There is a widespread perception that the tax-and-transfer system in the United States has become less progressive and less redistributive over the past 50 years, even as market-driven income inequality has increased. Indeed, this perception has attained the status of folk wisdom: ask most any reasonably knowledgeable person and they will confidently assert that progressivity and redistribution have declined.

Both progressivity and redistribution are important inputs into policy decisions, particularly as Congress is making substantial changes to both taxes and big transfer programs. If redistribution has indeed declined, regressive changes may make a problem worse. If redistribution has increased, however, those regressive changes may simply mean that the pattern of redistribution is reverting to that of the recent past.

Here, we assess how redistribution has changed over time. Our approach is based on the understanding that there are fundamental uncertainties in measuring progressivity and redistribution – uncertainties in both defining the relevant items to be measured and in evaluating the data. Moreover, these uncertainties allow researchers considerable leeway in interpreting the results, rendering them more vulnerable to bias.

To address this problem, we look for results that are robust across different methods and approaches since these are more likely to be solid than results that hold only under a particular set of choices. To this end, we looked at all publicly available studies of redistribution that meet our criteria. It turns out that only three make their data available and meet minimum credibility criteria.

• Thomas Piketty, Emmanuel Saez and Gabriel Zucman’s 2018 estimates (PSZ)
• 2024 estimates by Gerry Auten (U.S. Treasury) and David Splinter (Joint Committee on Taxation-AS)
• The Congressional Budget Office’s estimate of the distribution of household income (CBO)

We compared these three in detail. To ensure that they are not outliers, we have also reviewed other notable studies on the topic and looked at their top-line results. And at risk of spoiling the suspense, we found near-consensus that the tax-and-transfer system redistributes more now than it did in the past 40 to 60 years. The conventional wisdom, then, does not appear to be true.

But back up a moment. It’s important to note that we take no position on whether the tax-and-transfer system should redistribute more or less than it currently does. There may be good arguments on both sides of the question. It is crucial, however, to base reform proposals on an accurate assessment of the facts. Our goal is to help with that assessment.

Metrics and Measurement

It might seem that measuring redistribution and progressivity is a straightforward matter of accounting. Alas, it is anything but. Data limitations and disputes about what should be measured mean that estimates require subjective judgment – and to some extent, guesswork.

What Counts as Income?

A key problem here is defining what exactly we mean by income. Narrow measures often use the amounts reported on tax returns, or what we will call “fiscal income.” Research estimates that seemed to validate the widely accepted view that progressivity has declined have largely been based on fiscal income. But as the table to the right shows, this definition of income captures only a modest portion of the resources available to households.

The first row consists of estimates of the growth in real (inflation-adjusted) fiscal income per adult from the 45 years from 1979 to 2014, which amounted to 43.7 percent per adult overall – but a negative 26.0 percent for the bottom half of the distribution. Meanwhile the second row (using the data from the same source) shows the growth in a broader measure of income called national income, which was 55.9 percent overall and virtually nothing (0.5 percent) for the bottom half.

If we chose to use fiscal income as an approximation of economic income – the true resources available to individuals – we would, in effect, be accepting the implausible inference that real income fell by over one-quarter for the bottom half of the U.S. population. In fact, since fiscal income only measures income reported for tax purposes, it only indicates that the income reported on tax forms fell.

A second problem with using fiscal income in calculating inequality is that the share of true economic income that is captured on tax returns has changed differently for different income groups. As Figure 1 below shows, the share of economic income captured by fiscal income fell for the bottom 50 percent while it rose modestly for the top 10 percent. This means that using fiscal income to compare the bottom to the top produces not just inaccurate but positively misleading results: it would show an increase in inequality that is merely due to changes in what is reported on tax returns rather than to changes in actual resources available to individuals to consume or save.

The way to correct the distortion is to use a more comprehensive measure of income. But with such broad measures, there is no straightforward method of allocating many components of income to individuals. In some cases, such as income attributable to tax evasion, there is a correct answer (particular individuals, and not others, engaged in tax evasion). In other cases, such as how to assign the value of what economists call public goods – think national defense or public safety – there may be no conceptually coherent answer.

As a result, researchers have substantial latitude in allocating income to individuals. And the choices they make can have large effects on conclusions about distribution. To get a sense of the problems with using a comprehensive measure of income, we describe some of the core issues.

Underreported income. Income hidden to evade taxation is, by its nature, difficult to observe. Estimates of the gap between reported and underreported income is $500 billion per year, with enforcement efforts only narrowing the gap by $75 billion.

Timing of tax-preferred retirement income. Returns in tax-deferred retirement accounts – that is, accounts in which taxes are only collected when cash is withdrawn – constitute a large fraction of capital (sometimes called “unearned”) income. There are no straightforward ways to resolve the resulting ambiguities. The two basic options are to count retirement income when it is contributed (i.e., the year savings are deposited in a retirement account and on an accrual basis thereafter) or when it is distributed in retirement.

Allocating retirement income to individuals when contributed would mean that many retired individuals would have little or no income counted in the calculations, making many people who are living comfortably look poor. Including retirement income only when distributed, however, risks understating earnings during working years.

Taxes. Any estimate of the effects of taxes (and transfers) must make assumptions about their incidence – that is, who actually bears the burden or reaps the benefit. While there is substantial agreement on the incidence of some taxes, such as payroll taxes, estimates of the incidence of others, such as the corporate tax, are all over the place.

The incidence of government transfers is much less studied. But transfers, like taxes, could well be shifted to individuals other than the immediate recipients. Think EITC beneficiaries who use some of the money to support their extended families or, for that matter, landlords who raise rents when housing demand is increased by the distribution of government subsidy vouchers. And choices regarding likely incidence may prove central to reported results.

Capital gains and returns to holding wealth. A large fraction of the returns to holding wealth is not reported on tax returns, and much of it cannot easily be observed. The allocation of this income may significantly affect inequality, although it is not clear in which direction.

A simple method for dealing with the issue is to ignore it, including only realized income. This approach, however, does not capture most gains in tax-preferred accounts as well as gains that are excluded due to stepped-up tax basis at death, likely understating the capital income of wealthy individuals. Moreover, it bunches the timing of income that is earned over many years and includes a large inflation component, likely overstating the capital income of wealthy individuals. Other approaches can avoid these pitfalls, but still require imputing unobserved amounts of income to individuals.

Government spending. Total government spending in fiscal year 2024 was almost $10 trillion, comprising more than one-third of GDP. So the choice of how to allocate government spending among households can have significant effects on the measurement of posttax- and-transfer income. That said, there is no conceptually clear way to divide the benefits. We do not know, for example – indeed, we have no principled basis for deciding – how much individuals benefit from national defense.

Deficits and interest on government borrowing. Net interest on federal debt in 2024 was more than $1 trillion, which was almost 4 percent of GDP – and is almost certain to soar in coming years. (The 2024 federal deficit was nearly $2 trillion.) But we have no rigorous way of assigning the future burden of repayment to income groups.

Using narrower measures of income reduces the uncertainty and researcher discretion, but comes at the potential cost of misleading those who draw policy conclusions from the research. In short, the choice is between making sausage (using a comprehensive but messy measure of income) or using numbers that we know are misleading (using fiscal income).

Who Pays and/or Benefits?

In addition to choosing a measure of income, researchers must decide which units to use to assign the burdens and benefits. Possibilities include individuals, adults, households or tax filing units. If the units were stable over time, or even if they changed at the same pace over time, the choice would generally not matter much in calculating the distributional impact. But that’s a big if: the size and number of each of these units have in fact changed quite differently – and more importantly, have diverged for different groups along the income distribution.

For example, the percentage of tax filers who were married went down from 67 to 37 percent between 1960 and 2019, but only fell from 90 percent to 85 percent for the top 1 percent. This disproportionately increased the number of tax-filing units at the low end, automatically increasing measured inequality in any study that uses tax units as the baseline.

Still, the choice of units might seem of secondary importance. Not so. To illustrate the magnitude of the problem, compare rows 2 and 3 in the table on page 25. Both show estimates of growth rates of national income per unit. Row 2 shows PSZ’s estimate, which is 55.9 percent. Row 3 shows AS’s estimate, which is 75.6 percent, a stark difference for a number that can be estimated with substantial accuracy.

The primary reason for the difference is that PSZ reports income per adult and AS reports income per person: the number of adults has grown more rapidly than the number of people. This means that even though both research groups used the same numbers for national income, PSZ’s measure grew more slowly than AS’s measure.

The difference in units also explains a large part of the difference in PSZ’s and AS’s estimates of growth in income for the bottom 50 percent, which is larger than the overall differences. In particular, family size has shrunk more at the lower end of the income distribution than at the upper, and this divergence could account for much of the difference.

Where to Focus

Studies differ on which policies they aim to evaluate. Some only look at tax policy, while others examine both taxes and transfers. Though some policy analysts have argued that it is appropriate to study the progressivity of taxes alone, doing so can certainly be misleading. The government can achieve more or less the same ends by cutting taxes for a given group or by increasing transfers to that group. The net effect on the group’s resources is the same, but measures that look only at taxes would treat them as different. As a result, we only look at studies that examine both taxes and transfers.

Even if we include all taxes and transfers, studies vary on what they include in pre-tax-and-transfer income and after-tax-and-transfer income. For example, PSZ includes Social Security receipts in pre-tax-and-transfer income while AS treats them as transfers, which means that they show up only in after-tax-and-transfer income. This implies that the measures of pre-tax income, after-tax income and the extent of taxes and transfers differ for the two studies merely because of definitional choices.

How to Measure Progressivity? Redistribution?

Once we’ve agreed on what constitutes income, which units receive the income, and a set of policies to evaluate, we need to choose how to measure progressivity or redistribution. Changes in taxes and transfers can occur throughout the income distribution, which means that metrics looking at only part of the distribution, such as the top 1 percent, may miss important changes happening elsewhere. Even broad measures that look at the entire distribution often use mathematical formulas that reduce complex changes across the entire distribution of income to a single number – an approach which necessarily sacrifices information in favor of simplicity.

Moreover, the underlying distribution of income may change over time for reasons entirely distinct from changes to the tax-and-transfer system. A given level of progressivity or redistribution may mean something different for a society that has fairly equal pre-tax income than for a society that has unequal pre-tax incomes. As a result, comparing the extent of redistribution in, say, 1980 to 2020 is comparing apples to oranges in terms of their societal impact.

How We Did Our Own

Rather than trying to call balls and strikes ourselves, we view the issues described above as sources of fundamental uncertainty and seek results that are robust across different approaches. To this end, we performed a search for all studies of the effects of the taxand- transfer system on inequality since 2012 (10 years before we started the project).

We imposed some minimum criteria on our search results. Since we are interested in changes over time, we looked only at studies that estimate the effects of taxes and transfers over several decades. And for the reasons outlined above, we only considered studies that examine both taxes and transfers. Finally, because we focus on inequality across the entire income spectrum, we only looked at studies that report results across the spectrum – and not, for example, just the top 1 percent or the bottom 20 percent. We also excluded a small number of studies because we believe that they made implausible assumptions about unknown parameters.

We divided our review of the research into two groups of studies. The first group, AS, PSZ and CBO, makes most of their underlying data available, allowing for detailed comparisons. We provide those comparisons in the next section. The second group consists of all other studies that met our search criteria. We review those studies in the subsequent section.

Our primary measure of redistribution is the tax-and-transfer rate at each income level:

This is analogous to the average tax rate, but is more comprehensive in treating taxes and transfers symmetrically. Note that our sign convention is from the government’s perspective: taxes are inflows and transfers are outflows. A positive number means that the household pays net taxes (the taxes it pays exceed the transfers it receives), while a negative number means that the household receives net transfers.

This measure does not account for changes in underlying income. For example, a reduction in the tax-and-transfer rate to a particular income group may be offset by increases in that group’s pre-tax income driven by other factors. To account for this effect, we show changes in pre-tax and after-tax income over time for different income groups. Finally, as a summary measure, we show how the difference between the pre-tax Gini coefficient (a widely used metric of inequality) and the after- tax concentration coefficient, the Reynolds-Smolensky index, has changed over time.

CBO, AS and PSZ

As previously noted, three studies, CBO, AS and PSZ, make their data available and provide estimates of pre-tax and after-tax distributions over substantial periods. The three research groups used different definitions of income, different methods of allocating items of income to units and different units. Results that are common across the three studies are robust to these choices.

Start by comparing the tax-and-transfer rates produced by all three research groups. PSZ provides consistent rankings only for the bottom 50 percent, the 50-90 percentile (which they call the middle 40 percent), the top 10 percent, the top 1 percent and smaller groupings at the very top. To compare the results to PSZ, we use the AS data for the same groups. Figure 2 on page 31 shows the tax-and-transfer rates for each of these groups from 1966 (the first year of consistent data in AS) until 2019.

Looking at the top three lines in each panel (middle 40 percent, the top 10 percent and the top 1 percent), we can see differences between PSZ and AS. These differences are important and substantive, reflecting differences in both methodology and data. But they are not our primary focus: both datasets do show that the key change over the last 60 years has been a dramatic increase in transfers to the bottom half of the population. This increase swamps the impact of changes in tax rates for the top half.

While the two studies unearth similar trends, AS shows larger transfers to the bottom half of the population and a larger increase in those transfers. In 2019, PSZ estimates that the tax-and-transfer rate for the bottom half of the population was 24.4 percent while AS pegs the rate at almost twice that value, 45.6 percent (noting again that a negative transfer is positive to the recipient).

While there may be many factors driving this difference (measuring by adults as units versus individuals among them), one important one is how the two research groups treat Social Security. PSZ excludes Social Security from its definition of transfers (including it in its definition of pretax income), while AS does treat Social Security as a transfer. In spite of this difference, however, both studies show that the increase in transfers to the bottom half of the population is the central story of the tax-and-transfer system over the last half century.

Figure 3 compares AS (left-hand panel) to CBO (right-hand panel). Both these research groups break their data into quintiles, which allows us to isolate groups at the bottom – something we cannot do with the PSZ data. The results are consistent with those shown in Figure 2: there is a widening of tax rates between higher- and lower-income groups, but the impact is swamped by the increase in transfers to the bottom.

Comparing Figures 2 and 3, we see that aggregating the bottom half of the population into a single group masks differences within this group. Both AS and CBO show that almost all the decrease in the tax-and-transfer rate for the bottom half of the population is due to changes for the bottom fifth. The second fifth has some increase in transfers, but nothing like the increase to the bottom quintile. CBO attributes this increase in transfers to Medicaid and CHIP (Children’s Health Insurance Program) due to both an increase in the number of households qualifying for these benefits and the cost of those benefits. Two other big programs, the Supplemental Nutrition Assistance Program (aka food stamps) and Supplemental Security Income, were relatively level during this period.

All three studies yield similar qualitative results. While methodological differences affect the size of their estimates, the central narrative remains consistent: the dominant change in the tax-and-transfer system over the past half century has been an increase in transfers to the bottom.

Figures 2 and 3 showed the tax-and-transfer rates, which address questions about the extent of redistribution. Figure 4 presents similar information, but in terms of income levels and growth. Using AS data, it shows changes in pre-tax income and after-tax income for four different income groups, normalized to the 1966 income for that group. The vertical distance between these two lines tells us the extent of redistribution while the slope of the lines show how pre-tax and after-tax income has changed for each of the groups.

Focusing on pre-tax income (the solid blue lines), we see the standard story about inequality: since around 1980, the pre-tax-income of the top 1 percent grew much faster than the income of other group. By 2019, the income per capita of the top 1 percent was almost triple the level of 1966 income, with all that growth occurring since 1982.

Individuals in the bottom fifth also enjoyed growth from 1966 until 1999, when their income was 1.67 times their 1966 income (again, with all that growth occurring since 1982). In years following, though, their pre-tax income has stagnated: in 2019 it was 1.68 times their 1966 income, the same as it was in 1999.

The results for the middle quintile fell between these two: households in this group saw a steady increase in their income over the entire period (rather than stagnating until 1982). But the overall growth in their incomes was much smaller than for the top 1 percent, reaching 2.2 times 1966 income in 2019.

A look at after-tax income reveals a markedly different story. The four groups experienced almost the same growth rates in after-tax income during the period we examine, equal to just under 2.5 times 1966 income. That is, the tax system appears to be offsetting the changes in the growth of market income, doing more over time as the differences in market income grow. Most importantly, the bottom quintile has seen significant growth in their after-tax income over time. Whatever the story is about the top 1 percent capturing most of the market gains in recent years, after-tax income growth has been more evenly spread.

This outcome is consistent with Figure 2. It shows a sharp increase in transfers to the bottom quintile. We can see this in Figure 4 as well: the gap between the pre-tax and aftertax income of the bottom quintile grows over this period. The gap is also larger for the bottom quintile than for the middle (which also receives net transfers). A graph based on CBO data would show very similar results.

The figures so far demonstrate that the most important change to the fiscal system is the increase in transfers to bottom-income groups. To understand this better, Figure 5 isolates transfers, using AS data and definitions (which means that transfers administered as part of the tax system, such as the earned-income tax credit and the child tax credit, are not included). The left-hand panel shows transfers as a share of national income. The right-hand panel drills further down, showing just the transfers to the bottom quintile along with its income share.

Here we glimpse a somewhat more nuanced story. Although total transfers have gone up (from 5.2 percent of national income in 1966 to 15.3 percent in 2019), transfers to the bottom quintile peaked in 1975 (a period of very high unemployment) at 5.7 percent of national income and have declined since to 4.7 percent of national income.

The increase in transfers as a share of national income has instead largely accrued to the middle quintiles. The second and third quintiles saw large increases in transfers. The fourth and fifth quintiles also enjoyed increases in transfers, though smaller increases than for the middle groups.

The conclusion one might have drawn from the first set of figures is that government policy has increasingly shifted resources toward the bottom. But this last figure suggests the government has instead increasingly shifted resources toward the middle.

The reason that prior figures show that transfer rates have gone up at the bottom while Figure 5 shows transfer amounts have gone down is that income at the bottom has grown more slowly than national income. This can be seen in the right-hand panel of Figure 5, which shows the income share for the bottom quintile declining from 4.6 percent in 1966 to 3.2 percent in 2019.

Looking Elsewhere

To get some perspective on the results above, we reviewed other research on the progressivity and redistribution of the U.S. tax-and-transfer system. We limited our search to papers published since 2012 that show changes in progressivity or redistribution and examine the entire distribution – not just top income groups. The full list of papers along with a brief summary of each is published in an online appendix to the technical version of this paper.

Our review finds that almost all are consistent with the results from CBO, AS and PSZ. Where they differ, there are clear explanations that allow us to reconcile the results.

In our research survey, two facts stand out. First, the major studies come up with consistent results on the changes to the tax-and-transfer system over the past half century. While there are differences in their estimates of the effects at the top of the distribution, there is broad agreement that there have been large increases in the net transfer rate to the bottom and the size of those increases swamp any changes at the top. Second, the tax-and-transfer system has become more redistributive over the last half century, with much of that increase occurring in the last several decades.

In our research survey, two facts stand out. First, the major studies come up with consistent results on the changes to the tax-and-transfer system over the past half century. While there are differences in their estimates of the effects at the top of the distribution, there is broad agreement that there have been large increases in the net transfer rate to the bottom and the size of those increases swamp any changes at the top. Second, the tax-and-transfer system has become more redistributive over the last half century, with much of that increase occurring in the last several decades.

* * *

Why are our conclusions (and those of other rigorous studies) so different from conventional wisdom? We can only speculate. Inequality in the U.S. has risen over the past few decades, accompanied by important social and economic changes. This has inevitably driven academic, political and popular debate. One likely reason for the disconnect is that recent discussions of inequality have focused particularly on gains of the top 1 percent – a question with strong emotional resonance but that unfortunately distracts from the important fact that a great deal of income has been redistributed to the bottom 50 percent.

Whatever the explanation, the truth matters. It matters for shaping the debate and for shaping public policy. And our results raise more questions than they answer. Ballooning transfers have offset (to a greater or lesser extent) rising inequality in labor market outcomes, but what is driving the rising labor market disparities? Do rising transfers introduce disincentives for working and labor market attachment? Whatever the answers, and however one assesses the role of economic factors in explaining the deep social and political divisions in America, it is a stretch to pin it on fiscally driven changes in income inequality.