|Burkhauser, Richard V., Larrimore, Jeff, Simon, Kosali I.|
Researchers considering levels and trends in the resources available to the middle class traditionally measure the pre-tax cash (market) income of tax units or the pre-tax, post-transfer size-adjusted income of households. We demonstrate that these choices and others carry significant implications for assessing income trends. Based partially on estimates from Piketty and Saez (2003) and the
When we broaden the income definition to include taxes, transfers, and the ex-ante value of employer and government provided health insurance, the 28 year growth in tax unit median income increases from 3.2 percent to 18.2 percent. Even this measure, however, understates median income growth during this period. Many individuals share economic resources with individuals in their household that are in separate tax units, such as cohabiting couples or grown children living with their parents. Furthermore, the number of persons living in both households and tax-units has declined in recent years, meaning that the same income provides greater levels of consumption for each individual in these sharing units than in earlier years. Incorporating these additional factors, we find median income growth improves to 36.7 percent.
So which income series is superior? For researchers interested solely in how middle class Americans are compensated for their time in the labor market, it is more appropriate to use pre-tax, pre-transfer (market) income, although even here researchers who ignore the dramatic increases in the ex-ante value of employer health insurance will understate the returns to work and disproportionately do so for workers in middle class households. However, for those interested in the overall economic resources available to individuals, it is more appropriate to consider income defined as broadly as possible, consistent with the traditional Haig-Simons measure of comprehensive income as the best measure of ability to pay tax.
We demonstrate the importance of these choices by showing their impact on who gains from the employer provided health insurance tax exclusion. When individuals are arrayed by where they fall in the distribution of non-size-adjusted post-tax, post-transfer income plus the ex-ante value of health insurance of tax units, most benefits go to the top quintile and almost none to the bottom two quintiles. In contrast, by arraying these same individuals by where they fall in the size-adjusted household income distribution, the receipt of benefits is substantially more balanced throughout the income distribution.
The most basic measure of the economic resources available to the average American–median household income–has been consistently tracked by the
This highly visible measure of the decline in the real economic resources available to middle class Americans is often discussed alongside research using Internal Revenue Service ORS) administrative records by Piketty and Saez (2003) and Saez (2009) showing that the fraction of market income going to the top 10 percent of tax units is at its highest level since at least 1917. Together, these findings suggest that the middle class is not sharing proportionately in the fruits of American economic growth. Such concerns have been manifested in the popular new media (Johnson, 2007; Piketty and Saez, 2007; Goldman, 2008; Lahart and Evans, 2008; Leonhardt, 2008), and have led to calls for policies that would increase the share of income growth going to the middle of the income distribution. For instance, when forming the
In this paper, we offer new evidence that forms a "second opinion" on the extent to which middle class Americans have failed to benefit from economic growth over the past three business cycles (1979-2007). Using cross-sectional data to capture the economic resources available to individuals at the same point in the income distribution over time, we find that the evidence of a middle class decline is far from clear, and that such results are highly sensitive to how available resources are measured. (1) Thus, we will argue that the apparent failure of the median American to benefit from economic growth can largely be explained by the use of an income measure for this purpose that does not fully capture what is actually happening to the resources available to middle class individuals.
Researchers considering long term income trends have traditionally based their analyses on one of two data sources. The first, used by Piketty and Saez (2003) and others, is
The second data source, reported each year by the
The distinction between tax units or households as the sharing units to be analyzed or the choice of the resources counted as income are somewhat abstract concepts and may appear to be trivial. As a result, literatures that differ in their income measures of choice or sharing units are often viewed interchangeably. Indeed, it is often the case that an individual's tax unit and household unit are exactly the same. A tax unit typically consists of an adult, his or her spouse, and any dependent children. Such a tax unit would include all of the members of a "traditional family arrangement" household. However, there are many situations in which this is not the case. For example, cohabiters, roommates who share expenses, children who move back in with their parents, or older parents who live with their adult children are households that contain more than one tax unit.
If measures of the level and trend in the shared resources of middle class individuals were insensitive to the choice of sharing unit, then these differences would be immaterial. However, we will show that the choice of sharing unit and which of its resources are counted as income makes a substantive difference in measures of the resources available to middle class Americans, as will controlling for the number of people in the sharing unit. Furthermore, the inclusion of taxes and transfers, in addition to the value of employer provided health insurance benefits,
Several earlier papers have recognized the importance of income definitions in evaluating economic resources. For example, Karoly (1994) shows how much including different mixes of taxes and income transfers impacts family income inequality. Similarly, Meyer and Sullivan (2009b) show how pre-tax and post-tax income definitions impact measures of poverty in their argument that consumption based measures would more precisely capture poverty trends. Additionally, both the
Others have recognized the importance of including the value of in-kind compensation in measures of market income, since cash compensation alone does not provide a complete measure of payment for work. For example, Pierce (2001, 2010) uses the Employment Cost Index (ECI) data to consider how levels and trends in labor compensation change when he includes employer contributions to fringe benefits (including health insurance). Chung (2003) extends this insight by merging data from the ECI into the CPS. Additionally, previous research by Burkhauser et al. (forthcoming) demonstrates how choices of income definition impact measures of income inequality at the top of the income distribution, and Meyer and Sullivan (2009a) consider similar questions for income inequality measured using 90/10 ratios–the ratio of the income of the individual at the 90th percentile of the income distribution to that of the individual at the 10th percentile.
Here, we expand on this previous research by using March CPS data to report the median resources of Americans as well as the growth in those resources by quintile over the last three business cycles (1979-1989, 1989-2000, and 2000-2007), using different assumptions regarding the sharing unit, tax treatment of income, and the various sources of income we include in our resource measure. When we analyze median income in the CPS data using tax units, we find that the pre-tax, pre-transfer income (the market income) of the median tax unit decreased over the 2000-2007 business cycle. This is the case whether we focus solely on those tax units who file a return or on all tax units regardless of whether they file a return. Potentially more disturbing, the median pre-tax, pre-transfer income of all tax units (filers and non-filers) only increased by 3.2 percent in real terms over the entire period between 1979-2007. These results are consistent with the view that the typical American has not gained much from economic growth over the last 30 years.
But when we broaden the sharing unit to the household, account for economies of scale in household consumption, and recognize that the payment of taxes or the receipt of tax credits as well as government transfer income and in-kind benefits all impact the economic resources available to individuals, we find the story changes. Specifically, when using our broadest measure of available resources–post-tax, post-transfer, size-adjusted household income including the ex-ante value of in-kind health insurance benefits–median income growth of individual Americans improves to 36.7 percent over the period from 1979-2007, and by 4.8 percent between 2000-2007. Similarly, these choices impact the observed distribution of income and the extent to which incomes at the top of the distribution are growing faster than those of the middle and lower classes.
We then provide an example of why such a broader measure of available economic resources is of value in considering the distributional impacts of public policy. We do so by showing how the distribution of benefits from the tax exclusion of employer provided health insurance differs when it is measured across the not-size-adjusted income of tax units and the size-adjusted household income of individuals–two common approaches to capturing such policy effects. Using our broader measure of household-based income, we show that the value to the middle class of the tax exclusion of employer provided health insurance benefits is greater than that observed when focusing on tax unit-based income. We conclude that researchers would be well served by using these broader measures of income and sharing unit when considering the distributional impacts of public policy proposals.
To explore the trend in the available resources of middle class Americans over the past 30 years, we use the public use March CPS data set supplemented with cell-means to overcome topcoding of high incomes in the March CPS. (2) One limitation of the CPS data discussed by Ryscavage (1995) and Jones and Weinberg (2000) that remains even with the use of cell-means to overcome topcoding is a change in survey methods between 1992 and 1993 that limits comparability across these years. Thus, in all series the changes in income between 1992 and 1993 are suppressed and assumed to be zero given the trend-break resulting from survey redesign. The approach used in this analysis to overcome this break in the CPS data is similar to that used by Burkhauser et al. (forthcoming) and Atkinson, Piketty, and Saez (2011). Since incomes are being compared across years, all income is adjusted for inflation to
While the March CPS is commonly used for measuring levels and trends in income and its distribution in
To impute tax credits and liabilities, we enter marital status, state of residence, age, number of dependents, detailed income information, and other factors for each tax unit into
Along with tax credits and liabilities, we also consider the ex-ante value of in-kind health insurance benefits. While the March CPS does not capture the premiums paid for health insurance coverage, it does ask respondents whether they are insured and the source of that coverage. Using the type of coverage and information about the individual's employer, we impute the ex-ante value of employer contributions to health insurance and the value of public health insurance from outside sources. The value of employer contributions for health insurance comes from the cell-means of employer contributions from the Medical Expenditure Panel Survey Insurance Component (MEP-SIC). This includes the employer contribution for single and family plans separately, by state, year, and firm size. (7)
Using the March CPS data supplemented with the health insurance and tax data described above, we calculate income distributional statistics for sixteen different series, which are based on four income definitions, two sharing unit definitions, and two methods of size-adjusting income for different size tax units and households. The income definitions, sharing unit definitions, and size-adjustment methods are as follows.
A. Pre-Tax, Pre-Transfer (
This income series considers total cash market income of the sharing unit excluding realized and unrealized capital gains. Specifically, it includes income from wages and salaries, self-employment, farm income, interest, dividends, rents, trusts, and retirement pension income but excludes public cash transfers that are not included in market income. These CPS-based income sources closely match the taxable income sources Piketty and Saez (2003) include in their analysis of
B. Pre-Tax, Post-Transfer Income
This is the income measure the
C. Post-Tax, Post-Transfer Income
This income measure further broadens the income definition by incorporating tax credits and liabilities. The tax credits and liabilities are imputed using NBER TaxSim 9.0 and the procedure described in the previous section.
This income measure partially accounts for the fact that not all resources come in the form of cash compensation. While the cost of employer or government provision of in-kind benefits may be more or less than their value to the recipients, they have positive value and should be considered in a complete measure of available economic resources. The most important employer provided non-cash compensation is the ex-ante value of employer contributions to employee health insurance premiums. Since we consider post-transfer income, we include both the ex-ante value of employer provided health insurance and the ex-ante value of government provided health insurance via
E. Tax Unit Sharing Unit
For each series with the tax unit as the sharing unit, individuals living in a tax unit are assumed to only share their economic resources with other members of that tax unit and with no one else living in their household. The procedures discussed previously, which mirror those used by Piketty and Saez to determine the number of potential tax units, are used here to impute tax units.
Not all Americans file an income tax return, which is an important issue that Piketty and Saez (2003) address. (9) While Auten and Gee's (2009) finding that 91 percent of adults age 25-64 file a tax return illustrates that this problem may be less significant than some believe, focusing only on tax filers is likely to impact median income levels and trends. Thus, to more closely approximate the entire U.S. population, we consider both filing and non-filing tax units. Because the CPS is a random sample of the entire population, it naturally includes both filers and non-filers.
Nevertheless, given the uncertain importance of non-filers in tax return based research, we will briefly discuss a series that excludes non-filers. The filers-only series uses the
F. Household Sharing Unit
For each series with the household as the sharing unit, all individuals living in the same household are assumed to share economic resources. Thus, rather than aggregating income to the tax unit, this approach acknowledges sharing of resources within a household and aggregates income up to the household level. That is, the private incomes of all tax units within a household are combined. (10) Similar to the tax unit income series for filers and non-filers, this measure includes all households regardless of the filing status of the individuals in the household.
G. Non-Size-Adjusted Income of Sharing Units
Income series that are not specified as size-adjusted measure income at the sharing-unit level and treat sharing-units of all sizes equally. For example, a single-individual in a household making
H. Size-Adjusted Income of Persons
This measure moves from the sharing unit to the individual as the unit of analysis. In doing so, it acknowledges that the resources available to any person in a sharing unit, given some level of income, vary with the number of persons sharing that income. That is, an individual in a household that is comprised of only that individual and makes
Following the customary procedure in the income inequality literature, sharing unit income is deflated using an equivalence scale to account for economies of scale by dividing sharing unit income by the square root of the sharing unit's size (Atkinson and Brandolini, 2001; Gottschalk and Danziger, 2005; Smeeding, Rainwater, and Burtless, 2001; and Burkhauser et al., 2011). (11) This size-adjusted income series is commonly used by CPS based inequality researchers in the international literature. Auten and Gee (2009) also recognize the importance of using the individual as the unit of analysis in their work using
Using each combination of income definition, sharing unit definition, and size-adjustment method described above, we first track the growth in the economic wellbeing of middle class Americans over the past 30 years by measuring changes in the median income of the entire population. We then narrow our focus to six series that encompass some of the most widely used of these definitions to look in more detail at income trends over each peak-to-peak business cycle since 1979. (12) By comparing values over peak-year to peak-year of each business cycle, we are able to focus on long-term trends in income devoid of cyclical economic conditions.
There are, of course, numerous factors such as increases in education, the aging of the population, and the influx of immigrants which will impact income trends in the cross-sectional data we observe. While it is important to know which of these factors account for these income trends, our focus is on the sensitivity of measured income trends to alternative choices of sharing unit and income definition rather than the underlying causes of these trends. (13)
A. Median Income
The first measure we consider of the trend in the economic resources of the middle class is the median income of all Americans. Table 1 provides a matrix of median income growth for the 29-year period 1979-2007 using each possible combination of income definition, sharing unit definition, and size-adjustment method described above.
In the upper-left corner of the matrix is the pre-tax, pre-transfer (market) income of tax units, which most closely matches the definitions used by Piketty and Saez (2003). When using this definition, we observe a total increase in median income of just 3.2 percent in real terms over the 29-year period. Going down the first column, the income definition broadens but the sharing unit remains the tax unit and there are no adjustments for the number of people in the tax unit. Here, the observed median income growth improves to 6.0 percent for pre-tax, post-transfer income, 9.5 percent for post-tax post-transfer income, and 18.2 percent for post-tax post-transfer income including the ex-ante value of health insurance (all in inflation-adjusted terms). Thus, just broadening the income definition to recognize the growth of transfer income, the decline in taxes and increase in tax credits, and the fact that an increasing portion of middle class compensation comes in the form of non-cash benefits increases median tax unit income growth from 3.2 percent to 18.2 percent.
Moving from Column 1 (non-size-adjusted tax unit income) to Column 2 (non-size-adjusted household income), we see that using the tax unit as the sharing unit also limits measured income growth. For each income series, household median income growth is between 9 and 11 percentage points greater than when the sharing group is tax units. For instance, simply shifting from a tax unit to a household sharing unit almost quadruples market income growth from 3.2 to 12.5 percent. To some extent, this may reflect tax units joining into households because they are unable to afford their desired standard of living separately. In these circumstances, since the individuals prefer to live separately, household income may overstate their well being since it fails to account for the negative utility from sharing a dwelling. But it is also true that they are undoubtedly better off than if they had to live as separate tax units since they made the choice to combine into a single household. As previously mentioned, results using the market income of tax units are largely the basis for the view that those at the top of the income distribution have become wealthier while the income of middle class Americans has stagnated since 1979. However, the perspective portrayed in Column 1 of Table 1 is much gloomier than the one in Column 2 where the sharing unit is the household.
Panels A and B of Table 2 provide an explanation for these differences. Over the past three business cycles, the proportion of households with only one tax unit declined from 80.3 to 76.2 percent while the proportion with three or more tax units increased from 3.9 to 5.6 percent. Two major factors precipitated this shift in the number of tax units living in the same household: (1) an increase in the number of cohabiters and other unrelated individuals living in separate tax units but sharing the same dwelling; and (2) an increase in related individuals living in separate tax units but in the same dwelling, such as adult children living with their parents. The 4.1 percentage point drop in one tax unit households (Panel A of Table 2) is analogous to the 4.2 percentage point drop in households containing only one unrelated tax unit (Panel B of Table 2). However, the growth in three or more tax unit households originated largely from increases in related tax units– as shown by the much smaller increase in three or more unrelated tax unit households. Because the number of multiple tax unit households increased for both these reasons, measures of median income based on the tax unit will report smaller income growth than measures of median income at the household level that recognize that shared resources occur beyond the tax unit.
A criticism of each series discussed so far is that they make no adjustment for the fact that the number of persons per household (and tax unit) has declined over the past three business cycles (Panels C and D of Table 2). This has resulted in household resources being shared over fewer people which, holding sharing unit income constant, increases the available resources per person.
The third and fourth columns of Table 1 account for the change in sharing unit size over the period of our analysis by making the individual the unit of analysis and scaling available income based on the size of the tax unit and household respectively. When doing so, median tax unit income growth increases between 11-15 percentage points (comparing Column 1 to Column 3) and median household income growth increases between 8-10 percentage points (comparing Column 2 to Column 4). With the most inclusive income definition, which recognizes the sharing of resources between all individuals in a household, and recognizes the economies of scale in household consumption, median income growth was 36.7 percent (Row 4 of Column 4) over the past 3 business cycles. This figure is over 10 times the 3.2 percent growth (Row 1 of Column 1) observed in the initial series, which considers only the market income of tax units without adjusting for tax unit size. In summary, broadening the income definition, capturing households rather than tax units, and adjusting for the household and tax unit size each broadly increase measured median income growth over the past three business cycles.
We now turn to the trend in income over each separate business cycle and focus on just six series which encompass some of the most commonly used income and sharing unit definitions, as well as two of our broader measures. The importance of each intermediate change should not be viewed as absolute since the stacking order matters for the importance of each. For example, as demonstrated in Table 1, size-adjusting has a larger impact on tax unit income than on household unit income. Rather, these measures are chosen for additional emphasis given their widespread use in the
The first series, which is not shown in Table 3 since it only became available in 1993, is the pre-tax, pre-transfer (market) income of tax units that file a return. In this series we find that median income fell by 2.7 percent in the 2000-2007 business cycle. Column 1 of Table 3 adds non-filers to the sample and reports their pre-tax, pre-transfer (market) income. This series most closely matches the Piketty and Saez (2003) tax unit sample. As can be seen from Panel A of Table 3, which presents this information as the total median income change over each business cycle, the median pre-tax, pre-transfer tax unit income of the entire population of tax units declined even more (5.5 percent) in the 2000-2007 business cycle.
Column 2 of Table 3 adds cash transfer income to the income definition and allows income to be shared across all members of a household rather than only within a tax unit. This series approximates the one the
Column 3 of Table 3 presents the most common income definition in the CPS-based inequality literature. This series, household size-adjusted, pre-tax, post-transfer income of the median person, rose by 23.6 percent over the three business cycles–well above the 3.2 percent increase for the median market income of tax units–and fell by only 0.1 percent in the 2000-2007 business cycle.
Columns 4 and 5 of Table 3 provide our series with broader income definitions than have traditionally been used in the literature. In Column 4, it is evident that including taxes and measuring post-tax, post-transfer, size-adjusted household cash income results in even faster median income growth. The gains in median net of tax income are most significant over the 1980s business cycle but occurred over all three periods. Median income now rises by 29.3 percent over the entire period and by 1 percent between 2000 and 2007. Declines in tax rates over the past 30 years together combined with the rise of numerous credits have substantially positively impacted the median person's net of tax income.
Finally, Column 5 of Table 3 adds the ex-ante value of employer and government provided health insurance to post-tax, post-transfer, size adjusted household income. Because these data are only available for the last two business cycles, we conservatively assume the ex-ante value of employer and government provided health insurance increased at the same rate as all other post-tax, post-transfer household income over the 1979-1989 business cycle. However, in both the 1989-2000 and 2000-2007 business cycles, accounting for health insurance payments further increased median income growth. While the 2.2 percent increase in median income growth from the increased value of health insurance was dwarfed by the overall 16.6 percent growth in income in the 1990s, accounting for the increased fraction of middle class compensation received in the form of health insurance in the 2000s results in median income growth increasing from 1 to 4.8 percent over 2000-2007. Hence at least part of the recent decline in the economic resources of the median American observed previously is the result of a shift in the way compensation and government transfers have been provided over the last two decades, as more compensation and transfers have come in the form of in-kind benefits.
Panel B of Table 3 reports the annualized median income growth in each business cycle to account for the different lengths of each business cycle. However, the patterns discussed above with respect to Panel A of Table 3 are the same. For instance, an anemic growth rate of 0.12 percent (Row 4, Column 1) in the market income of tax units over the entire period increases by 10 times to 1.31 percent (Row 4, Column 5) for the post-tax, post-transfer, household size-adjusted income including the ex-ante value of health insurance of persons. Likewise, an annual drop of 0.79 percent over the 2000-2007 business cycle in the market income of tax units (Column 1) is transformed into an annual increase of 0.68 percent in Column 5.
B. Ouintile Incomes
While median income is a straightforward measure of how the average American is faring, it is also valuable to use a measure of the middle class that allows one to compare growth in the middle of the distribution to growth in the two tails. (15) We do so below by comparing income trends for each quintile of the population.
Table 4 shows the growth in mean income within each quintile of the distribution for each of our primary income series, holding the boundaries of the middle quintiles constant in real terms over each of the last three business cycles. We also provide the mean income of the top 10 and the top 5 percent of the distribution as well as the
During the 1980s business cycle (Panel A), relative income growth across the five quintiles of the distribution follows almost the same pattern in each income series. In each series income growth is more rapid for the third and fourth quintiles than for the first two quintiles, and the top quintile of the distribution showed the fastest growth.
Income growth in the top 5 percent of the distribution is the greatest of all. This result is consistent with earlier findings that income inequality rose rapidly during the 1980s and that inequality growth occurred throughout the distribution (Piketty and Saez, 2003; Burkhauser et al., 2011).
Ostensibly the first income series, which only considers market income, is an outlier across the five measures. That is, using this first series, it appears that over the 1980s the poorest two quintiles actually become poorer–i.e., they experienced negative growth. However, when in-cash government transfers and taxes are included and adjustments are made for household size (Column 4), these negative values become positive, albeit at very low levels. Unlike for the results shown in the first column that considers only market income, the broader measures of income used in the other columns suggest that while income growth was highest in the top quintile, it was substantial in the third and fourth quintiles and at least positive in the bottom two quintiles.
To show how the different growth pattern in Panel A translates into overall changes in inequality, in the bottom two rows we report the
These patterns changed dramatically during the 1990s business cycle. Growth occurred in mean income in all quintiles across all five measures. It was higher in the bottom quintile than the other quintiles across all measures and even greater than in the top 5 percent in all but one measure. But like the 1980s business cycle, the middle quintiles made the most significant gains in terms of mean income growth when shifting to measures that better capture sharing units, household size, government taxes and transfers, and health insurance. Comparing Column 1 with Column 5 shows that economic growth is higher in all five quintiles, but the two lowest quintiles actually experienced the greatest growth over this period with the top three quintiles growing at about the same rate. Additionally, while income inequality (measured by the
A similar picture emerges for the most recent business cycle. When only considering private income at the tax unit level, the top quintile of the distribution saw the smallest decline in their incomes (-1.6 percent for the top quintile, compared to -4.9 percent for the middle quintile and-43 percent for the bottom quintile). However, adjusting for household size and including transfers, taxes, and health insurance compensation changes results in income growth instead of declines in all quintiles. This growth is balanced across the distribution with the middle three quintiles (e.g., 4.9 percent for the middle quintile) displaying more rapid growth than either the top (3.1 percent) or bottom (2.2 percent) quintiles. This more complete income measure also shows a slight decline in inequality of 0.002
Finally, Panel D illustrates the change in income for each quintile over the entire 29-year period. Due to the previously observed rapid growth in top incomes in the 1980s, under all five income series the top quintile had greater income growth than the other four quintiles and the top 5 percent had the greatest income growth. Hence income inequality rose as measured by the
Given that income growth is more balanced throughout the distribution with the more comprehensive income measure, had Piketty and Saez (2003) used this income measure their observed growth in top income shares may have been less substantial. This is particularly true during the two business cycles since 1989 when the income growths for the top 5 and 10 percent of the income distribution were slower than that for the middle quintile with the most comprehensive income definition, but were faster than that of the middle quintile when using the cash market income of tax units definition.
Hence, these more inclusive measures of access to economic resources suggest that income inequality increased in
V. IMPACT ON PUBLIC POLICY DEBATES
Thus far, we have focused on how sensitive income measures are to changes in sharing unit and income definitions. However, these choices are also important when considering the impact of policy changes. The importance of income definitions for understanding policy debates should be self-evident. Taxes and transfers have a real impact on the well-being of the individuals paying the taxes and receiving the transfers, which is why proposed changes in these policies are often controversial. Since these components of incomes and expenses impact Americans' available resources, we should include their effects in the statistics used to evaluate the success of public policies. By considering post-tax, post-transfer income including the value of health insurance, we can better capture the effect of program changes on people's economic resources. Had we been able to include irregularly received income such as capital gains from investments or home sales, other non-cash government transfer income and employer compensation, or imputed rents, median income growth likely would have been even stronger since many of these uncaptured sources increased over the past three business cycles. (16)
Less obvious, but similarly important, is how the choice of sharing unit influences the observed distribution of benefits of a given policy change. Although the distinction is often overlooked, this choice and the choice of whether to adjust for the size of the sharing unit can have profound effects on where individuals fall in the income distribution and consequently on the distribution of the benefits of policy changes. While an individual's location in the income distribution is positively correlated across sharing unit measures, this correlation is not perfect. Some low-income individuals in one distribution will be reported as having high-income in another depending on which sharing unit measure is employed.
Table 5 illustrates the extent to which the distributions of "not size-adjusted tax unit income" and "size-adjusted household income" differ–even when the income definition (post-tax, post-cash transfer including the ex-ante value of health insurance) is the same. If individuals in each of the quintiles in the tax unit distribution fell in the same quintiles of the household unit distribution, they would all lie on the diagonal of Table 5. This is not the case. Just 57 percent of individuals in the bottom quintile of the tax unit income distribution are also in the bottom quintile of the household income distribution (11.5 percent of the 20 percent of individuals in this quintile). The other 43 percent live in households with income above the bottom quintile. Overall, only 46 percent of all individuals are in the same quintiles of both distributions.
The different alignments of the distribution based on the sharing unit subsequently influence the perceived economic progressivity of public policies. To illustrate this general statement, we consider who benefits from the tax exemption for employer provided health insurance. A similar analysis could be performed for any policy, including restructuring
Table 6 provides the relative mean benefit from the tax exemption of employer provided health insurance by quintile across both the not-size-adjusted tax unit income distribution and the size-adjusted household income distribution, respectively. In both cases, we normalize the mean benefit of the entire population to 100 to focus on the relative benefits across the distribution. When we focus on the tax unit, it appears that the value of the health insurance tax exemption is largely concentrated among the top of the distribution while the bottom of the distribution receives little benefit–the average individual in the top quintile received 94.2 percent above the mean benefit in the population while the average individual in the bottom quintile received just 1.2 percent of the mean benefit in the population. Thus, the average individual in the top quintile receives 166 times the benefit of the average individual in the bottom quintile.
Although the relative size of the mean benefit is still highest near the top of the distribution when we consider size-adjusted household income, the spread is less extreme. Using this income series, the top quintile of the distribution receives 40.9 percent above the mean benefit and those in the bottom quintile receive 18.1 percent. Thus, using this series, the mean benefit going to the top quintile is less than 8 times that going to the bottom quintile. More generally, the values of the mean benefit in the 2nd and the 3rd quintile also rise substantially and the size of the benefit in the 4th quintile is now greater than the benefit in the highest quintile.
Table 7 provides insight into why the spread of benefits differ across these series. In this table, individuals are divided into cells based on their location in the joint distribution of not-size-adjusted tax unit income and size-adjusted household income. Each cell contains the mean benefit from the health insurance tax exemption relative to both the population mean when using not-size-adjusted tax unit income and size-adjusted household income.
Among individuals along the diagonal where the quintile of both income distributions is the same, the ratio of mean quintile benefit to mean population benefit is similar for both series. This is not the case for individuals in quintiles off the diagonal (i.e., those who switch quintiles depending on which unit of measurement is employed). Individuals below the diagonal are in a higher quintile for household income than they are for tax unit income. This can occur, for instance, when low-income grown children live with their higher income parents. For these individuals, the relative mean benefits from tax-deductible employer provided health insurance observed for household units exceeds that observed for tax units. This is particularly evident among individuals in the bottom quintile of tax units, where almost no benefits are observed at the tax unit level but substantial benefits may be observed at the household level. Thus, by missing the benefits for these individuals, using tax units exacerbates the perceived disparity in the tax advantage of exempting employer provided health insurance across the income distribution. In contrast, above the diagonal there is less of a distinction between the two series as even individuals in the bottom quintiles of the household income distribution are still receiving benefits from the income exclusion. Therefore, an exclusive focus on the tax unit obscures the fact that within low income quintiles there is a range of benefits and some households at the bottom of the distribution do, in fact, receive substantial benefits from this policy.
Much of the previous research on income and its distribution is based on the types of income captured in the data and the sharing unit over which it was collected, without fully considering the implications of those choices. In this paper, we demonstrate that such choices can substantially change the view of how the average American has fared over the past three business cycles (1979-2007) and who benefits from public policy choices going forward. When using the most restrictive income definition–pre-tax, pre-transfer tax unit cash (market) income–the resources available to the middle class have stagnated over the past three business cycles. In contrast, once broadening the income definition to post-tax, post-transfer, size-adjusted household cash income, middle class Americans are found to have made substantial gains, and these increases are even larger when including non-cash income such as the ex-ante value of health insurance. Additionally, as we demonstrate using the example of the benefits of the tax exclusion of employer provided health insurance, these measurement decisions impact the extent to which we view policy benefits as skewed towards the top or bottom of the income distribution or view them as distributed more widely to individuals of all incomes.
So which income series is superior? This depends on the research question. For researchers interested in how middle class Americans are compensated for their time in the labor market, for example, it is more appropriate to use pre-tax, pre-transfer (market) income, although even here researchers who ignore the dramatic increase in the ex-ante value of employer health insurance will understate the returns to work in
In most cases, it is more important to know how a given policy impacts people arrayed by available resources within their sharing unit than how that policy impacts their market income. In such cases, researchers should broaden the definition of income to include taxes and non-cash benefits. This will more accurately reflect the total financial resources available to individuals. Additionally, they should do so across households rather than tax units and adjust for the number of people in those households. As we have demonstrated, doing so provides a markedly different picture of how middle class Americans have fared over the past several decades.
ACKNOWLEDGEMENTS AND DISCLAIMERS
Atkinson, Anthony B., and
Atkinson, Anthony B., Thomas Piketty, and
Auten, Gerald, and
Buhmann, Brigitte, <person>Lee Rainwater,
Burkhauser, Richard V., Shuaizhang Feng,
Burkhauser, Richard V., Shuaizhang Feng,
Burkhauser, Richard V., and
Burkhauser, Richard V., and
Chung, Wankyo, 2003. "Fringe Benefits and Inequality in the Labor Market." Economic Inquiry 41 (3), 517-529.
Cutler, David M., 2004. Your Money or Your Life: Strong Medicine for
Elmendorf, Douglas W.,
Feenberg, Daniel, and
Goldman, David, 2008. "Work Harder, Take Home Less." CNN.com,
Gordon, Robert J., 2009. "Misperception About the Magnitude and Timing of Changes in American Income Inequality." NBER Working Paper No. 15351.
Gottschalk, Peter, and
Jones, Arthur F., Jr., and
Karoly, Lynn A., 1994. "Trends in Income Inequality: The Impact of, and Implications for, Tax Policy." In Slemrod, Joel B. (ed.), Tax Progressivity and Income Inequality, 95-129.
Lahart, Justin, and
Leonhardt, David, 2008. "For Many a Boom That Wasn't."
Meyer, Bruce D., and
Meyer, Bruce D., and
Obama, Barack, 2009. Memorandum for the Heads of Executive Departments and Agencies:
Pierce, Brooks, 2001. "Compensation Inequality."
Pierce, Brooks, 2010. "Recent Trends in Compensation Inequality." In Abraham, Katharine G.,
Piketty, Thomas, and
Piketty, Thomas, and
Ruggles, Patricia, 1990. Drawing the Line: Alternative Poverty Measures and Their Implication for Public Policy.
Ryscavage Paul, 1995. "A Surge in Growing Income Inequality?" Monthly Labor Review 118 (8), 51-61.
Saez, Emmanuel, 2009. "Striking it Richer: The Evolution of Top Incomes in
Stewart, Kenneth J., and
(1) An alternative way to measure how middle class Americans have fared over time is to use panel data that actually follows the same individuals over time. Such longitudinal analyses are not possible with the March CPS data. For an example of this type of analysis, see Auten and Gee (2009).
(2) In this procedure, all individuals with income from an income source above the topcode threshold have their income replaced with the mean source-level income of all individuals who are topcoded on that income source. The incomes from all income sources (both the topcoded sources replaced with cell-mean values and the actual values for all other sources) are then combined to yield total personal income.
(3) A more detailed discussion of this issue as it relates to this paper is contained in a data appendix available upon request from the authors.
(4) Since the March CPS data do not have detailed information on mortgage interest, property taxes paid, charitable contributions, or other factors relevant for itemized deductions, this imputation likely understates the value of deductions for individuals not taking the standard deduction.
(5) Given that many students are dependents even if they file their own tax return, we replicated our analysis assuming that students age 20-24 were part of their parents' tax unit rather than independent tax units. These results, which are largely unchanged, are contained, along with a more detailed discussion of our methods for calculating them, in a data appendix available upon request from the authors.
(6) In the small number of cases where never-married individuals under age 20 live in a household without a parent or guardian, we assign them to the tax unit of the household's primary family or the oldest adult in the household when there is no primary family. Only households with no adults over age 20 are treated as their own tax unit.
(7) These data are available at "
(9) This is a common problem for researchers using
(10) For an example of shifting from the tax unit to the household as the sharing unit to measure the impact of tax changes on economic well-being, see Elmendorf et al. (2008). This procedure is embedded in the
(11) Dividing by the square root of household size is the most commonly used case of the economies of scale size-adjustments proposed by Buhmann et al. (1988) where size-adjusted income = total income / size([alpha]) with [alpha] = 1 implying no economies of scale (per capita income) and [alpha] = 0 implying infinite economies of scale (the implicit assumption of those who do not adjust for size). Dividing by the square root of household size ([alpha] = 0.5) closely matches the adjustments for household size implied by official
(12) Peak years of business cycles are defined based on peaks in median income, which generally lag macroeconomic growth. However, the results are not sensitive to reasonable adjustments to the choice of peak years used in the analysis.
(13) Burkhauser and Larrimore (2011) analyze the factors causing the declines in median pre-tax, post-transfer, size-adjusted household income over the first years of the last four recessions. To our knowledge, no researchers have considered how the observed impact of such factors differs across income definitions, sharing units, or size-adjustments, which may be a valuable area for future research.
(14) Our results do not exactly match the Census P-60 reports because the
(15) An alternate approach to evaluating the extent to which economic gains are captured by the middle class is to compare median income growth in the CPS to GDP growth, productivity growth for non-farm businesses, or mean per-capita income growth from the National Income and Products Accounts. For a flavor of this approach, and a discussion of some of its challenges, see Gordon (2009).
(16) A fuller measure of resources would also include most or all of these sources which are not collected by the March CPS and thus are not analyzed here. For this reason, it may be beneficial for the U.S. Census to more rigorously attempt to capture these income sources.
Table 1 Comparing the Total Percentage Growth from 1979-2007 Using Each Sharing Unit, Size-Adjustment, and Income Series Combination Size- Size Tax Adjusted Adjusted Unit Household Tax Unit Household Pre-tax, pre-transfer 3.2 12.5 14.5 20.6 Pre-tax, post-transfer 6.0 15.2 17.0 23.6 Post-tax, post-transfer 9.5 20.2 25.0 29.3 Post-tax, post-transfer + 18.2 27.3 33.0 36.7 health insurance Notes: Changes in income between 1992 and 1993 are suppressed and assumed to be zero given the trend-break resulting from the CPS redesign in those years; see main text for details. Health insurance information is not available prior to 1988. The rate of growth in the value of health insurance from 1979-1989 is assumed to match that of post-tax, post-transfer income. Source: Public Use March CPS data. Table 2 Trends in the Size of Tax Units and Households Panel A: Tax Units per Household Mean Tax Tax Units Households Units per (Thousand) (Thousand) Household 1979 98,958 79,399 1.25 1989 119,705 93,626 1.28 2000 137,810 106,512 1.29 2007 153,322 116,881 1.31 Panel B: Unrelated Tax Units per Household Mean Unrelated Unrelated Tax Tax Units Households Units per (Thousand) (Thousand) Household 1979 83,690 79,399 1.05 1989 100,606 93,626 1.07 2000 117,146 106,512 1.10 2007 128,751 116,881 1.10 Panel C: Individuals per Tax Unit Mean Individuals Tax Units Individuals (Thousand) (Thousand) per Tax Unit 1979 217,965 98,958 2.20 1989 243,886 119,705 2.04 2000 271,359 137,810 1.97 2007 292,895 153,322 1.91 Panel D: Individuals per Household Mean Individuals Households Individuals (Thousand) (Thousand) per Household 1979 217,965 79,399 2.75 1989 243,886 93,626 2.60 2000 271,359 106,512 2.55 2007 292,895 116,881 2.51 Panel A: Tax Units per Household Percent of Households with One, Two, or More Tax Units One Two [greater than or equal to] Three 1979 80.3 15.8 3.9 1989 78.4 16.8 4.8 2000 77.1 18.0 4.9 2007 76.2 18.2 5.6 Panel B: Unrelated Tax Units per Household Percent of Households with One, Two, or More Unrelated Tax Units One Two [greater than or equal to] Three 1979 95.3 4.2 0.5 1989 93.4 6.0 0.6 2000 91.2 7.9 0.9 2007 91.1 8.0 0.9 Panel C: Individuals per Tax Unit Percent of Tax Units with One, Two, or More Individuals One Two [greater than or equal to] Three 1979 36.3 28.6 35.2 1989 41.7 27.7 30.6 2000 45.0 26.9 28.1 2007 47.2 26.7 26.0 Panel D: Individuals per Household Percent of Households with One, Two, or More Individuals One Two [greater than or equal to] Three 1979 22.7 31.2 46.1 1989 24.8 32.2 43.0 2000 26.2 33.2 40.6 2007 27.6 33.2 39.3 Source: See Table 1. Table 3 Percent Growth in
Median Incomes Using Alternative Income Series PanelA: Total Median Income Growth (%) in Each Business Cycle Household Household Size- Size- Tax Unit Household Adjusted Adjusted Pre-Tax Pre-Tax Pre-Tax Post-tax Pre- Post- Post- Post- Transfer Transfer Transfer Transfer 1979-1989 0.2 6.6 9.2 12.0 1989-2000 9.1 9.3 13.4 14.4 2000-2007 -5.5 -1.2 -0.1 1.0 1979-2007 3.2 15.2 23.6 29.3 Panel B: Annualized Median Income Growth (%) in Each Business Cycle Household Household Size- Size- Tax Unit Household Adjusted Adjusted Pre-Tax Pre-Tax Pre-Tax Post-tax Pre- Post- Post- Post- Transfer Transfer Transfer Transfer 1979-1989 0.02 0.66 0.92 1.20 1989-2000 0.82 0.85 1.22 1.31 2000-2007 -0.79 -0.17 -0.02 0.14 1979-2007 0.12 0.54 0.84 1.05 A: Total Median Income Growth (%) in Each Business Cycle Household Size-Adjusted Post-Tax Post-Transfer + Health Insurance 1979-1989 12.0 (1) 1989-2000 16.6 2000-2007 4.8 1979-2007 36.7 (1) Panel B: Annualized Median Income Growth (%) in Each Business Cycle Household Size-Adjusted Post-Tax Post-Transfer + Health Insurance 1979-1989 1.20 (1) 1989-2000 1.51 2000-2007 0.68 1979-2007 1.31 (1) (1) Health insurance information is not available prior to 1988. The rate of growth in the value of health 3 insurance from 1979-1989 is assumed to match that of post-tax, post-transfer income. Source: See Table 1. Table 4 Percent Growth in Mean Quintile Income by Business Cycle Using Each Income Series Panel A: 1979-1989 Business Cycle Mean Income Growth (%), by Income Quintile Household Size- Tax Unit Household Adjusted Pre-Tax Pre-Tax Pre-Tax Pre- Post- Post- Transfer Transfer Transfer Bottom quintile -0.2 5.0 0.0 2nd quintile -5.0 0.2 -0.7 Middle quintile 0.0 6.3 9.1 4th quintile 4.0 9.6 12.9 Top quintile 17.6 19.7 23.4 Top 10 percent 21.8 23.0 19.7 Top 5 percent 25.6 26.3 27.2 1979 Gini 0.515 0.424 0.384 1989 Gini 0.547 0.451 0.423 Panel B: 1989-2000 Business Cycle Mean Income Growth (%), by Income Quintile Household Size- Tax Unit Household Adjusted Pre-Tax Pre-Tax Pre-Tax Pre- Post- Post- Transfer Transfer Transfer Bottom quintile 17.8 10.6 17.2 2nd quintile 10.8 8.3 12.6 Middle quintile 7.5 10.7 13.1 4th quintile 10.7 12.3 13.3 Top quintile 14.7 14.0 16.2 Top 10 percent 15.0 14.3 14.0 Top 5 percent 14.4 13.8 13.9 1989 Gini 0.547 0.451 0.423 2000 Gini 0.556 0.459 0.427 Panel C: 2000-2007 Business Cycle Mean Income Growth (%), by Income Quintile Household Size- Tax Unit Household Adjusted Pre-Tax Pre-Tax Pre-Tax Pre- Post- Post- Transfer Transfer Transfer Bottom quintile -43.0 -5.8 -0.2 2nd quintile -10.2 -3.9 -2.9 Middle quintile -4.9 -2.0 -0.4 4th quintile -2.5 -0.1 1.0 Top quintile -1.6 -1.4 -1.0 Top 10 percent -2.4 -2.4 -1.4 Top 5 percent -4.0 -4.0 -4.0 2000 Gini 0.556 0.459 0.427 2007 Gini 0.566 0.462 0.430 Panel D: 1979-2007 Mean Income Growth (%), by Income Quintile Household Size- Tax Unit Household Adjusted Pre-Tax Pre-Tax Pre-Tax Pre- Post- Post- Transfer Transfer Transfer Bottom quintile -33.0 9.5 9.9 2nd quintile -5.5 4.3 8.6 Middle quintile 2.2 15.3 22.8 4th quintile 12.3 23.0 29.2 Top quintile 32.7 34.6 42.0 Top 10 percent 36.7 37.3 34.6 Top 5 percent 37.9 38.0 39.1 1979 Gini 0.515 0.424 0.384 2007 Gini 0.566 0.462 0.430 Panel A: 1979-1989 Business Cycle Mean Income Growth (%), by Income Quintile Household Household Size- Size-Adjusted Adjusted Post-Tax Post-tax Post-Transfer Post- + Health Transfer Insurance (1) Bottom quintile 0.4 0.4 2nd quintile 1.0 1.0 Middle quintile 11.7 11.7 4th quintile 15.6 15.6 Top quintile 28.1 28.1 Top 10 percent 27.4 33.7 Top 5 percent 32.0 39.5 1979 Gini 0.349 0.330 1989 Gini 0.394 0.372 Panel B: 1989-2000 Business Cycle Mean Income Growth (%), by Income Quintile Household Household Size- Size-Adjusted Adjusted Post-Tax Post-tax Post-Transfer Post- + Health Transfer Insurance Bottom quintile 20.4 23.2 2nd quintile 15.2 18.2 Middle quintile 14.5 16.8 4th quintile 13.8 15.5 Top quintile 14.8 15.5 Top 10 percent 17.0 15.2 Top 5 percent 16.6 15.1 1989 Gini 0.394 0.372 2000 Gini 0.390 0.364 Panel C: 2000-2007 Business Cycle Mean Income Growth (%), by Income Quintile Household Household Size- Size-Adjusted Adjusted Post-Tax Post-tax Post-Transfer Post- + Health Transfer Insurance Bottom quintile -4.8 2.2 2nd quintile -1.2 4.7 Middle quintile 1.2 4.9 4th quintile 2.3 5.2 Top quintile 1.5 3.1 Top 10 percent -2.0 1.3 Top 5 percent -3.4 1.5 2000 Gini 0.390 0.364 2007 Gini 0.396 0.362 Panel D: 1979-2007 Mean Income Growth (%), by Income Quintile Household Household Size- Size-Adjusted Adjusted Post-Tax Post-tax Post-Transfer Post- + Health Transfer Insurance Bottom quintile 15.0 26.4 2nd quintile 15.0 25.0 Middle quintile 29.5 36.9 4th quintile 34.6 40.4 Top quintile 49.4 52.6 Top 10 percent 46.1 56.0 Top 5 percent 48.7 63.0 1979 Gini 0.349 0.330 2007 Gini 0.396 0.362 Note: See Table 1. (1) Health insurance information is not available prior to 1988. The rate of growth in the value of health insurance from 1979-1989 is assumed to match that of post-tax, post-transfer income. Source: See Table 1. Table 5 Comparing the Quintile Distributions of the Size-Adjusted Household Income Distribution and Not Size-Adjusted Tax Unit Income Distribution (2007) Quintile of Quintile of Not Size-Adjusted Tax Unit Income Size-Adjusted Household Income Bottom 2nd Middle 4th Top Total Bottom 11.5 7.1 1.4 0.0 0.0 20 2nd 3.5 6.8 7.9 1.9 0.0 20 Middle 2.2 3.3 6.5 7.4 0.6 20 4th 1.7 1.7 3.1 7.7 5.8 20 Top 1.2 1.1 1.2 3.1 13.5 20 Total 20 20 20 20 20 100 Notes: In both series the unit of analysis is the individual so each quintile contains 20 percent of individuals in the population and income is measured using post-tax, post-transfer income including the ex-ante value of health insurance benefits. Source: See Table 1. Table 6 Relative Benefit of Health Insurance Tax Exclusion by Quintile of the Distribution in Each Income Series in 2007 (Population Mean Benefit Normalized to 100 in Each Series) Relative Health Relative Health Insurance Tax Insurance Tax Quintile of Tax Exclusion Benefit, Exclusion Benefit, Unit Income Not Size-Adjusted Size-Adjusted Bottom 1.17 18.08 2nd 25.55 79.27 Middle 101.95 116.93 4th 177.12 144.85 Top 194.22 140.88 Note: In both series the unit of analysis is the individual so each quintile contains 20 percent of individuals in the population and income is measured using post-tax, post-transfer income including the ex-ante value of health insurance benefits. Source: See Table 1. Table 7 Comparing Relative Benefits of Health Insurance Tax Exclusion by the Joint Quintile of the Size-Adjusted Household Income and Not Size-Adjusted Tax Unit Income Distributions in 2007 (Population Mean Benefit is Normalized to 100 in Each Series) Quintile of Size-Adjusted Quintile of (Not Size-Adjusted) Tax Unit Income Household Income Bottom 2nd Middle 4th Bottom HH: 3.2 HH: 31.1 HH: 76 HH: N/A TU: 1 TU: 35.7 TU: 104.1 TU: N/A 2nd HH: 25.5 HH: 31.7 HH: 124.5 HH: 162 TU: 1.3 TU: 28.4 TU: 145.6 TU: 226.4 Middle HH: 63.1 HH: 21.7 HH:107.4 HH: 181.5 TU: 1.6 TU: 10.4 TU: 98.8 TU: 220.7 4th HH: 108.9 HH: 61.4 HH: 36.7 HH: 175.7 TU: 1.6 TU: 11.8 TU: 28.7 TU: 170.1 Top HH: 128.7 HH: 84.1 HH: 53.1 HH: 74.1 TU: 0.8 TU: 8.9 TU: 16.8 TU: 60.5 All HH: 27.1 HH: 62.2 HH: 113.2 HH: 144.6 TU: 1.2 TU: 25.6 TU: 101.9 TU: 177.1 Quintile of Quintile of (Not Size- Size-Adjusted Adjusted) Tax Unit Income Household Income Top All Bottom HH: N/A HH: 18.1 TU: N/A TU: 20.3 2nd HH: 119.8 HH: 79.3 TU: 225,9 TU: 88.3 Middle HH: 154.8 HH: 116.9 TU: 222.2 TU: 122 4th HH: 195.4 HH: 144.9 TU: 239.9 TU: 141.4 Top HH: 169 HH: 140.9 TU: 173.3 TU: 128 All HH: 153 TU: 194.2 Note: HH is the ratio of the mean benefit to size-adjusted household income in the joint quintile to the mean benefit to size-adjusted household income for the population. TU is the ratio of the mean benefit to not-size-adjusted tax unit income in the joint quintile to the mean benefit to not-size-adjusted tax unit income for the population. In both series the unit of analysis is the individual so each quintile contains 20 percent of individuals in the population. Source: See Table 1.
|Copyright:||(c) 2012 National Tax Association|