A Guide to Statistics on Historical Trends in Income Inequality
May 15, 2018
The broad facts of income inequality over the past six decades are easily summarized:
- The years from the end of World War II into the 1970s were ones of substantial economic growth and broadly shared prosperity.
- Incomes grew rapidly and at roughly the same rate up and down the income ladder, roughly doubling in inflation-adjusted terms between the late 1940s and early 1970s.
- The income gap between those high up the income ladder and those on the middle and lower rungs — while substantial — did not change much during this period.
- Beginning in the 1970s, economic growth slowed and the income gap widened.
- Income growth for households in the middle and lower parts of the distribution slowed sharply, while incomes at the top continued to grow strongly.
- The concentration of income at the very top of the distribution rose to levels last seen more than 80 years ago (during the “Roaring Twenties”).
- Wealth — the value of a household’s property and financial assets, minus the value of its debts — is much more highly concentrated than income. The best survey data show that the share of wealth held by the top 1 percent rose from just under 30 percent in 1989 to nearly 49 percent in 2016, while the share held by the bottom 90 percent fell from just over 33 percent to less than 23 percent over the same period. Put another way, the top 1 percent now have more than twice as large a share of the nation’s wealth as the bottom 90 percent.
Data from a variety of sources contribute to this broad picture of strong growth and shared prosperity for the early postwar period, followed by slower growth and growing inequality since the 1970s. Within these broad trends, however, different data tell slightly different parts of the story (and no single source of data is better for all purposes than the others).
This guide consists of four sections. The first describes the commonly used sources and statistics on income and discusses their relative strengths and limitations in understanding trends in income and inequality. The second provides an overview of the trends revealed in those key data sources. The third and fourth sections supply additional information on wealth, which complements the income data as a measure of how the most well-off Americans are doing, and poverty, which measures how the least well-off Americans are doing.
I. The Census Survey and IRS Income Data
The most widely used sources of data and statistics on household income and its distribution are the annual survey of households conducted as part of the Census Bureau’s Current Population Survey (CPS) and the Internal Revenue Service’s (IRS) Statistics of Income (SOI) data compiled from a large sample of individual income tax returns. The Census Bureau publishes annual reports on income, poverty, and health insurance coverage in the United States based on the CPS data, and the IRS publishes an annual report on individual income tax returns based on the SOI. While the Federal Reserve also collects income data in its triennial Survey of Consumer Finances (SCF), the SCF is more valuable as the best source of survey data on wealth.
Each agency produces its own tables and statistics and makes a public-use file of the underlying data available to other researchers. In addition, the Congressional Budget Office (CBO) has developed a model that combines CPS and SOI data to estimate household income both before and after taxes, as well as average taxes paid by income group back to 1979. Economists Thomas Piketty and Emmanuel Saez have used SOI data to construct estimates of the concentration of income at the top of the distribution back to 1913. More recently, they have expanded that work to examine trends in wealth concentration and to incorporate the portion of national income not captured in the tax or survey data into the analysis of income inequality. CBO and Piketty-Saez regularly release reports incorporating the latest available data.
Concepts of Income Measured in Census and IRS Data
Census Money Income
The Census Bureau bases its report on income and poverty on a sample of about 95,000 addresses conducted through the Annual Social and Economic Supplement (ASEC) to the monthly Current Population Survey, which is the primary source of data for estimating the unemployment rate and other household employment statistics. The ASEC, also called the March CPS, provides information about the total annual resources available to families — including income from earnings, dividends, and cash benefits (such as Social Security), as well as the value of tax credits such as the Earned Income Tax Credit (EITC) and non-cash benefits such as nutritional assistance, Medicare, Medicaid, public housing, and employer-provided fringe benefits.
The income measure used in the Census report is money income before taxes, and the unit of analysis is the household. The latest data, for 2016, were released in September 2017. The statistics on household income are available going back to 1967. Census has statistics on family income that go back to 1947, but because Census defines a “family” as two or more people living in a household who are related by birth, marriage, or adoption, those statistics exclude people who live alone or with others to whom they are not related.
Census’s standard income statistics do not adjust for the size and composition of households. Two households with $40,000 of income rank at the same place on the distributional ladder, even if one is a couple with two children and one is a single individual. An alternative preferred by many analysts is to make an equivalence adjustment based on household size and composition so that the adjusted income of a single person with a $40,000 income is larger than the adjusted income of a family of four with the same income. Equivalence adjustment takes into account the fact that larger families need more total income but less per capita income than smaller families because they can share resources and take advantage of economies of scale. In recent reports, Census has supplemented its measures of income inequality based on household money income with estimates based on equivalence-adjusted income.
For reasons having to do with small sample size, data reporting and processing restrictions, and confidentiality considerations, Census provides more limited information about incomes at the very top of the income distribution than it does for incomes elsewhere in the distribution. For example, Census does not collect information about earnings over $1,099,999 for any given job; earnings above that level are recorded in Census data as $1,099,999.
Income Tax Data
The income tax data used in distributional analysis come from a large sample of tax returns compiled by the IRS’s Statistics of Income Division. For 2015, the sample consisted of about 339,000 returns scientifically selected from the roughly 150 million returns filed that year. For the population that files tax returns and for the categories of income that get reported, these administrative data are generally more accurate and more complete than survey data, such as the CPS, which is prone to underreporting of some kinds of income.
However, not all people are required to file tax returns, and tax returns do not reflect all sources of income. Those who do not file returns are likely to have limited incomes; hence tax data do not provide a representative view of low-income households (the mirror image of inadequate coverage of high-income households in the CPS). Like Census money income, income reported on tax returns excludes non-cash benefits such as food stamps, housing subsidies, Medicare, Medicaid, and non-taxable employer-provided fringe benefits.
The exclusion of non-filers is a major limitation of the tax data for distributional analysis. A further complication is that the data are available only for “tax-filing units,” not by household or family (members of the same family or household may file separate tax returns).
SOI tax data are also less timely than Census data. Final statistics for tax year 2015 were released in September 2017.
Key Historical Series Constructed from Census and IRS Data
CBO’s Distribution of Household Income
CBO produces annual estimates of the distribution of household income and taxes that combine information from the CPS and the SOI. Thus, these estimates have relatively detailed information about very high-income households and taxes paid (the strengths of the SOI) and about low-income households and income and non-cash benefits (the strengths of the CPS). Accordingly, CBO’s measure of household income includes more sources of income than either CPS- or SOI-based measures alone.
In producing its estimates, however, CBO must make judgments about how to value non-cash benefits like government-provided health insurance, and in presenting its estimates CBO must make judgments about what specific measures of income to feature, e.g., income before taxes and government assistance (transfers) versus income after taxes and transfers. As discussed below, these methodological choices affect how CBO’s estimates of trends in household income compare with other estimates.
CBO’s methodology for analyzing the distribution of household income and taxes changed little between 2001 and 2012. The primary measure CBO used to rank households and calculate average federal tax rates was a broad measure of “before-tax income” that included both “market income” and a broad set of government transfers. The latter included both social insurance benefits (Social Security, Medicare, unemployment insurance, and workers’ compensation) and means-tested transfers, both cash and in-kind, such as Medicaid and Children’s Health Insurance Program (CHIP) benefits, Supplemental Nutrition Assistance Program (SNAP, formerly food stamps) benefits, and Temporary Assistance for Needy Families (TANF) cash assistance. “After-tax income” was this “before-tax income” minus federal individual and corporate income, payroll (social insurance), and excise taxes.
In 2012, CBO changed the way it valued government-provided health insurance such as Medicare and Medicaid. Whereas CBO’s previous method sought to measure the extent to which this coverage frees up income that a household can then use to meet basic food or housing expenses, CBO’s revised method reflects the government’s average cost of providing health insurance to the household. This is the same approach CBO has always used to value employer-provided health insurance benefits. The new method increases before-tax income (as CBO measures it) for low-income households with limited cash income, but it does not increase their ability to meet basic needs. Because medical benefits make up a sizeable and growing share of income in CBO’s series, CBO’s treatment of government provided health insurance can lead to a difference between trends in CBO’s income data, which include these benefits, and trends in other income series that do not include these benefits.
In 2018, CBO made another substantial change to its methodology, adopting a new measure — “income before transfers and taxes” — to rank households and calculate effective tax rates. Broadly speaking, this new measure consists of market income plus social insurance benefits, such as Social Security and Medicare. More specifically, it includes all cash income (including non-taxable income not reported on tax returns, such as child support), taxes paid by businesses, employees’ contributions to 401(k) retirement plans, and the estimated value of in-kind income such as Medicare and employer-paid health insurance premiums). One effect of the change appears to be to shift more seniors with substantial Medicaid benefits into the bottom fifth of the income distribution.
As part of this 2018 revision, CBO created a second new measure, “income after transfers and taxes.” It consists of the former “after-tax income” plus means-tested transfers, such as Medicaid and SNAP.
As it has since 2001, CBO makes a simple equivalence adjustment based on household size to determine each person’s household income for ranking purposes: each household’s income is divided by the square root of the number of people in the household. Thus, the adjusted household income of a single person with $20,000 of income is equivalent to that of a household of four with $40,000.
With the 2018 changes, CBO’s distributional tables now rank people by their adjusted household income before transfers and taxes and construct five income groups (quintiles), each containing roughly an equal number of people. The quintiles contain slightly different numbers of households, depending on the average household size at different points in the income distribution. CBO states that the former method of using after-tax income for ranking was appropriate for analyzing the effects of federal taxes, but with the growing importance of means-tested transfers, the change allows the agency to analyze both means-tested transfers and taxes on the same basis. Together with the 2012 change in the treatment of government-provided health insurance, however, this change appears to strongly affect income trends for the poorest households, as discussed in Section II.
The latest CBO report on the distribution of household income, released in March 2018, includes data for 1979-2014 on income before and after transfers and taxes as well as taxes paid for each quintile and for the top 1, 5, and 10 percent of households.  Because of the effort involved in preparing these analyses, CBO’s annual updates tend to lag behind other sources of income data, often by a couple of years.
Piketty-Saez Data on Income Concentration
Economists Thomas Piketty and Emmanuel Saez have constructed income statistics based on IRS data that go back to 1913 to provide a long-term perspective on trends in the concentration of income within the top 10 percent of the distribution.
Because they have no direct data on non-filers and because in any year only about 10 to 15 percent of potential tax units had to file an income tax return prior to World War II, Piketty and Saez focus on the share of income received at the top of the distribution.
Their income concept is market income before individual income taxes. They define market income as the sum of all income sources reported on tax returns (including realized capital gains and taxable unemployment compensation). Other non-taxable non-cash income sources, such as nutrition assistance and employer-provided health care benefits, are not included.
Some people with market income are not required to file income tax returns; hence they do not show up in the population of tax filers, and their income does not show up in the total income reported on tax returns. Piketty and Saez address these omissions by estimating the number of non-filers and their income and adding these to the population of tax filers and the market income calculated from the income tax data. They compute total income as all market income reported on tax returns plus their estimate of market income for non-filers. The top 10 percent, top 1 percent, etc. are defined with respect to this total income and to the population of potential tax units (filers plus non-filers). Piketty and Saez do not make an adjustment for family size in their analysis.
The primary advantage of the Piketty-Saez data is that they provide the longest historical series of annual data on income at the top of the distribution. The key limitation is that they are based exclusively on tax return data. As a result, they do not include data for individual non-filers (and therefore provide no information about the distribution of income among non-filers). They also don’t account for government cash transfers or for public and private non-cash benefits (such as government health and nutrition assistance benefits and employer-paid health insurance benefits).
The share of personal income coming from the public and private non-cash benefits that are missing from the Piketty-Saez income measure has increased over the years. As a result, total income as computed by Piketty and Saez has accounted for a decreasing share of personal income in the national income and product accounts over time. This could distort their estimates of what share of the growth of total income has come at the top of the distribution. For example, employer-sponsored health insurance benefits are most likely a much smaller fraction of income for the top 1 percent than for the vast majority of middle-income tax units; not including them could understate income growth in the middle of the distribution relative to growth at the top.
II. Broad Trends in Income Inequality
Because each individual source of readily available data on income distribution has different advantages and limitations, no single source illustrates all of the major trends in inequality over the past six decades or so. Ideally, we would look at a comprehensive measure of income that covers a long time span, allows us to compare before- and after-tax income at different points in the income distribution, and accounts for changes in the size and composition of households. CBO data satisfy many of these criteria but only go back to 1979 and are sensitive to particular methodological choices; the historical Census family income data series and Piketty-Saez income concentration data cover a longer time span but use less-comprehensive measures of income and do not adjust for changes in the size and composition of households.
The Loss of Shared Prosperity
Census family income data show that from the late 1940s to the early 1970s, incomes across the income distribution grew at nearly the same pace. Figure 1 shows the level of real (inflation-adjusted) income at several points on the distribution relative to its 1973 level. It shows that real family income roughly doubled from the late 1940s to the early 1970s at the 95th percentile (the level of income separating the 5 percent of families with the highest income from the remaining 95 percent), at the median (the level of income separating the richer half of families from the poorer half), and at the 20th percentile (the level of income separating the poorest fifth of families from the remaining 80 percent). Then, beginning in the 1970s, income disparities began to widen, with income growing much faster at the top of the ladder than in the middle or bottom.
While the Census family income data are useful for illustrating that the widening of income inequality began in the 1970s, other data are superior for assessing more recent trends.
Widening Inequality Since the 1970s
Census family income data show that the era of shared prosperity ended in the 1970s and illustrate the divergence in income that has emerged since that time. CBO data allow us to look at what has happened to comprehensive income measures since 1979 — both before and after taxes — and offer a better view of what has happened at the top of the distribution.
As Figure 2 shows, from 1979 to 2007 (just before the financial crisis and Great Recession), average income after transfers and taxes quadrupled for the top 1 percent of the distribution. The increases were much smaller for the middle 60 percent and bottom 20 percent of the distribution.
The CBO data also show income growth for the bottom 20 percent over this period that’s comparable to the 81st through 99th percentiles and substantially greater than the middle 60 percent. But this appears to be a methodological anomaly. In CBO analyses published before the agency’s 2012 change in its method for valuing government-provided health insurance, incomes grew more slowly in the bottom 20 percent than in the middle 60 percent from 1979 to 2011. CBO’s 2018 change in the income measure it uses to rank households primarily affects which households are in the bottom versus the next-to-the-bottom 20 percent. Although the new methodology does a better job of highlighting the growing importance of transfers and taxes for low earners (including seniors), it may well substantially exaggerate the rise in low-income households’ true standard of living.
After-tax incomes fell sharply at the top of the distribution in 2008 and 2009 but have since partially recovered. The up-and-down pattern in 2012-13 may reflect, in part, decisions by wealthy taxpayers to sell assets in 2012 that had increased in value since they were first purchased in order to pay taxes on those capital gains before income tax rates increased in 2013. The Piketty-Saez data discussed below, which go through 2015, show a generally upward trend since 2009 that is consistent with this explanation.
Although the average after-tax income of the top 1 percent of households remains well below its 2007 peak, the percentage increase in their average after-tax income from 1979 to 2014 was more than five times larger than that of the middle 60 percent and more than three times larger than that of the bottom fifth. (See Table 1.) Moreover, CBO’s latest baseline assumptions predict earnings to grow faster for high-income earners than for others in the next decade, suggesting that the Great Recession and financial crisis may have had only a temporary impact on the rising trend of income gains at the top, much as the impact of the dot-com collapse in the early 2000s was only temporary.
|Change in CBO Comprehensive Income by Income Group and Time Period|
|Before transfers and taxes||40%||32%||67%||274%|
|After transfers and taxes||58%||42%||72%||314%|
|Before transfers and taxes||26%||28%||69%||221%|
|After transfers and taxes||69%||42%||73%||228%|
Trends in income before transfers and taxes look very similar. Because average tax rates have fallen for all income groups since 1979, income before transfers and taxes grew somewhat more slowly larger income after transfers and taxes from 1979 to 2014. (See the box for more on the effect of transfers and taxes on income.)
Federal Transfers and Taxes Are Progressive
But Income Is Highly Concentrated Both Before- and After- Transfers and Taxes
The chart below shows that federal transfers and taxes are progressive. In 2014, the top 20 percent of households had a smaller share of total income after transfers and taxes than before transfers and taxes, while the opposite is true for the other 80 percent of the distribution.
Income is highly concentrated under either measure, however. In 2014, the top 1 percent of households received 17 percent of income before transfers and taxes and 13 percent of income after federal transfers and taxes; the comparable figures for the bottom 80 percent of households were 46 and 53 percent, respectively.
As CBO’s latest analysis of trends in income distribution from 1979 to 2014 shows, both federal transfers and federal taxes reduce income inequality, but the reduction due to transfers is considerably larger.
Income Concentration Has Returned to 1920s Levels
The Piketty-Saez data put the increasing concentration of income at the top of the distribution into a longer-term historical context. As Figure 3 shows, the top 1 percent’s share of income before transfers and taxes has been rising since the late 1970s, and in the past decade has climbed to levels not seen since the 1920s. The vast majority of the increase occurred among the top 0.5 percent of households.
The increase in income concentration since the 1970s reversed the prior, long-term downward trend. After peaking in 1928, the share of income held by households at the very top of the income ladder declined through the 1930s and 1940s. Consistent with the shared prosperity found in the Census data on average family income, the share of income received by those at the very top changed little over the 1950s, 1960s, and early 1970s. The sharp rise in income concentration at the top of the distribution since the late 1970s was interrupted briefly by the dot-com collapse in the early 2000s and again in 2008 with the onset of the financial crisis and deep recession.
Top incomes generally have been on the rise since 2009. The Piketty-Saez data show the same up-and-down pattern in 2012-13 as CBO’s, but the additional data for 2014 and 2015 show the rise in top income share continuing.
III. The Distribution of Wealth
A family’s income is the flow of money coming in over the course of a year. Its wealth (sometimes referred to as “net worth”) is the total stock of assets it has as a result of inheritance and saving, less any liabilities. Wealth is much more highly concentrated than income, and concentration at the top has risen since the 1980s.
The main source of data for the distribution of household wealth is the Federal Reserve’s Survey of Consumer Finances (SCF), which is conducted every three years. SCF data go back to 1983; the latest published data are for 2016. The SCF is based on a sample of about 6,300 families. The data sources discussed in the preceding sections on income distribution are superior to the SCF for measuring income distribution, but none of those sources has comparable data for looking at the distribution of wealth.
The Federal Reserve publishes detailed statistics on wealth and income based on the SCF. Figure 4 shows the distribution of income and wealth in 2016, based on the SCF data. As the chart illustrates, wealth is much more concentrated than income. It should be noted that while there is considerable overlap, the top 1 percent of the income distribution does not contain the identical group of people as the top 1 percent of the wealth distribution. The SCF data show that the top 1 percent of the income distribution received roughly a quarter of all income in 2016, while the top 1 percent of the wealth distribution held nearly two-fifths of all wealth. Similarly, the top 10 percent of the income distribution received a little more than half of all income, while the top 10 percent of the wealth distribution held more than three-quarters of all wealth.
SCF data show rising concentration of wealth for the top 1 percent, little change for the rest of the top 10 percent, and a declining share for the bottom 90 percent. In particular, the share of wealth held by the top 1 percent rose from just under 30 percent in 1989 to 38.6 percent in 2016, while the share held by the bottom 90 percent fell from 33.2 percent in 1989 to 22.8 percent in 2016.
While the SCF is invaluable, it has its limitations, especially for detecting trends in wealth concentration at the very top. Recently, Emmanuel Saez and Gabriel Zucman have used tax-return information on income derived from wealth to infer the underlying distribution of wealth over time. Figure 5 shows Saez and Zucman’s estimates of the share of wealth held by the top 1 percent and top 0.5 percent since 1913. As with income, these data show a long historical decline in the concentration of wealth from the late 1920s into the late 1970s. Concentration at the top has increased markedly since then, driven by a rising share of wealth at the very top.
The Official Poverty Measure
The official U.S. poverty measure was developed in the 1960s. The Census Bureau uses money income (as described above) to determine a person’s poverty status. Each family or unrelated individual in the population is assigned a money income threshold based on the size of his or her family and age of its members. A person is defined as living in poverty if his or her family income is below the threshold for that family size and composition (the threshold for a couple with two children was $24,339 in 2016). The poverty thresholds are adjusted each year to reflect changes in the consumer price index. The poverty rate is the percentage of people living in poverty.
The official poverty statistics show a sharp decline in the poverty rate between 1959 and 1969 but little real change since then, apart from fluctuations due to the business cycle. For a number of reasons, however, the official measure is an unreliable guide to trends in poverty since 1970 and significantly understates progress in reducing poverty since then. The official poverty measure is based on Census money income, which includes cash assistance but does not count non-cash assistance like SNAP (formerly known as food stamps) and rental vouchers. The official poverty measure also omits the impact of the tax system, including tax credits for working families like the EITC and Child Tax Credit (CTC).
Alternatives to the Official Poverty Measure
Over the years, researchers have raised a number of serious conceptual and measurement concerns about how the official poverty rate is calculated. Following the publication of an important National Academy of Sciences (NAS) report on poverty measurement in 1995, the Census Bureau and the Bureau of Labor Statistics (BLS) explored a number of experimental measures reflecting NAS recommendations. NAS-based measures use a more complete definition of income that includes the value of non-cash benefits and tax credits while subtracting taxes and certain expenses. The NAS also recommended using a modernized poverty line that varies with local housing costs.
Census, with support from BLS, unveiled the newest refinement of the NAS-based measures, called the Supplemental Poverty Measure (SPM), in November 2011. This measure reflects recommendations from a federal interagency technical working group that drew on the NAS report and subsequent research. The Census SPM is available from 2009 to 2016. Unlike the official measure, which counts only a family’s cash income, the SPM counts non-cash benefits (SNAP, housing assistance, WIC, school lunch, and home energy assistance) and tax credits (the EITC and CTC) as income and subtracts various expenses, namely federal and state income and payroll taxes, child care and other work expenses, out-of-pocket medical expenditures, and child support paid. In addition, it updates the poverty line each year based on Americans’ shifting patterns of spending on basic needs, and it varies the poverty line based on local housing costs and the family’s type of housing (such as renters versus owners with a mortgage). Unlike in the official poverty measure (and most previous implementations of the NAS measure), unmarried partners are counted in the same SPM family.
Long-Term Poverty Trends
Since non-cash and tax-based benefits constitute a much larger part of government assistance than 50 years ago, the official poverty measure’s exclusion of these benefits masks progress in reducing poverty. Trying to compare poverty in the 1960s to poverty today using the official measure yields misleading results; it implies that programs like SNAP, the EITC, and rental vouchers — all of which were either small in the 1960s or didn’t yet exist — have no effect in reducing poverty, which clearly is not the case.
While the federal government has only calculated the SPM back to 2009, Columbia University researchers have estimated the SPM from 1967 to 2012. We have updated their estimates through 2016, and found that government economic security programs are responsible for a decline in the poverty rate from 25 percent in 1967 to 14 percent in 2016, based on an “anchored” version of the SPM that uses a poverty line tied to what American families spent on basic necessities in 2012 adjusted back for inflation. (See Figure 6.) Without government assistance, poverty would have been about the same in 2016 as in 1967 under this measure, which indicates the strong and growing role of antipoverty policies. Also, the child poverty rate fell to a record low in 2016 based on the SPM, largely due to increasingly effective government assistance policies. These findings underscore the importance of using the SPM rather than the official poverty measure when evaluating long-term trends in poverty.
Effectiveness of Economic Security Programs Against Poverty
Economic security programs cut poverty nearly in half in 2016. These programs (i.e., the safety net of government assistance policies) lifted 36 million people, including 7 million children, above the poverty line and reduced the poverty rate from 25.3 percent to 14.0 percent, according to CBPP’s analysis of SPM data.  (See Figure 7.)
Poverty also rose much less in the Great Recession when measured by the SPM rather than the official rate. Between 2007 (the year before the recession) and 2010 (the year after the recession), the anchored SPM rose from 14.8 percent to 15.3 percent, a rise (in unrounded data) of about 0.6 percentage points. This increase was one-fifth the size of the rise in the official poverty rate, which went from 12.5 percent to 15.1 percent (a rise of 2.6 percentage points) over the same period. The smaller increase under the SPM largely reflects the wider range of economic security programs included in the SPM and their success in keeping more Americans from falling into poverty during the recession.
Measuring “deep” poverty, often defined as income below half of the poverty line, poses particular challenges due to underreporting of certain benefits, reflecting respondents’ forgetfulness, embarrassment about receiving benefits, or other reasons. Census’s counts of program participants typically fall well short of the totals shown in actual administrative records. Such underreporting is common in household surveys and can affect estimates of poverty and, in particular, deep poverty because people who underreport their benefits naturally make up a larger share of those with the lowest reported incomes. (While respondents may also underreport earned income, the net rate of underreporting in the CPS is thought to be much lower for earnings than for benefits.)
In an analysis that corrects for underreporting of Temporary Assistance for Needy Families (TANF), SNAP, and Supplemental Security Income (SSI) benefits and uses a comprehensive NAS-based poverty measure similar to the SPM, CBPP analysts find that starting in the mid-1990s — when policymakers made major changes in the public assistance system — the share of children living in poverty fell but the share living in deep poverty rose, from 2.1 percent in 1995 to 3.0 percent in 2005.
Notably, uncorrected CPS figures — whether using the official poverty definition or CBPP’s broader NAS measure — do not show this rise in deep child poverty. By the official measure, the share of children below half the poverty line fell from 1995 to 2005, from 8.5 percent to 7.7 percent. When counting non-cash benefits and taxes but not correcting for underreporting, the figures are essentially flat, at 4.9 percent in 1995 and 4.7 percent in 2005. Only the corrected figures show the increase. (See Figure 8.)
The increase in deep poverty for children was largely due to means-tested benefits becoming less effective at shielding children from deep poverty. Over the 1995-2005 period, TANF cash assistance programs served a shrinking share of very poor families with children.
From 2005 to 2010, by contrast, the children’s deep poverty rate fell from 3.0 percent to 2.6 percent after correcting for underreporting. (See Figure 9.) The decline, occurring despite the Great Recession, shows the striking effectiveness of economic security programs during this period, when policymakers supplemented programs’ built-in responsiveness through recovery policies such as expansions in tax credits and SNAP and temporary measures such as the Making Work Pay tax credit.
 The authors would like to acknowledge the contributions of Hannah Shaw, who helped create this guide and was one of the original authors, as well as William Chen, Brandon DeBot, and Emily Horton.
 Internal Revenue Service, “SOI Tax Stats — Individual Income Tax Returns Publication 1304,” multiple years available, https://www.irs.gov/uac/soi-tax-stats-individual-income-tax-returns-publication-1304-complete-report
 Jesse Bricker et al., “Changes in U.S. Family Finances from 2013 to 2016: Evidence from the Survey of Consumer Finances,” Federal Reserve Bulletin, vol. 103, no. 3, September 2017, https://www.federalreserve.gov/publications/files/scf17.pdf.
 Emmanuel Saez, “Striking it Richer: The Evaluation of Top Incomes in the United States,” University of California, June 30, 2016, https://eml.berkeley.edu/~saez/saez-UStopincomes-2015.pdf.
 See Emmanuel Saez and Gabriel Zucman, “Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data,” Quarterly Journal of Economics, Vol. 131, No. 2, May 2016, http://eml.berkeley.edu/~saez/SaezZucman2016QJE.pdf; and Thomas Piketty, Emmanuel Saez, and Gabriel Zucman, “Distributional National Accounts: Methods and Estimates for the United States,” Quarterly Journal of Economics, Vol. 133, No. 2, May 2018, http://gabriel-zucman.eu/files/PSZ2018QJE.pdf.
For a discussion of distributional analyses and frameworks currently in use, see Kevin Perese, “CBO’s New Framework for Analyzing the Effects of Means-Tested Transfers and Federal Taxes on the Distribution of Household Income,” Congressional Budget Office, December 2017, pp. 41-45, https://www.cbo.gov/system/files/115th-congress-2017-2018/workingpaper/53345-workingpaper.pdf.
 About 70,000 households responded and were actually in the sample used to compute income statistics for 2016.
 Census also collects data on income, poverty, and health insurance coverage through the American Community Survey (ACS), which has replaced the long-form decennial census questionnaire. For its more limited set of categories, the ACS provides better data at the state and local level than the CPS, but Census advises that the CPS data provide the best annual estimates of income, poverty, and health insurance coverage for the nation as a whole.
 Examples of money income — sometimes referred to as “cash income” — include: wages and salaries; income from dividends; earnings from self-employment; rental income; child support and alimony payments; Social Security, disability, and unemployment benefits; cash welfare assistance; and pensions and other retirement income. Census money income does not include non-cash benefits such as those from the Supplemental Nutrition Assistance Program (food stamps), Medicare, Medicaid, or employer-provided health insurance.
 Census uses a three-parameter scale for equivalence adjustment that takes into account family size and composition (so that, for example, a two-adult, one-child family has a different adjustment than a one-adult, two-child family).
 This is generally referred to as “top-coding” and is done to preserve confidentiality. In addition, earnings well below this limit are suppressed and replaced with group average values in the public-use data files of the ASEC made available to researchers.
 “Market income” is labor income (wages, salaries, benefits, and the employer’s share of payroll taxes), business income (net income from business and farms owned solely by their owners, partnership income, and income from S corporations), realized capital gains, other capital income (dividends, rental income, and imputed corporate income taxes), income received in retirement for past services, and income from other sources. Note that this definition of “market income” differs from the market income concept used in the Piketty-Saez analysis discussed in the next section (see footnote 25).
 See definitions of social insurance benefits and means-tested programs in CBO (2018), pp. 37-39.
 Prior to 2012, CBO valued government-provided health insurance on the basis of the Census Bureau’s “fungible value” estimates, which essentially cap the value at the amount that a household could afford to pay for insurance. The cap is set at the amount by which the household’s income exceeds what it needs to meet basic food and housing expenses. See Congressional Budget Office, “The Distribution of Household Income and Federal Taxes, 2008-2009,” July 10, 2012, https://www.cbo.gov/publication/43373.
For low-income households, the fungible value of government-provided health insurance can be substantially less than the average cost to the government of providing it. Consider a household with $5,500 in income above what it needs to meet basic food and housing expenses. If government-provided health insurance for this type of household costs an average of $10,000, CBO would value the benefit at the full $10,000 under its current approach but at $5,500 under the prior approach, since that is all that the household could afford to spend on insurance in the absence of government-provided insurance. See the supplemental data accompanying Congressional Budget Office, “The Distribution of Household Income and Federal Taxes, 2014,” March 19, 2018, https://www.cbo.gov/publication/53597.https://www.cbo.gov/publication/53597.
 Changes in the nature of health care spending also could affect measured income differently than they affect household well-being. For example, advances in medical technology could enhance the value to households of health care spending in ways that the income data would not fully capture. Alternately, increases in spending on wasteful medical procedures or larger profit margins in the medical, insurance, or prescription drug industries could result in increases in health care spending that CBO counts as added income but do not enhance recipients’ well-being.
 CBO’s estimates of household income before transfers and taxes include the imputed value of taxes paid by businesses because CBO assumes that businesses would pay equivalently higher wages in the absence of those taxes.
 CBO does not subtract other federal taxes (such as estate and gift taxes) or state and local taxes when calculating income after transfers and taxes. Also, it should be noted that for some low-income households, CBO’s estimated income after transfers and taxes is higher than their estimated income before transfers and taxes due to refundable tax credits.
 Households with negative income are excluded from the lowest income category but are included in the totals.
 The 2012 change in CBO’s methodology for valuing government-provided health insurance discussed in footnote 17 also introduces some important changes in the ranking of certain households. As CBO explains in its July 10, 2012 report (p.18):
[T]he higher valuation of government provided health insurance causes about one-eighth of the households in the bottom quintile under CBO’s earlier methodology (roughly 3 million households) to be classified in the second quintile under CBO’s new methodology, and it causes a corresponding number of households to be classified in the bottom quintile rather than the second quintile. The households who moved out of the bottom quintile generally had much lower cash income than did those who moved into it. . . .
 For details on their methods, see Thomas Piketty and Emmanuel Saez, “Income Inequality in the United States: 1913-1998,” Quarterly Journal of Economics, February 2003, or, for a less technical summary, see Saez’s latest update: https://eml.berkeley.edu/~saez/saez-UStopincomes-2015.pdf. Their most recent estimates are available at http://eml.berkeley.edu/~saez/TabFig2015prel.xls.
 Piketty and Saez make available three different data series, each of which treats capital gains slightly differently and therefore yields somewhat different estimates of the share of income going to each group. (For example, estimates of the share of income going to the top 1 percent in 2015 range from 18.39 percent in one series to 20.31 percent in a second series to 22.03 percent in the series we rely on here.) We follow the income concept in Saez’s most recent report and focus on the series that includes capital gains income both in ranking households and in measuring the income that households receive.
 More technically, Piketty and Saez calculate market income by taking the Adjusted Gross Income reported on tax returns and then adding back all adjustments to gross income (such as deductions for health savings accounts, student loan interest, self-employment tax, and IRAs). Note that this definition of market income is not the same as the “market income” concept used in the recent CBO report described above.
 People with income below certain thresholds are not required to file personal income tax returns. Thresholds are determined according to age and filing status. For example, in 2017, the filing threshold for a non-elderly married couple was $20,800; the threshold for an elderly single person was $11,950. Many people who are not required to file tax returns nonetheless pay considerable federal taxes, such as payroll and excise taxes, as well as state and local taxes.
 They estimate the total number of potential filers from Census data by summing the total of married men, widowed or divorced men and women, and single men and women over the age of 20. The number of non-filing tax units in their analysis is the difference between their estimated total and the number of returns actually reported in the IRS data. This methodology assumes the number of married women filing separately is negligible, and it has been quite small since 1948. Before that, however, married couples with two earners had an incentive to file separately, and Piketty and Saez adjust their data to account for that.
 For the years since 1943, non-filers, who account for a small percentage of all filers and of total income, are assigned an income equal to 20 percent of the average income of filers (except in 1944-45, when the percentage is 50 percent). For earlier years, when the percentage of non-filers and their share of income were much higher, Piketty and Saez assume, based on the ratio in subsequent years, that total market income of filers plus non-filers is equal to 80 percent of total personal income (less transfers) reported in the National Income and Product Accounts for 1929-1943 and as estimated by the economist Simon Kuznets for 1913-1928. For those years, the total income of non-filers is the difference between estimated total income and income reported on tax returns.
 According to data from the Bureau of Economic Analysis, wages and salaries now provide about 81 percent of employee compensation; supplemental benefits such as contributions to health and retirement plans provide the rest. In 1980, 84 percent of compensation came through wages and 16 percent through benefits; in 1950, 93 percent came through wages and 7 percent through benefits.
 For an effort to address these limitations, see Thomas Piketty, Emmanuel Saez, and Gabriel Zucman, “Distributional National Accounts: Methods and Estimates for the United States,” Quarterly Journal of Economics, Vol. 133, No. 2, May 2018, http://gabriel-zucman.eu/files/PSZ2018QJE.pdf.
 As discussed in footnote 17, the 2012 change in CBO’s method of valuing government-provided health insurance has a substantial impact on comprehensive income and the ranking of households near the bottom of the income distribution. Since there is no consensus on the ideal approach to valuing government-provided health insurance, researchers are likely to continue exploring how alternative approaches affect the interpretation of historical trends and year-to-year changes in household income and its distribution.
 When income increases by 100 percent, it doubles. When it increases by 300 percent, it quadruples.
 CBO’s 2012 published data showed the average income of the bottom 20 percent rising somewhat faster over the entire 1979-2011 period than that of the middle 60 percent. However, CBO provided data showing that if the old method had been used, the 1979-2011 growth in the bottom quintile would have been 26 percent, lower than the 38 percent growth for the middle 60 percent.
 See the discussion of CBO’s methodology in Part I.
 In the Piketty-Saez data, the average income for the top 1 percent of households in 2015 was about $1.4 million. The average income for the top 0.5 percent was about $2.2 million.
 Assets include such things as savings, stocks, vehicles, homes, and business and financial assets. Liabilities include such things as credit card debt, mortgages, and past-due bills.
 Jesse Bricker et al., “Changes in U.S. Family Finances from 2013 to 2016: Evidence from the Survey of Consumer Finances,” Federal Reserve Bulletin, vol. 103, no. 3, September 2017, https://www.federalreserve.gov/publications/files/scf17.pdf.
Ibid., “Box 3: Recent Trends in Income and Wealth.”
 Emmanuel Saez and Gabriel Zucman, “Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data,” Quarterly Journal of Economics, Vol. 131, No. 2, May 2016, http://eml.berkeley.edu/~saez/SaezZucman2016QJE.pdf.
 A recent analysis by Federal Reserve researchers tries to reconcile differences between the SCF and Zucman-Saez findings through measures such as including estimates for the wealth of the Forbes 400 in the SCF and adjusting upward the assumed rate of return on fixed income assets held by those at the top in Zucman-Saez. The Fed researchers are able to narrow the gap between the two estimates of top-income shares, and they conclude that their estimates “concur that inequality, at least as reflected in top income and wealth shares, has been rising in recent decades.” See Jesse Bricker, Alice Henriques, Jacob Krimmel, and John Sabelhaus, “The Increase in Wealth Concentration, 1989-2013,” Federal Reserve Board, June 2015, http://www.federalreserve.gov/econresdata/notes/feds-notes/2015/increase-in-wealth-concentration-1989-2013-20150605.html.
 There are 48 official poverty thresholds. These thresholds reflect an equivalence adjustment, but not the same three-parameter scale Census uses when it equivalence-adjusts household income. CBO uses another equivalence adjustment, based on the square root of the number of household members.
 Census publishes eight experimental NAS-based poverty rates in addition to the official poverty rate, each calculated using a slightly different methodology. Estimates of these alternative poverty rates are available for each year from 1999 through 2016. The latest tables are available here: https://www.census.gov/data/tables/2016/demo/supplemental-poverty-measure/nas-2016.html. NAS measures also use a three-parameter equivalence scale to adjust for family size and composition. For the purpose of measuring poverty, the NAS report recommended against treating the value of medical benefits as income, noting ways in which medical benefits do not serve the same role as cash. Instead, the report recommended subtracting out-of-pocket medical expenditures from income, since money spent on medical needs is not available to meet the basic needs of food, clothing, shelter, and utilities upon which the NAS poverty threshold is based.
 For more detail, see Liana Fox, “The Supplemental Poverty Measure: 2016,” U.S. Census Bureau, September 2017, https://www.census.gov/content/dam/Census/library/publications/2017/demo/p60-261.pdf.
 WIC — formally known as the Special Supplemental Nutrition Program for Women, Infants, and Children — provides nutritious food, counseling on healthy eating, and health care referrals to low-income pregnant and postpartum women, infants, and children under age 5 who are at nutritional risk.
 Isaac Shapiro and Danilo Trisi, “Child Poverty Falls to Record Low, Comprehensive Measure Shows Stronger Government Policies Account for Long-Term Improvement,” Center on Budget and Policy Priorities, October 5, 2017, https://www.cbpp.org/research/poverty-and-inequality/child-poverty-falls-to-record-low-comprehensive-measure-shows.
 Center on Budget and Policy Priorities, “Chart Book: Economic Security and Health Insurance Programs Reduce Poverty and Provide Access to Needed Care,” updated March 21, 2018, https://www.cbpp.org/research/poverty-and-inequality/chart-book-economic-security-and-health-insurance-programs-reduce.
 In unrounded data from the Columbia University Population Research Center for 2007 and U.S. Census Bureau for 2010, the SPM poverty rate rose from 14.75 percent in 2007 to 15.31 percent in 2010, an increase of 0.56 percentage points.
 For more detail, see Arloc Sherman and Danilo Trisi, “Deep Poverty Among Children Worsened in Welfare Law’s First Decade,” Center on Budget and Policy Priorities, July 23, 2014, http://www.cbpp.org/files/7-23-14pov2.pdf and Arloc Sherman and Danilo Trisi, “Safety Net for Poorest Weakened After Welfare Law But Regained Strength in Great Recession, at Least Temporarily,” Center on Budget and Policy Priorities, May 11, 2015, http://www.cbpp.org/sites/default/files/atoms/files/5-11-15pov.pdf.
 CBPP corrects for undercounting using data from the TRIM microsimulation model, a policy simulation tool developed and maintained by the Urban Institute for the U.S. Department of Health and Human Services Office of the Assistant Secretary for Planning and Evaluation. TRIM starts with person-by-person Census data from the CPS and adjusts it to better match true numbers of recipients of assistance from program records.
 Ife Floyd, LaDonna Pavetti, and Liz Schott, “TANF Reaching Few Poor Families,” Center on Budget and Policy Priorities, updated December 13, 2017, http://www.cbpp.org/research/family-income-support/tanf-continues-to-weaken-as-a-safety-net.
 CBPP’s analysis finds that corrections for underreporting have a particularly large effect on the poverty-reduction estimates for the SNAP program. SNAP lifted 10 million people above the SPM poverty line in 2012 with corrections, compared with 5 million people without these corrections. See Arloc Sherman and Danilo Trisi, “Safety Net More Effective Against Poverty Than Previously Thought,” Center on Budget and Policy Priorities, May 6, 2015, http://www.cbpp.org/research/poverty-and-inequality/safety-net-more-effective-against-poverty-than-previously-thought.
 Arloc Sherman, “Poverty and Financial Distress Would Have Been Substantially Worse in 2010 Without Government Action, New Census Data Show,” Center on Budget and Policy Priorities, November 7, 2011, www.cbpp.org/cms/?fa=view&id=3610.