplemental security income received. The BLS (in the CE), uses a similar definition, though it also adds the value of the Supplemental Nutrition Assistance Program (SNAP, formerly food stamps) received, but not other in-kind government transfers. The Federal Reserve Board’s SCF also adds realized capital gains, as does the after-tax and transfer income in the CBO definition (Table 2-1A, column 7). In NIPA, both realized and unrealized capital gains are reflected only in the capital accounts and not reflected as part of personal income.

The income concept used by the IRS for tax enforcement (adjusted gross income) is likewise tailored to purpose, focusing on realized employment, business, and investment income. Administrative microdata sources

___________________

¹⁶ In the CE, households are asked: “If someone were to rent your home today, how much do you think it would rent for monthly, unfurnished and without utilities?”, which is often used to measure imputed rent for owner-occupied housing.

such as the IRS’s Statistics of Income include components of income that are taxable or potentially taxable (adjusted gross income).¹⁷ Hence, employee contributions to Social Security and defined contribution retirement plans are excluded, as are in-kind government (or private) transfers, while some taxable benefits from government or employers are included as are realized capital gains.

The Federal Reserve’s concept of family income in the SCF is similar to money income but expanded to include fungible in-kind benefits, such as SNAP, and realized capital gains. The inclusion or exclusion of safety net benefits in the definition of income is also an important distinction, driven by intended uses—for example, for the purpose of measuring poverty versus of reconciling NIPA. The income or family resources concept used in the Supplemental Poverty Measure published by the Census Bureau—which is the measure used most in poverty policy research—include many safety net transfers. The construct guiding the United Nations’ Canberra Group definition of income is whether the source is potentially available to the household for current expenditures (i.e., “Could the income component be spent today?”).¹⁸ Elsewhere the concept of taxable income, established by tax law, is typically different from income measures grounded in economic concepts, such as a Haig-Simons income definition (Slemrod, 2016). Studies of income inequality base estimates on administrative tax data to provide detailed coverage of the top end of the population distribution (e.g., Auten & Splinter, forthcoming; Piketty et al., 2018). Such data of course reflect the specifications of the tax system and are also affected by the reporting behavior of taxpayers.

Conclusion 2-5: Definitions of income that do not completely satisfy the budget identity or aggregate to the national accounts may still be useful for some statistical and policy purposes. An example is after-tax- and-transfers income as defined by the Congressional Budget Office (see Table 2-1A) but that includes both realized and unrealized capital gains.¹⁹

Hard-to-Measure Components

A review of the varying concepts of income used by different agencies indicates that much of the discretion in defining ICW centers around

___________________

¹⁷ For example, the IRS concept of income includes Social Security income even if it falls below the thresholds for taxation; non-filers incomes are also included even if none of the income is taxed.

¹⁸ See Canberra Group (2011).

¹⁹ It is worth noting that current realized capital gains include unrealized gains from previous years.

several hard-to-measure components where determining what should and should not be included is not so clear-cut. Among the most common of these methodological choices are (a) the treatment of retirement income (e.g., plan contributions and pensions, and the timing of subsequent income flows); (b) valuation of in-kind benefits and services, especially government-employer-provided health insurance; (c) treatment in income and wealth estimates of realized and unrealized capital gains; and (d) alternatives to valuing homeownership, most prominently imputed rent for homeowners.

Conclusion 2-6: Just as there are tailored definitions of income, consumption, and wealth to serve various research and policy purposes, the construction of conceptually complicated components, such as retirement income, health insurance benefits, capital gains, corporate profits, and the value of homeownership, will also entail tailored definitions.

While the annex to this chapter provides a more detailed discussion, the overarching conclusions reached by this panel regarding each of these four difficult definitional aspects of income are as follows:

1. Retirement income. There are four primary types of inflows for retirement income: employer contributions, employee contributions, interests and dividends earned, and capital gains on retirement accounts. The retirement outflows—namely benefits received or withdrawals made from retirement accounts—are the usual focus of published income distribution analysis. Table 2-1A shows how income definitions from different agencies or sources differ in how they treat these inflows and outflows. Some of the differences arise from trying to achieve consistency with NIPA. In line with the guidance about sources and uses of household resource flows, for the concept of income it will typically be desirable to include benefits received. For the purpose of most distributional analyses, such as those produced by CBO, the preferred income concept will include retirement income distributions. The CBO approach is similar to that used by OECD but different from that used in BEA’s (unadjusted) disposable personal income estimates. The different types of pension schemes (e.g., Social Security vs. occupational plans) and benefit formulas (i.e., defined benefit vs. defined contribution) and their inclusion or exclusion may affect estimates of ICW (see annex to Chapter 2).
2. Health insurance benefits. Health insurance provided by employers or the government has a value to recipients. A job that includes employer-sponsored health insurance is preferable to an otherwise identical job, paying the same amount, that does not. Similarly, a

2. 65-year-old with Medicare is surely better off than an uninsured but otherwise identical 64-year-old. As suggested in the CBO measure of after-tax and transfer income, and OECD measures of adjusted disposable income, total health benefits from employer-provided health insurance and government-provided health benefits are included at their full cost. This is similar to suggestions in Levy (2023) and Carpenter and Simon (2023). However, recent research (see Case & Deaton, 2020; Finkelstein et al., 2023) suggests that only some portion of the total cost of health benefits should be included in income (see the annex for further discussion).
3. Capital gains. While treatment of capital gains is perhaps most central to defining and measuring wealth (and is therefore discussed in greater detail in the next section and in the annex to this chapter), for some purposes it may be appropriate to consider the impact of capital gains on income. While all gains, interest, and change in value will be captured in the total wealth of households, it may be difficult to capture the changes in wealth between one year and the next. These changes are the capital gains that could be taxed if they are realized or left to continue to gain value if unrealized. It is these changes that, for some applications, are useful to capture in estimates of income (as in the CBO definition of market income). The CBO definition of after-taxes-and-transfers income including realized capital gains is useful, even if it does not completely satisfy the budget identity or aggregate to NIPA.
4. Imputed homeowner rent. Imputed or implicit rent is an estimate of the amount of income that homeowners earn from effectively renting their residences to themselves. Homeownership frees up resources that would otherwise have to be spent on rent. For this reason, for most distributional analyses, estimates of the level of resources available to individuals and households attempt to include homeowner access to housing assets that deliver a cash flow of rental income or a flow of housing services to the owner. Incorporating implicit rent into resource calculations is a fairly common practice in economic statistics. As identified in Table 2-1A, Disposable Personal Income as conceptualized in the NIPA produced by BEA and by the Canberra Group Handbook is a prominent example. Several surveys, among them the CE, collect information on imputed rent for homeowners by asking respondents who own a house how much it would rent for. The panel believes this type of question, while imperfect, should be retained and used to measure the value of the services from the most important durable that most households own, their house.

Consumption and Consumption Expenditures

As mentioned above, consumption—the activity whereby families consume goods or services such as food, housing, transportation, education, health, childcare—holds some advantages relative to disposable income as a measure of wellbeing or of living standards, because consumption captures the long-run expectations of one’s available resources.

In the United States, data on consumption expenditures are collected by the BLS through the CE and aggregate measures are constructed for the Personal Consumption Expenditures. The primary goal of the CE is to measure expenditure shares for revising the relative importance of goods and services represented in the market basket underlying the Consumer Price Index (CPI). While the CE is a short rotating panel,²⁰ other panel datasets collecting more long-run information on consumption, such as the Panel Study of Income Dynamics, report substantial variation in consumption over time for individual families, indicating the existence of a transitory term that complicates the use of its consumption estimate as a measure of wellbeing.

Consumer expenditures as collected in the CE include some elements that are not in NIPA or not included in a comparable manner. One reason for this is that the latter must balance estimates of income and production at the level of the economy as a whole (which is not limited to consumer spending). For example, the treatment of used motor vehicles differs; aggregate expenditures calculated from the CE tend to be much higher than those calculated from the PCE for this category because person-to-person sales are included in the former but not in the latter.²¹ Also, published CE means (as opposed to the variables available in the CE database) include expenses related to owner-occupied housing that NIPA treats as property income expenses (in the case of mortgage interest payments), investments (in the case of major repairs), or intermediate costs (in the case of maintenance). NIPA also has a different definition of financial services paid, in that it includes food produced and consumed on farms, and it also includes imputed financial services such as banking services and imputed rent from owner-occupied housing. Another set of issues involves the evaluation of medical services when they are covered partially or totally by insurance companies or the government (Medicare/Medicaid), since they do not represent spending for the individual household, but they do represent consumption and are also treatable as a source of income (in-kind transfers, as in the examples above).

___________________

²⁰ Households in the CE are followed for at most four quarters, so much of the variation referred to here is attributable to seasonality.

²¹ www.bls.gov/cex/pce_compare_0203.pdf

TABLE 2-1A Established Definitions of Income

NOTE:

+ denotes inclusion in income; (-) means subtracted from income; AGI = adjusted gross income; BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; CBO = Congressional Budget Office; CE = Consumer Expenditure Survey; EITC = Earned Income Tax Credit; FRB = Board of Governors of the Federal Reserve Board; HH = household; LIHEAP = Low Income Home Energy Assistance Program; OCED = Organisation for

Economic Co-operation and Development; PSZ = Piketty, Saez, and Zucman; SNA = System of National Accounts; SNAP = Supplemental Nutrition Assistance Program; SSI = Supplemental Security Income; TANF = Temporary Assistance for Needy Families; UI = unemployment insurance; WIC = Woman, Infants, and Children Program.

^a Since employer and employee contributions are both added and subtracted, they are effectively absent.

^b Interest paid (except on mortgage loans) is subtracted from income in the OECD measure and treated as an outlay in the BEA measure.

^c Cash or in-kind benefits from others (e.g., other households, institutions, governments, abroad) would normally be captured in the SNA, although benefits from others may not show up in the aggregated amounts if these benefits are within the household sector, but conceptually, it is still part of income.

^d The SNA only recognizes in-kind transfers from the government and nonprofit institutions serving households. Any other in-kind benefits (such as gifts received from another household) would normally be recorded as a benefit in cash with the recipient household consuming the relevant good/service.

^e As with note d, this would not apply to transfers from one household to another.

^f Social transfers in-kind included in adjusted disposable income.

SOURCES: Panel-generated from a variety of sources (Auten & Splinter, forthcoming; Canberra Group, 2011; CBO, 2023; Fixler et al., 2020; Guzman & Kollar, 2023; Piketty et al., 2018; Zwijnenburg, 2019).

TABLE 2-1B Established Definitions of Consumption

NOTE: Consumer expenditures also includes household contributions to pensions. BLS = Bureau of Labor Statistics; CE = Consumer Expenditure Survey; OECD = Organisation for Economic Co-operation and Development.

^a Net foreign travel is spending by U.S. residents abroad less spending by nonresidents in the United States.

^b Part of this may be captured by government spending on health and education. The part that relates to interhousehold transfers would appear as transfers in cash, with the actual purchases being reflected as done by the donor household (so captured in the various underlying categories).

^c Only home production of goods is included as part of production and consumption in the SNA. This item also appears on the income side (ensuring consistency with the budget identity).

SOURCES: Panel generated from various sources (European Commission, 2024; Fisher et al., 2015; Garner et al., 2023; Meyer & Sullivan, 2023).

TABLE 2-1C Established Definitions of Wealth

NOTE: DB = defined benefits; DC = defined contributions; OECD = Organisation for Economic Co-operation and Development; SCF = Survey of Consumer Finances.

^a Consumer durables are not included in the SNA wealth estimates; however, consideration is being given to including it in their distributional measures as a supplementary item.

^b Life insurance and unfunded DB pensions are included in the SNA wealth estimates.

^c The SNA/OECD measure includes business wealth, such as other dwellings, farmland, machinery, and inventories. However, it does not include detailed information on miscellaneous assets and liabilities at the household level (e.g., money owed between households) that appear in other sources such as the SCF. The SNA also distinguishes financial derivatives and other accounts receivable and valuables, although most of these items are very small.

^d While not included as part of SNA wealth, public pension entitlements are being considered for distributional analyses as supplementary item.

SOURCES: Panel generated from various sources (Batty et al., 2019; European Commission, 2024; Saez & Zucman, 2016).

As with income and wealth, there are practical issues to consider when defining consumption. While some purchases are clearly nondurable (e.g., sport event tickets, food outside the home, gasoline, hotel bills) and others are clearly durable (e.g., refrigerators, cars, furniture, laptops), for other goods the distinction is less obvious (e.g., some apparel, sundry items). Moreover, even when the distinction is clear-cut, measuring, say, the flow of services from durables is problematic. Some households wear down a given durable at a faster rate than others because of different usage; some households purchase durables of differing quality, which may provide unequal services per unit of time. Surveys that contain information on durables do not provide a precise inventory of all the durables owned by households; and when they do, information on brand or vintage is typically missing. As shown in Table 2-1B, some definitions of consumption include the service flows from durables instead of the purchase price. For example, the CE contains information on the service flow of durables that are implicitly included in rents. Unfortunately, BLS stopped collecting an inventory of the stock of durables owned from the second quarter of 2013 onwards, and only information on the stock of vehicles is now collected in the CE.²²

As discussed above, one important aspect of measuring consumption is that, in the data, it is often proxied by expenditures. Some measures produced by statistical agencies reflect individuals’ or households’ current-period expenditures on food, housing, nondurables, transportation, durables, education, health, and childcare. Others attempt to measure consumption more fully by including current-period expenditures on goods and services, plus imputed service flows from durables, including those purchased in previous periods. The most quantitatively substantive methodological choices in measuring consumption will often concern the treatment of owner-occupied housing services and government-provided health care. Including an imputed rental income—as is done in BEA’s PCE or in the proposed Meyer/Sullivan concept of consumption—or government spending on health care will often swamp the impact of other definitional choices.

Because expenditures do not equal consumption, BLS has embarked on a program to produce a more comprehensive measure of economic wellbeing than is possible using data on expenditures alone. The measure will be “based on the theoretical concept of consumption and will include the value of goods and services purchased within a certain time period and assumed to be consumed within the same period (e.g., food purchased at a restaurant, household utilities), the flow of services for owner-occupied housing and the stock of vehicles, in-kind benefits, and home production.”²³

___________________

²² Using data from the CE (interview component), Cho et al. (2024) estimate a consumption measure for automobiles based on a user-cost approach.

²³ www.bls.gov/cex/consumption-home.htm

This concept of consumption should more closely capture the wellbeing implications for people.

Conclusion 2-7: An additional definition of consumption that would be useful for distributional and other analyses is one that would include rental equivalence (for homeowners) and service flows from vehicles and add health and education spending but excludes time use. The measure proposed by the Bureau of Labor Statistics in its initiative to develop a comprehensive consumption measure would go part of the way by adding (to the Consumer Expenditure Survey) estimates of imputed rental income and the consumption component of education (but not investment components, of which tuition is one), but also including time use and excluding expenditures on health insurance.

Wealth

As with income, wealth is defined in a variety of ways across agencies and by purpose, although, relative to the multitude of income specifications, its treatment tends to be more similar across datasets. Table 2-1C lists the range of definitions of wealth, which essentially reduce to the total value of assets owned by the household minus the value of liabilities or debts. In the SCF (first column of the table), which is the source of the most detailed information on wealth in the United States, unrealized capital gains typically enter the calculation of net worth through their inclusion in the potential liquidation value of a given asset (i.e., how much the household would get from selling the asset). The distinction between liquid and illiquid wealth is particularly important because, over the lifecycle, people’s wealth typically shifts to less liquid forms. It is the panel’s view that most distributional analyses of wellbeing should follow the SCF approach to estimating wealth.

While this preferred conceptual approach may be comparatively simple to implement for assets that have well-identified market prices, such as bonds or shares of publicly listed companies, the issues are more complex for shares of private companies, for housing, and for other real assets. The inclusion of some assets is also controversial (e.g., durables besides vehicles); while other assets are notoriously difficult to price (e.g., art pieces). Further, debts include secured debt (mortgages, vehicles) and unsecured consumer debt (credit cards), but also student loans, legal debts, medical debts, and possibly others such as child support debt. In many cases these debts are hard to evaluate; for instance, would one include all medical debt or only medical debt in collection, a much smaller but still notable amount (Rae et al., 2022).

The most important decision in scoping the definition of wealth is how to treat pension benefits. The value of retirement wealth held in defined contribution funds is typically included in estimates of wealth. However, how to treat the value of defined-benefit retirement wealth is less obvious, because estimates rest on several assumptions, not all of them uncontroversial. In principle, from a household perspective, one could include in the definition of net worth the present discounted value of future defined-benefit payments, with appropriate discounting (including discounting from life tables, survivors’ rights, etc.). Furthermore, it may be debated whether pension wealth should also include Social Security benefits or only occupational pensions and life insurance. As indicated in Table 2.1C, the SCF and the Federal Reserve’s Distributional Financial Accounts differ on this choice.

Conclusion 2-8: A useful definition of wealth for assessing the ability of households to finance current and future consumption is that used for the Bulletin publication for the Survey of Consumer Finances, which includes both debts and net assets.

The annex to this chapter includes a more detailed look at these contentious definitional elements underlying the definition and estimation of families’ wealth.

ANNEX TO CHAPTER 2: DIFFICULT DEFINITIONAL AREAS IN ICW MEASUREMENT

This chapter annex discusses a number of difficult issues regarding measurement of concepts entering the definition of income and, in some cases, definitions of wealth and consumption. To support the panel’s guidance on how to handle these issues, this annex describes complexities associated with categorization and measurement in four topic areas: (a) retirement income, (b) in-kind benefits and services (especially, government-employer-provided health care), (c) realized and unrealized capital gains, and (d) imputed rent for homeowners.

Issues regarding treatment of retirement assets to wealth definitions differ depending on whether retirement wealth is from private or public sources and, for private retirement wealth accounts, whether it comes from defined contribution (DC) or defined benefit (DB) plans (Table 2-1C). In general, for private retirement plans, a clearer relationship exists between contributions made, interest earned, and benefits received over a lifetime. A present discounted value of such wealth can be constructed using either a contributions or distributions approach—the two are equal so long as the

appropriate discount rates are used. One could also differentiate DB from DC plans, but the distinction is not a central one in the sources and uses of funds for the consumption approach emphasized in this chapter. That said, some of the wealth definitions in Table 2-1C exclude DB plans, especially unfunded ones, from the concept of wealth considered.

Differences in pension funding may also be relevant in the evaluation of public (and private) policy choices. It is well known that private DB plans are being slowly phased out. (In 1990, the aggregate asset value of DB plans was 1.6 times larger than the aggregate asset value of DB plans; in 2018, the aggregate asset value of DB plans had declined to two-fifths of that of DC plans (see Jang & Youchang, 2020). For public pensions, there is still a theoretical connection between contributions and benefits (e.g., through Average Indexed Monthly Earnings formulas).

The challenge of estimating values of retirement resources in the measurement of wealth is that they do not have precise (and stable) current market values. The usual measures of the expected discounted value of accrued DB pension wealth are conceptually comparable with current wealth, and the approach entails relatively few assumptions. But such a measure omits what may be seen as an insurance value of possible future accruals and time-related kinks in benefits. An alternative, the expected discounted value calculated at the point when benefits begin (for those not yet retired) is often used in inequality assessments. However, computing such a measure requires additional assumptions that typically have a high degree of uncertainty. Moreover, the approach may not be comparable with those used to estimate other wealth values; at best, to make them comparable would require projecting forward all other wealth items to the same period, with corresponding levels of uncertainty.

As is the case for private retirement wealth, evaluating Social Security wealth depends on the appropriate discount rate and assumptions about other factors. However, as indicated in Table 2-1C, Social Security wealth is typically excluded from most definitions of wealth—including that used in the SCF definition—in part because it may be unfunded and, in some cases, governments do not have a legal obligation to payments as would be the case with many occupational pensions.²⁴ As expected, when Social Security is taken into account, the distribution of wealth across the population becomes less unequal (Sabelhaus & Volz, 2019).²⁵

For generating retirement income, there are four primary types of inflows to consider: employer contributions, employee and household contributions, interest and dividends earned, and capital gains on retirement

___________________

²⁴ unstats.un.org/unsd/nationalaccount/aeg/2016/7_1_Pension_Entitlements.pdf

²⁵ See also Aubry et al. (2023) on the racial wealth gaps for Social Security wealth.

accounts. And, basically, there are two methods of including pensions in income—either accounting for it when earned or contributed, or accounting for it when used or distributed. Table 2-1A shows the different allocation methods and definitions used to serve the purposes of various organizations. For example, the Census Bureau’s money income and CBO’s after-tax- and-transfer income both include the distribution of pensions but treat the contributions to pension funds as savings in the working years.

The contributions approach, however, has an important drawback for analyses of income distribution. There is no place for retirement benefits received in the contributions approach, and, consequently, the sources of funds available for retiree consumption appear very low. To address this shortcoming, one could adopt a comprehensive measure of sources of funds available for consumption that included, for example, the annuitized value of accumulated retirement assets, but there are both conceptual and empirical difficulties with estimating accumulated retirement assets and the annuitized value. The most common treatment in the inequality literature focuses on benefits received and withdrawals made—the so-called “distributions” method. Table 2-1A shows that the Census Bureau’s money income and the CBO methods follow this treatment, which counts the distributions from retirement accounts as income. Again, employee contributions are part of earned income (and hence included).

Distributing pension contributions when they are earned vs. when they are distributed can yield very different measures of inequality and income estimates for the aged population. As discussed in the panel presentations by Poterba (2023), Gale (2023), Quinby (2023), and Mitchell (2023), it is important in measuring the economic wellbeing of the elderly to include distributions from retirement accounts in their measure of income. Brady and Bass (2023) show the composition of income after age 55 and demonstrate that retirement income eventually becomes the largest share of income (see Figure 2A-1; and Figure 15 in Brady & Bass, 2023). As a result, measuring retirement income when it is earned (as employees) will mean that the elderly will not have much income; however, their consumption and wealth will still be significant. Treating retirement income at distribution, however, creates a disconnect between saving when young and dis-saving when old. It becomes more complicated because many households are saving, contributing to pensions, and obtaining distributions from pensions all at the same time (see Monaco & Pierce, 2015). Nevertheless, the panel believes that to consistently measure inequality, using the distribution method for disposable income provides the best alternative.

One possible method is to distribute retirement flows in a way that is consistent with national income accounting. Employee contributions are part of earned income, so they are already part of national income. Adding employer contributions gives us a comprehensive measure of the current share of national income that is allocated to future retirement benefits. The interest and dividends earned on private retirement accounts completes the allocation of national income—nothing else is needed to fully distribute current national income.

The BEA approach (see Table 2-1A) also follows the NIPA method; however, it distributes “income” from retirement using the contributions approach. This can also be seen in the BEA Table 2.1 distributions. Social Security employer and employee contributions are both added and subtracted, so they are effectively absent. This approach ignores private pensions (DB and DC) and government DC benefit distributions (see Fixler et al., 2020).

The SNA/OECD (see Table 2-1A) starts with primary income, capturing the income received as part of involvement in production (i.e., labor income and property income), then makes adjustments to obtain (adjusted) disposable income, which better reflects the resources available to households. Basically, pension contributions as paid by the employer are included in primary income as part of the return on labor input. They are deducted as paid in by households into the social insurance pension schemes when moving to (adjusted) disposable income. Distributions (or social insurance pension benefits) are included in (adjusted) disposable income. The methods are similar for investment income on DB and DC schemes. The returns are regarded as part of property income in the SNA; in the redistribution account (going from primary to disposable income), they are then deducted as (imputed) social contribution paid in by the household. Any contributions are deducted from primary income to arrive at (adjusted) disposable income, whereas any benefits are added to primary income. Contributions include both the employer contributions explicitly acknowledged as part of primary income as well as those paid directly by the household itself and not earmarked to a specific source (i.e., they may be funded in different ways). Because all social contributions (both employer related and employee contributions) are deducted from primary income to arrive at (adjusted) disposable income, this means that whereas social contributions as paid by the employer are feeding into primary income, they do not affect disposable income, as they are deducted going from primary to (adjusted) disposable income, and hence there is no double counting. Social benefits, both from employer-related DB and DC schemes as well as from

Social Security, are only reflected in the redistribution account (see OECD Guidance Note).²⁶

On average, about “half of welfare state transfers in rich nations are in-kind benefits” (Garfinkel et al., 2006). A long-standing challenge in the construction of income (or joint ICW) statistics, and the underlying datasets, is how to treat these in-kind benefits provided by employers or government. Most estimates of income produced by statistical offices take into account the value of in-kind benefits, in part because it is important to be able to track the impact on wellbeing of these benefits. As itemized in Table 2-1A, BEA’s disposable income concept includes the value of employer-provided health insurance, plus a value for government-provided Medicare, Low Income Home Energy Assistance Program, and a range of other safety net programs. As discussed in greater detail later in the report, a range of options exist valuing in-kind benefits. CBO (2023), for example, assigns the government’s cost of in-kind transfer program benefits (such as for Medicare and SNAP) to those program’s participants. For public goods, CBO has previously allocated the resources by “using a combination of two methods: half allocated in proportion to each household’s share of the population and half allocated in proportion to each household’s share of total income” (p. 8).

Because of their significant impact on the estimates of the resources available to people with lower incomes, the inclusion of in-kind (and cash) programs has a significant impact on income distribution statistics. As expected, once the impact of government benefits and taxes is taken into account, the final incomes of households in the lower quintile groups are higher than market incomes, while final incomes at the top are lower (see

___________________

²⁶ The SNA explains that “Pension schemes are treated in the SNA as having liabilities towards the households with claims on the schemes. The payments of pension contributions into the schemes and the receipts of pensions by pensioners constitute the acquisition and disposal of financial assets. However, this may not accord with the perception of the households concerned, especially pensioners’ households, who tend to regard the pensions they receive as income in the form of current transfers. […] In order to present income information that may be more useful for analysing the behaviour of the households concerned, the payments of pension contributions to all pension schemes and to social security and the receipts of pensions by pensioners’ households under both pension schemes and social security are recorded in the secondary distribution of income account as social contributions and social insurance benefits, respectively. They therefore affect the level of disposable incomes of households. […] However, as is clear in the financial account, the change in pension entitlements is part of household net worth. It is therefore necessary to adjust saving for the difference between contributions payable and benefits receivable shown in the secondary distribution of income account” (European Commission et al., 2009).

Figure 2A-2). This is a result that is replicated across developed countries.²⁷ The impact of various in-kind programs on the incomes of those in the lower income ranges are also captured in poverty statistics (for estimates of the impacts, see Guzman and Kollar, 2023; National Academies, 2019).

This section focuses on the value of health insurance benefits, because it is the in-kind benefit that affects the most households and that has the greatest quantitative impact on estimates of income levels and distribution.²⁸ The health insurance example is also sufficiently complex to illustrate the main issues that arise when valuing in-kind benefits more generally.

___________________

²⁷ The magnitude of these effects can be seen, for example, in the UK Office of National Statistics estimates of income distributions before and after transfers that are reported in their annual articles on “Effects of Taxes and Benefits on Household Incomes” (www.ons.gov.uk/peoplepopulationandcommunity/personalandhouseholdfinances/incomeandwealth/bulletins/theeffectsoftaxesandbenefitsonhouseholdincome/financialyearending2022); Australia produces similar estimates (www.abs.gov.au/statistics/economy/finance/government-benefits-taxes-and-household-income-australia/latest-release).

²⁸ Mean health insurance premiums for employer provided health insurance was $11,764 in 2019 (Finkelstein et al., 2023); In 2021, the average Medicare cost per beneficiary in the United States was $15,309 (The Board of Trustees, 2021).

Health care, and health insurance as a means for obtaining it, is a fundamental element of people’s wellbeing. Reflecting this view, many countries guarantee the right to medical care services. In the United States, health insurance coverage is more complex and more heterogeneous. Beginning in the 1940s, a distinctive feature of U.S. medical care emerged: health insurance became increasingly bundled in total compensation packages by employers.²⁹ Now, about half of the U.S. population, and most with private insurance, receive coverage through employer-sponsored health insurance (ESI). Others in the population receive health insurance through government programs such as Medicare and Medicaid, buy it privately, or are uninsured. And coverage differs with income, with higher-income household more likely to receive ESI and lower income households receiving government program coverage.

How health insurance benefits are treated is also highly consequential to the measured distribution of economic wellbeing. While including government- and employer-provided health insurance in estimates of income tends to reduce measured inequality overall, a measure of the impact that combines the two obscures the fact that employer-provided benefits are skewed toward the high-income earners (Figure 2A-3 demonstrates the distributional differences of including health insurance benefits in measured income). Carpenter and Simon (2023), Burkhauser et al. (2012), and others have shown that it is government-provided health insurance coverage, valued at cost of provision, that reduces inequality in the population distribution of wellbeing, relative to money income measures (CBO, 2023). Of course, one would not observe this distributional impact in a country with national health plans.

The focus of this section is on how to attribute health insurance benefits in estimates of income or resources available to families, as opposed to how to value consumption of health care more broadly.³⁰ It is useful to begin with a few observations about the unique nature of health insurance transactions. In some respects, health insurance is not unlike other goods and services. Some people would assign a value to a given product that is higher than its price (for them, generating consumer surplus), while others will value it by far less than its cost. For people who fall into the latter group, theory tells us that they are unlikely to voluntarily purchase the

___________________

²⁹ The 1942 Stabilization Act, designed to combat war-time inflation by limiting employers’ freedom to raise wages, created an incentive for employers to offer health benefits instead to compete for scarce workers.

³⁰ In the context of designing National Health Accounts, National Research Council (2010) addresses the question of how to conceptualize inputs (expenditures) and outputs (health gains) of the U.S. health care system.

product. However, when an in-kind service such as health care is provided to an individual or family by employers or government, the total cost of the plan no longer directly enters the consumer’s decision rule; the result is that some people consume health insurance who might not otherwise. The same scenario applies to any in-kind transfer that cannot be resold (Currie & Gahvari, 2008). It is this “nonmarket” element that creates complexity in estimating the income value of health insurance benefits.

As recommended above, and following the suggestions in Levy (2023) and Carpenter and Simon (2023),³¹ the panel favors including medical care benefits provided by employers or government for most research purposes, including production of distributional statistics. The logic for arguing that at least some portion of health insurance benefits should be counted as income is straightforward. Employer- or government-provided health insurance has value to recipients; thus, zero is clearly too low a value to assign to such benefits. Additionally, the provision of benefits by employers has been shown to reduce the wages paid to employees, further confirming

___________________

³¹ These white papers, commissioned by the panel to inform its deliberations, are available at www.nationalacademies.org/our-work/an-integrated-system-of-us-household-income-wealth-and-consumption-statistics-to-inform-policy-and-research.

its central role in compensation schemes.³² In other countries, including the United Kingdom, where the state provides medical care, wages do not adjust in this way even though the health benefits are valuable, as they are in the United States.

Hence, the main question is whether statistics designed to track the distribution of income within the population should value employer-provided or government-provided health insurance at cost, dollar for dollar (α = 1). Or whether the benefit should be discounted to something less than full cost (α < 1), because benefit may not be fully fungible in the way that wage income is (i.e., since the benefit cannot be traded for cash, even at a discount, although in many cases it does free up resources that can be used elsewhere). As with the income definitions, the choice of α may depend on the particular research question. Table 2-1A lays out purposes and applications of income data along with the appropriate, associated treatment of insurance and other in-kind benefits. In their use of the ICW microdata sets, researchers will continue to determine the value of α that is appropriate for their analyses. However, for published joint ICW distribution statistics, agencies will need to make choices about how health insurance benefits are to be treated in their income (and, possibly, consumption) estimates.

In estimating individuals’ or households’ income, when does it make sense to treat a health insurance benefit essentially the same as wages? This question presents an instance where a micro–macro measurement disconnect may arise. For many statistical purposes—perhaps, most prominently, national income accounting—measures that count the premium cost of an insurance benefit as income are needed. To monitor aggregate economic activity, the central purpose of NIPAs, all income in the economy must be included so that accounting identities match. Therefore, the BEA measure of personal income includes all private and public health insurance transfers

___________________

³² This compensating wage differential response may also effect the employment process by weighing the preferred benefit profile of some workers more heavily than others (Gruber & Krueger, 1991; Tilipman, 2022). Saez and Zucman (2020) point out that, because health insurance premiums are similar across workers, the wage penalty is the same for a low-wage secretary as it is for a highly paid executive, and therefore is regressive for moderate-income workers. It is also worth noting that, when compensation is paid in the form of nontaxable health insurance instead of cash, high-wage earners receive a larger (absolute) tax break than do low-wage earners because of their position in a higher marginal tax bracket.

(implicitly, α must = 1).³³ For the same reasons, other organizations’ (e.g., OECD, Canberra) concepts of disposable personal income also include health insurance benefits, even if health care is covered by national health care schemes (see Table 2.1). In their commissioned papers, Levy (2023) and Carpenter and Simon (2023) both recommend using α = 1 for Medicaid, Medicare, and ESI, with some suggestions that if the recipient value was desired, then α could be less than 1 for Medicaid.

Hence, if for a given research or policy purpose, a value of α less than 1 is desired for the household income calculation, then to maintain accounting identities, the (1-α) portion of costs must be shifted to another sector—for example, to the corporate profits of hospitals, insurers, pharmaceutical companies, device manufacturers, lobbyists for the industry—and NIPA income would still need to be reconciled.³⁴

Saez and Zucman (2020) apply a “distributional national accounts” framework to study inequality both over time and, ultimately, across countries. A comprehensive income estimate that includes health insurance transfers (which requires imputing and assigning an average amount to beneficiaries of each program) is therefore integral to their approach. The authors do, however, explicitly acknowledge the problem that this “market” value will not typically be equal to the “recipient value”—essentially an amount that beneficiaries would be willing to pay for the provided plan. To address this disconnect, Saez and Zucman suggest alternative analyses that “assign the perceived cash value of individualized in-kind transfers to recipients, while treating the rest as a collective public good” (p. 20).

The above described “market value” approach, based on unit averages for program spending, is practiced elsewhere in government.³⁵ For example, in its annual reports on the distribution of household income (Congressional Budget Office, 2023), CBO applies an after-transfers-and-taxes concept wherein the full cost of employer contributions and a quasi-market value of Medicare and Medicaid are used. The government transfer element is “defined to equal the average cost to the government for providing those

___________________

³³ Specifically, the value of employer contributions to health insurance premiums is included in the category “supplements to wages and salaries,” and the values of Medicare, Medicaid, and Premium Tax Credits for the purchase of private nongroup marketplace coverage are all included in the category “personal current transfer receipts” as “government social benefits to persons” (Bureau of Economic Analysis, 2023).

³⁴ In the SNA, a logical option would be to shift the remaining amount from individual consumption (i.e., the part that is benefiting households directly) to collective consumption by government (i.e., the part that is benefiting the society as a whole, including elements such as public administration, the police force, road maintenance). This also seems in line with the approach by Saez and Zucman explained below.

³⁵ One issue is that there is not a singular market price for health care services. For example, Medicare and Medicaid rates are around 70 and 50%, respectively, of what private insurers pay (Lopez et al., 2020; Zuckerman et al., 2009, 2021).

benefits” (Congressional Budget Office, 2023, p. 40). Of the four in-kind benefits that CBO incorporates into its income measure, Medicaid is by far the largest.³⁶

In the broader research literature, investigators (e.g., Armour et al., 2014; Burkhauser et al., 2013; Larrimore et al., 2021) have applied the Haig-Simons framework that includes the value of health insurance and other in-kind transfers in estimates of income. The framework explicitly accounts for the increase in potential consumption that accompanies such transfers (Alm, 2018, p. 382). Burkhauser and others (U.S. Council of Economic Advisers, 2019) applied the concept in their development of a Full-Income Poverty Measure, which incorporates a value of health insurance equal to full market costs (estimated as the average premiums for employer-provided health insurance) which is then incorporated to estimates of post-tax, post-transfer income.

Another approach has been proposed in the poverty measurement context that caps the “market” valuation of insurance in estimates of household income/resources. For example, the Health Inclusive Poverty Measure developed by Korenman and Remler (2016) includes health insurance as a basic need in its calculation of a poverty threshold. The basic need amount is set to the cost of a defined “benchmark” health insurance plan (National Academies, 2023). To balance the equation, on the resource side employer-provided insurance is capped at the same benchmark level, thereby setting a ceiling on the value of the benefit in cases where employers provide high-coverage plans. These high-coverage plans might be more generous than some employees would choose to pay on their own, even if wages were increased by the amount of the provided plan instead. As stated by Levy (2023), the Health Inclusive Poverty Measure approach cleverly solves the problem of non-fungibility by offsetting the imputed value of the transfer added to resources with an equivalent addition to the estimate of needs. However, in the context of measuring inequality, this method—capping the income value of the insurance benefit at the cost of a standard insurance plan—is somewhat problematic because, in some cases, it would estimate that a person with a high-value plan is no better off than a person with a basic/standard plan. This suggests a possible solution that estimates the income value of a basic health insurance plan using an α = 1, but a lower (but still positive) α for plans above a certain threshold.

Finkelstein et al. (2023) suggest that, in some situations, households may experience an α > 1. They note that, if health insurance is only available

___________________

³⁶ This full measure of income used by CBO is defined as income before transfers and taxes + means-tested transfers - federal taxes, defining the values of Medicare, Medicaid, and the Children’s Health Insurance Program as the average cost to the government of providing these benefits.

through an employer, employee risk aversion could produce a dollar-for-dollar value of health insurance that eclipses that of wages. The ex-ante value may also be higher than cost if individuals are unable to purchase insurance of the same quality or at the same price as that negotiated by employers or government.³⁷

As described above, several important statistical purposes require an income concept that includes the full premium cost of an insurance benefit as income. However, for other purposes it may be desirable to define and measure income in a way that has other properties but may not satisfy the budget identity at the aggregate level. This deviation occurs most commonly when the concern is with micro-level questions where the recipient value is most relevant. The drawback of aggregate data is that, by design, it does not readily provide insight into the distribution of economic resources or their growth or decline.

Theoretically, there are several reasons why the value of an in-kind health insurance plan (such as Medicaid or Medicare), reflecting its utility to the recipient, may be less than its “market” or full cost (Smeeding, 1982). Of course, individuals value all goods and services differently, and hence make different consumption choices. Therefore, in the absence of an ESI-based system (and government-provided health care), not all individuals/families would choose the same policy among those offered by the market, and, in turn, they would spend different amounts on insurance. For example, young or healthy individuals would typically buy less expensive, perhaps high-deductible plans, while older or less healthy individuals would be willing to upgrade for fuller coverage; indeed, young adults are the most likely to go without insurance altogether (Conway, 2020).

Additionally, a health insurance benefit has the characteristic of not being fully fungible—individuals cannot typically trade in an employer-provided benefit for cash that can be reallocated elsewhere in their budgets. For the purpose of estimating family resources in their poverty measures, during the period 1988–2005, the Census Bureau developed a fungible value approach to estimating the contribution of Medicaid and Medicare to people’s income, by assuming that the recipient value for poor households was negligible (see Levy, 2023). For a period, CBO followed suit, using a fungible values approach in its income construct. While the approach offered a conceptually attractive way of thinking about the period-to-period

___________________

³⁷ Additionally, as detailed in National Academies (2022), the ex-poste value of an insurance policy often deviates from its ex-ante value because utilization is unpredictable. Such a scenario arises for holders of a policy that costs, say, $10,000 but who end up using no medical care, or who end up consuming $100,000 of medical care during a covered time period.

resources available to families, the agencies found it problematic to estimate in practice (Levy, 2023).

In part because the above-described idiosyncrasies of the U.S. health care system, Finkelstein et al. (2019) estimate that Medicaid beneficiaries value each dollar that is actually spent on them at $0.50 to $1.20. Further, using data from the Oregon Health Insurance Experiment, they find that 60% of Medicaid spending for the study population did not go to beneficiaries, meaning that the overall ratio of beneficiary willingness-to-pay for Medicaid to gross cost (corresponding to what is referred to above as market value) is between 20 and 40%, the same value that Smeeding found in his work (Levy, 2023; Smeeding 1982).

An argument that has been raised in opposition to equating the cost of all health insurance benefits with cash income has to do with inefficiencies in health care provision, which have the effect of transferring resources from plan beneficiaries to hospital executives, pharma companies, and some doctors. Case and Deaton (2020) estimate that, judging by costs elsewhere (e.g., Switzerland), the United States has upwards of $1 trillion in excess costs. With these leakages in mind, the authors are adamant that the recipient value of private insurance should be discounted when considering the level of resources available to families:

Additionally, Cutler (2006) has shown that cost increases in U.S. medical care, even when reflecting new and improved treatments, are not as valuable to low-income workers as to higher-income workers, which carries distributional effects. Another part of the increase in insurance costs, he notes, is attributable to high-wage earners going to high-priced providers, which is worth it to them but not others. For most analyses, it is consistent with established estimates used for income inequality to set α = 1 for all sources of health insurance. However, researchers and agencies may want to use alternative estimates for α to assess the importance of health

insurance benefits in both vertical and horizontal equity, and especially for government programs such as Medicaid and Medicare. These options could include α = 1, but with a capped level that relates to income (something along the lines of the old Census Bureau “fungible income” method) and using different alphas for different insurance coverages.

To understand the impact of health insurance in determining economic wellbeing it would be useful to provide the impacts on inequality of including Medicaid, Medicare, and Veteran’s Health Care, along with ESI. It would also be useful to consider alternative values for α where it is less than one for both private and public health insurance. Ultimately, the choice of whether to set α = 1, α = 0, or to something in between will be dictated by what is appropriate for the specific question or analysis. This is what researchers already do, and it could be a user option for the ICW dataset. Even the ICW statistics could include two income series, one assigning α = 1 to the valuation of employer- and government-provided health insurance, and the other assigning α a value less than one.

The various definitions of income in Table 2-1A differ with regard to the treatment and inclusion of capital gains and losses. Only a few definitions include realized capital gains, and none include unrealized capital gains. Excluding unrealized capital gains is problematic because it means that series are affected by business decisions to operate as pass-throughs vs. corporations, an important issue when comparing inequality over time and/or across countries. These considerations reverberate onto the definitions of wealth. Any increase in the value of one’s assets, whether they are realized or not (e.g., housing), is typically incorporated in wealth, as it is in the accounting identities discussed above. Capital gains may translate into higher consumption opportunities, even if unrealized, so long as people can borrow against them, as shown in the “wealth effect” literature. Indeed, financial markets have created products, such as home equity lines of credit, that are designed to allow consumers to achieve precisely these goals. It is of course complicated in some cases to measure unrealized capital gains. Some countries have administrative data on individual listed shares owned at the end of each year, so knowledge of (publicly available) prices is all that is needed to compute unrealized gains or losses. When only the value of one’s stock portfolio is available, and no sales or purchases are recorded, it is complicated to distinguish between unrealized capital gains on the stocks owned at the beginning of a holding period and additions/subtraction to the portfolio that take place during the year. For other assets (e.g., housing, private businesses), valuations are difficult because of missing market prices. In some cases, evaluations are elicited through a survey. For housing,

surveys such as the CE ask respondents about the value of their house if they were to sell it. In the SCF, entrepreneurs are asked about the value of their businesses. Ideally, for some purposes, it would be useful to have the capacity to count unrealized capital gains and losses.

Another challenging definitional question when constructing ICW statistics is how to think about the value of owner-occupied housing as a resource to households and, indeed, as a consumption item. Ideally, estimates of available resources should reflect that homeowners have an asset (a home) that delivers a flow of rental income, whether they are renting it out to someone else or to themselves. “Imputed or implicit rent is an estimate of the amount of income that homeowners earn from effectively renting their residences to themselves” (National Academies, 2023, p. 69). Perhaps more intuitively, the benefit of homeownership, at least if a home is owned outright, is that resources are freed up that can be redirected to finance other household budget needs. Some countries (e.g., Iceland, Luxembourg, the Netherlands, Slovenia, and Switzerland) even tax the flow of implicit rent earned by homeowners, as if they were actually paying themselves (Andrews & Sanchez, 2011).

The essential idea underpinning the “rental equivalence” method is that the valuation of owner-occupied housing services can be approximated based on the dollar amount that homeowners would pay for their home if they were renting it. In keeping with this construct, the SNA states as a guiding principle that “housing services produced are deemed to be equal in value to the rentals that would be paid on the market for accommodation of the same size, quality and type” (European Commission, 2008, p. 187).

One problem with the alternative—not including implicit rent in the income estimate—is that it could lead to inequitable treatment of households, even if they have the same incomes and identical homes. A simplified example (simplified in the sense that it abstracts away things like differential tax treatment) from National Academies (2023, p. 69) illustrates:

Incorporating implicit rent into resource calculations is fairly common practice in economic statistics, including Disposable Personal Income as conceptualized in the NIPA produced by BEA and by the Canberra Group Handbook. Within the Canberra taxonomy, the value of owner-occupied housing falls into the category “Income from household production of services for own consumption” (Canberra Group, 2011). Rent is included in income on a net basis—that is, the imputed value of the shelter services received subtracts owners’ housing-related costs, including interest paid, incurred.³⁸ The Handbook further specifies that imputed rent estimates should be “presented separately from estimates for other services, so that data is available to support different types of analysis,” including those for which investigators wish to exclude the resource contribution of owned housing. The BEA treatment of owner-occupied property is similarly imputed as the net income of the owner (Mayerhauser & Reinsdorf, 2007).³⁹ Specifically, in calculations of GDP, depreciation, maintenance and repairs, property taxes, and mortgage interest expenses are netted out.⁴⁰ In this calculation, BEA estimates the dollar value of shelter services based on the amount that homeowners would pay if they were to rent a comparable unit in the same local market.

As indicated in Table 2-1B, estimates of the flow of owner-occupied housing service are also included in measures of consumption and expenditure. BEA’s PCE and OECD’s consumption measures are examples. Elsewhere, in 2021, BLS initiated a program to produce a consumption measure that would provide a more comprehensive measure of economic wellbeing than is possible with expenditures alone that includes the service value of housing and other consumer durables.⁴¹

___________________

³⁸ Garner and Short (2009) consider the strengths and weaknesses of different methods for valuing net implicit rental income.

³⁹ When looking at disposable income of the homeowner, the SNA approach would likewise deduct mortgage interest. However, for GDP calculations, only the implicit financial services related to the mortgage interest payments, not the interest payments themselves, are deducted. The main reason is that interest is not regarded as resulting from production (that would only be the implicit financial services provided by the financial institution) but just a remuneration for the provision of financial capital (in the same way as dividends). This is then recorded as a property income flow in the primary income account instead of as resulting from the production account.

⁴⁰ See Mayerhauser and Reinsdorf (2007) for a description. The CE includes questions about homeowner costs such as mortgage payments, homeowner’s insurance, property taxes, and maintenance and repairs. The Annual Social and Economic Supplement to the Current Population Survey does not ask about property-related expenditures.

⁴¹ www.bls.gov/cex/consumption-home.htm

Operationally, in the context of poverty measurement, National Academies (2023, p. 76) suggests that, given currently available data, the Census Bureau has three options for estimating rental equivalence:

That report ultimately recommends the first approach, based on Fair Market Rent values that adjust with family size, minus user costs. For renters, that report similarly recommends imputing the rental value for renters who receive government rental assistance or live in public housing.

Determining whether homeowners’ user costs should include the principal portion of mortgage payments in a poverty measure is an unresolved question. For establishing thresholds in the Census Bureau’s Supplemental Poverty Measure, mortgage payment cost estimates include both interest and principal components. However, most economists would agree that the interest-only portion of mortgage payments represents a more accurate depiction of user cost because payments to principal are effectively a long-term investment into an asset. For this reason, the report of the Interagency Technical Working Group on Evaluating Alternative Measures of Poverty (Bureau of Labor Statistics, 2020) states that, in calculating homeowners’ net implicit income, costs incurred should include only mortgage interest payments.

It is important to note that economic statistics do not universally invoke a rental equivalence approach. As referenced in Table 2-1A, CBO’s definition of after-tax-and-transfer income and the Federal Reserve Board’s SCF definition of family income do not include owner-equivalent rents; nor does the IRS’s definition of adjusted gross income (for obvious reasons). For price measurement purposes, 11 of 17 OECD members do not include owner-occupied housing in their CPIs (Organisation of Economic Co-operation and Development, 2013a). The U.S. CPI does use an owner-equivalent rent approach.

While for some measurement purposes it may be conceptually inappropriate to include imputed owner rents in income, the arguments against doing so for statistical measures are often practical in nature, having to do with the fact that the rental equivalence values of owner-occupied homes

are not observed. On surveys, it may be difficult for respondents to understand the concept; additionally, accurate estimates require access to reliable data about the market rent that could be charged for a given home. However, as outlined in National Academies (2022, 2023) and in the documentation provided by the statistical agencies, considerable progress has been made in fine-tuning the needed data and estimation methods.