## Abstract

Scholars have increasingly drawn attention to rising levels of income inequality in the United States. However, prior studies have provided an incomplete account of how changes to specific transfer programs have contributed to changes in income growth across the distribution. Our study decomposes the direct effects of tax and transfer programs on changes in the household income distribution from 1967 to 2015. We show that despite a rising Gini coefficient, lower-tail inequality (the ratio of the 50th to 10th percentile) declined in the United States during this period due to the rise of in-kind and tax-based transfers. Food assistance and refundable tax credits account for nearly all the income growth between 1967 and 2015 at the 5th percentile and roughly one-half the growth at the 10th percentile. Moreover, income gains near the bottom of the distribution are concentrated among households with children. Changes in the income distribution were far less progressive among households without children.

## Introduction

The rise of income inequality has emerged as a central focus of academic and political discourse throughout the past decade. This is particularly true in the United States, the country reporting both the highest levels of income inequality and largest increases in recent decades (OECD 2015). Despite increased attention to rising inequality in the United States, however, the literature on changes in the U.S. income distribution during the past 50 years still has at least three shortcomings.

First, studies of income inequality have often relied on a single indicator, such as the Gini coefficient, to summarize changes in the income distribution. Although the Gini coefficient is a useful instrument for some purposes, it reveals little about changes across different parts of the income distribution (Allison 1978; Atkinson 1970). As a result, studies of income inequality that rely on the Gini have tended to conceal meaningful, nonmonotonic changes in income growth at different points in the overall distribution.

Second, few investigations have shown how changes in the tax and transfer system shaped changes in income growth across the U.S. distribution. This is sometimes due to an exclusive focus on pre-tax, pre-transfer incomes, such as those in most U.S. Census Bureau reports (Navas-Walt and Proctor 2015). Even in studies explicitly assessing incomes before and after taxes and transfers, little attention has been given to the specific policy changes that have promoted more equal income growth.

Third, changes in the income distribution have usually been evaluated only for the population as a whole, even though taxes and transfers affect the incomes of some populations far more than others. Given that transfers from the American welfare state are increasingly targeted at low-income households with children, for example, it is plausible that taxes and transfers have contributed to more progressive income growth among households with children while doing far less to equalize the income distribution of childless households.

In this article, we document and decompose changes in the American income distribution from 1967 to 2015 to address each of these three shortcomings. Measuring trends in economic resources for families with different incomes and needs is a central motivation for studying inequality. Thus, we move beyond summary statistics, such as the Gini coefficient, to provide a more complete account of changes in the 5th through 95th percentiles of the income distribution over the 48 years for which we have data. We also document the direct effect of taxes and transfers on changes in the distribution by decade and household type. This study is among the first to offer a detailed decomposition of how changes in the American tax and transfer system have directly affected income growth across the full U.S. income distribution from 1967 onward.

We make three primary contributions to the literature on income inequality. First, we show that despite the rise in the Gini coefficient, the economic resources of those near the bottom of the distribution grew at about the same rate as the resources of those at the median of the distribution from 1967 to 2015. In fact, after accounting for taxes and transfers, lower-tail inequality (the p50-p10 ratio) declined in three of the five decades that we examine.

Second, we clarify the changing role of taxes and transfers in shaping the American income distribution. Specifically, we show that increases in in-kind transfers, such as benefits from the Supplemental Nutrition Assistance Program (SNAP, or “food stamps”) and refundable tax credits, such as the Earned Income Tax Credit (EITC), explain nearly all the increase in economic resources at the 5th percentile of the distribution from 1967 to 2015 and roughly one-half the increase at the 10th percentile. Indeed, taxes and transfers largely account for the slight decline in lower-tail inequality during this period and are largely responsible for the 5th through 15th percentiles experiencing income growth at the same rate as the 50th percentile.

Finally, we decompose the effect of taxes and transfers on changes in the income distribution by decade and household type. In doing do, we demonstrate that among the decades examined, post-tax, post-transfer income growth was the most progressive between 1967 and 1977, and least progressive between 1977 and 1987. We also show that the strengthening of the tax and transfer system primarily affected households with children, which are the primary recipients of SNAP and EITC benefits. Among households without children, changes in the income distribution were far less progressive from 1967 to 2015, leading to an increase in both upper-tail and lower-tail inequality for these households.

## Background

### Measuring Trends in Income Inequality

The primary purpose of inequality research is to compare different economic groups’ gains and losses over time. The Gini coefficient is the indicator most frequently used to achieve this goal (Atkinson and Jenkins 2020; Bofinger and Scheuermeyer 2019; Guillaud et al. 2020; Jouini et al. 2018; Nolan et al. 2019; OECD 2011). The Gini summarizes the distribution of income using a single indicator that ranges from 0 (complete equality) to 1 (complete inequality). However, the Gini has several shortcomings that limit its value for comparing the gains and losses of different economic groups (Allison 1978; Atkinson 1970). First, the Gini is very sensitive to the income distribution’s extremes. It can rise, for example, if the top 1% or 2% gains more than anyone else or if the bottom 1% or 2% gains less (or loses more) than anyone else. A related weakness of the Gini coefficient is its tendency to mask nonmonotonic changes in income shares across the distribution. For example, a rising Gini indicates that the distribution of income is growing more unequal but offers no insight into whether this is solely due to the top percentiles of the distribution pulling away from rest or whether the middle of the distribution is also pulling away from the bottom.

Scholars have constructed other indicators of income inequality to sidestep these shortcomings. For example, indicators of the 90th, 50th, and 10th percentiles of the overall income distribution are widely used to measure inequality because they provide separate estimates for inequality in the top and bottom halves of the distribution (Förster and Vleminckx 2004; Gottschalk and Smeeding 1997). As a result, they can provide different insight into the potential causes of overall changes in inequality. Growth incidence curves (GICs) also allow for analysis of income growth across the entirety of the distribution. Although GICs have often been applied in the international poverty literature (Ravallion and Chen 2001) and in analyses of global inequality (Milanovic 2016), they have seldom been used to describe changes in the American income distribution.

We adopt both approaches in this study. Trends in the 90th percentile relative to the median (p90-p50 ratio), as well trends in the median relative to the 10th percentile income (p50-p10), offer year-to-year evidence on whether gaps among the 10th, 50th, and 90th percentiles are widening or narrowing. The GIC offers more detailed insight on income growth across the entire income distribution. If we observe that income growth at the 10th percentile of the distribution tends to outpace the median (a declining p50-p10 ratio), for example, is this because the bottom of the distribution in general is experiencing faster income growth? Or might the 10th percentile be a misrepresentative benchmark of the bottom part of the distribution? The GIC provides evidence to answer these questions.

The measurement of inequality is often imbued with value judgments about why one should care about rising income inequality. Those who worry that the rich have too much political influence may care more about the rising share of income going to the top 1.0% or even the top 0.1% (Piketty 2014; Saez and Zucman 2016). Those who worry about the perpetuation of poverty from one generation to the next care more about the apparent decline in the share of income going to the bottom 20%. Although unlikely that any study can be completely devoid of such judgments, we strive here simply to understand income growth across the distribution and the role of taxes and transfers in shaping those changes. We largely avoid trying to analyze trends within the top and bottom 5 percentiles because evidence suggests that household survey data tend to perform poorly at accurately measuring the extremes of the distribution (Meyer and Mittag 2015; Parolin 2019). We therefore focus primarily on the distribution between the 5th and 95th percentiles of the income distribution.

### Measuring Household Resources

Since 2002, the U.S. Census Bureau’s annual report on household incomes during the previous calendar year has included annual estimates of the Gini coefficient for each year since 1967 (Navas-Walt and Proctor 2015). Many scholars treat the Census Bureau’s annual report on changes in the Gini coefficient as the best available indicator of trends in economic inequality. However, the bureau’s published measures have several problems. First, they rely on households’ total pre-tax money income, despite the current widespread agreement that pre-tax money income is an increasingly antiquated measure of household resources, especially near the bottom of the income distribution (Blank and Greenberg, 2008; Fox et al. 2015; Iceland 2013; National Research Council 1995; Wimer et al. 2016). Instead, refundable credits (such as the EITC and Child Tax Credit (CTC)) and in-kind benefits (such as SNAP and housing assistance) have become increasingly important determinants of low-income families’ standard of living. Ignoring them, as the pre-tax measure of income does, can therefore give a quite misleading picture of trends in both poverty and overall economic inequality. For example, recent research on long-term trends in poverty using an adjusted supplemental poverty measure (SPM) that includes tax transfers and in-kind benefits demonstrated that poverty rates have declined substantially relative to trends based on the official poverty measure (Fox et al. 2015; Wimer et al. 2016). The income definition applied within the SPM is comparable to comprehensive measures of income applied frequently in comparative poverty and inequality research. Estimates of inequality from the OECD or Luxembourg Income Study (LIS), for example, tended to use the same measure of post-tax, post-transfer resources (LIS 2016). In this study, we show trends in income growth using both the pre-tax money income concept and the SPM income concept, which we refer to as measuring “post-tax, post-transfer resources.”1

A number of studies have documented how moving from a pre-tax/transfer definition of income to a post-tax/transfer definition changes patterns of income inequality. Comparative research has shown, for example, that taxes and transfers usually do less to reduce levels of inequality in the United States than in other advanced economies (Filauro and Parolin 2019; Gornick and Jäntti 2016; OECD 2011). Albeit useful, this collection of studies leaves several unanswered questions relating to the effects of taxes and transfers on the evolution of the U.S. income distribution. In particular, we highlight three reasons for a deeper analysis of the direct effects of taxes and transfers on the evolution of the American income distribution.

The first reason relates to the critique of the Gini coefficient offered in the prior section. Studies analyzing the effect of taxes and transfers on inequality have tended to compare a pre-transfer Gini coefficient with a post-transfer Gini coefficient. Although a useful first step, it offers little insight into the distributive effects of taxes and transfers across the income distribution. For example, the Gini does not tell us whether a pre-post change in the measure is due to high taxes on high-earning households (reducing net incomes at the top of the distribution), generous redistribution to low-earning households (increasing net incomes at the bottom), or both. Alternative measurement tools, such as GICs, allow us to visualize the effects of different types of taxes and transfers across the entire income distribution.

Second, pre-post analyses reveal little about which types of taxes and transfers have contributed to reductions in inequality. This is especially true regarding changes over time in taxes and transfers on inequality. The American tax and transfer system has undergone substantial changes in recent decades. Cash assistance through Temporary Assistance for Needy Families (TANF) and its predecessors, for example, has steeply declined since the mid-1990s. At the same time, SNAP benefits have gradually increased in participation and value, and refundable tax credits have been substantially expanded. An ideal pre-post analysis would shed light on how changes to each of these programs have affected changes in the income distribution.

Third, the effects of both taxes and in-kind transfers are likely to vary greatly by household type in the United States. Indeed, an overlooked question in inequality research is the role of taxes and transfers in shaping income growth for individuals in households with children versus individuals in households without children. Given that transfers from the American welfare state tend to be concentrated among low-income households with children, we might expect that taxes and transfers more strongly affect levels and trends of these households’ incomes. Households with children are also of special interest because many observers assume that widening income disparities among parents will make children’s educational and economic opportunities more unequal (Duncan et al. 2013; Kaushal et al. 2011; Kornrich and Furstenberg 2013; Reardon 2011).

## Data and Methods

To explore trends in income inequality, we rely primarily on the Census Bureau’s Annual Social and Economic Supplement (ASEC) to the Current Population Survey (CPS). The CPS is a monthly household survey of 100,000 to 200,000 U.S. residents, depending on the year. The ASEC survey is primarily conducted in March and asks detailed questions about how much income each household member aged 15 and older received from various sources during the previous calendar year. We use the surveys conducted from 1968 through 2016, which provide data on current household members’ income during the previous calendar year (1967–2015). Importantly, the data exclude people who are homeless, including many of the 1.3 million children that the U.S. Department of Education estimated to be living in temporary housing in 2015 (Parolin and Brady 2019). This may bias the study’s results when looking at income changes at the very bottom of the distribution and is part of the reason why we exclude the bottom five percentiles in our primary analysis. In the online appendix, however, we also show results that do include the bottom five percentiles.

The most widely cited measures of income inequality, including the Gini, are measured at the household level. Until recently, however, the Census Bureau’s inequality measures ignored variation in household size, treating households with pre-tax income of $50,000 for two people as equivalent to households with$50,000 for eight people. To adjust for household size, we follow the common practice of equivalizing resources by household size, assuming that a household’s economic needs rise in proportion to the square root of its size (Atkinson et al. 1995; Buhmann et al. 1988).2 We assign each household member their household’s size-adjusted pre-tax income and post-tax resources. As a sensitivity check, we also present results using the SPM equivalence scales calculated at the household level (Fig. S10 in the online appendix). The results are substantively similar to our primary analysis.

Our estimates of household resources are based on an augmented ASEC data set used to calculate a version of the SPM that the Census Bureau and Bureau of Labor Statistics recently released (Fox et al. 2015; Wimer et al. 2016). Our measure of post-tax resources differs from these authors’ historical SPM measure, however. As with the SPM, we subtract both federal and state income and payroll taxes from pre-tax money income and add refundable tax credits, like the EITC. As with the SPM, we also add the estimated value of noncash benefits, such as SNAP, the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), the National School Lunch Program, the Low Income Home Energy Assistance Program (LIHEAP), and means-tested housing subsidies. Unlike the SPM, however, we do not subtract household payments for child support, childcare, work-related expenses, or out-of-pocket medical expenses. Also unlike the SPM, we use the household as the unit of analysis instead of the SPM poverty unit, which treats unrelated adults living together as separate units. Our post-tax, post-transfer income definition, which we describe in more detail later, is closer in spirit to household income definitions shown by various Congressional Budget Office reports (e.g., Congressional Budget Office 2018) and is virtually identical to the LIS disposable household income measure.

We should also note that, like the SPM, we do not try to estimate the cash value of Medicare and Medicaid benefits. We view medical insurance as a means to offset expenses that are unevenly distributed and would otherwise make households with medical problems worse off than households with the same money income but no medical problems. Similarly, we do not view these benefits as gifts that make the recipients better off than others with the same income. For instance, a large Medicaid payment to cover the cost of treating a heart attack does not make those who get “free” treatment for a heart attack better off than those who had neither insurance nor a heart attack.

Because the CPS has never asked respondents about their income tax liabilities, payroll tax deductions, or refundable tax credits, we use estimates provided by a tax calculator. For the years 1980–2015, our data set contains the Census Bureau’s tax estimates, which are available in public use data sets. For the years 1967–1979, for which the bureau’s tax estimates are not available, the data set uses estimates from Feenberg and Coutts’s (1993) TAXSIM program, provided by the National Bureau of Economic Research. TAXSIM estimates federal income and payroll taxes for 1967 to the present and state income taxes for 1977 to the present. The data set is further augmented with imputations of noncash benefits in years when these benefits are not available in the CPS (for full details, see Fox et al. 2015: technical appendix). These imputations are mostly prior to 1979 and undoubtedly contain random error. However, most noncash programs were fairly small before 1979, so random errors in assigning their value are unlikely to have a large impact on our estimates of trends in income inequality.

Finally, we adjust household incomes for inflation. Unlike the Census Bureau’s published estimates, which adjust for inflation using a modified version of the consumer price index (CPI-U-RS), we use the implicit price index for personal consumption expenditure (PCE). Although results using the CPI-U-RS are similar to those based on the PCE, there is some debate over which price deflator to use (Meyer and Sullivan 2013). The major disadvantage of the CPI-U-RS is that for many years, it did not account for consumers substituting less expensive goods for more expensive goods when relative prices changed. Ignoring substitution created upwardly biased inflation estimates, especially prior to 1979. Because our measures begin in 1967, we use the PCE. In the online appendix (Fig. S8), we show that the relative growth rates across the distribution are similar using the CPI-U-RS.

### Measuring Inequality

As described earlier, we focus primarily on inequality measures for the total population. However, we also compare trends in income growth for individuals in households with and without children. We focus primarily on three measures of trends in income inequality: the Gini coefficient, the ratio of the 90th income percentile to the median (p90-p50 ratio, or “upper-tail” inequality), and the ratio of the 50th income percentile to the 10th percentile (p50-p10 ratio, or “lower-tail” inequality). We also present growth incidence curves to measure the cumulative income growth of the 5th through 95th percentiles of the distribution. Formally, the cumulative growth of a percentile, or g(p), from 1967 to 2015 is
$gp=y2015py1967p−1.$
(1)

In Eq. (1), y is the income of the SPM unit at percentile p of the income distribution in a given year. The growth incidence curve thus measures the relative income growth from 1967 to 2015 across the income distribution (changes in absolute income growth can be found in the online appendix). Generally, p ranges from 0 to 1, representing the bottom and the top of the distribution, respectively. For this study, p ranges from .05 to .95 because measurement error and sampling frame challenges preclude reliable estimates of incomes in the bottom and top 5 percentiles of the CPS. We present results that include the bottom 5% of the distribution in the online appendix (Figs. S11–S12), but we urge caution in interpreting changes at the very bottom because of the noted measurement error and sampling frame problems.

Building on this, we can also measure the direct contribution of changes to certain social programs to changes in the income distribution. For example, we can measure the direct effect of changes in households’ receipt of food and nutrition support (primarily through SNAP, but also through WIC and free school lunches) on changes in income growth across the distribution. To denote this formally, we use subscripts a and b to represent two different income definitions. Let a represent our measure of post-tax resources and b represent post-tax resources minus food and nutrition assistance. Applying Eq. (1), we can straightforwardly measure g(pa) and g(pb)—the growth of a given percentile, p, given the income definition a or b as indicated. To measure the direct contribution of food and nutrition support on growth at a given percentile, denoted as g ′ (pb), we can calculate
$g′pb=gpa−gpb.$
(2)

For example, imagine that income growth at the 10th percentile from 1967 to 2015, when measured using post-tax resources, equals 50% of the 1967 level, so that g(10a) = .5. Assume that food and nutrition support increased over time for households toward the bottom of the income distribution, so that income growth for the 10th percentile measured without SNAP, WIC, and school lunches is only 30%, making g(10b) = .3. Following Eq. (2), we can estimate the direct contribution of food and nutrition support to income growth at the 10th percentile is .5 – .3 = .2, or 20 percentage points of the baseline level. Among percentiles of the distribution that do not receive the means-tested food and nutrition support, we will find that g(pa) = g(pb), so that the contribution of food and nutrition support is 0. We can then estimate the contribution of each set of taxes and transfers to income growth at each point of the income distribution.

We decompose all taxes and transfers into four groups to document changes in their contributions over time: (1) food and nutrition support, (2) housing support, (3) cash transfers, and (4) tax-based transfers minus tax liabilities. Food support includes SNAP, WIC, and school lunches. Housing support includes housing subsidies and LIHEAP. Note that housing subsidies include both public housing and rent subsidies, such as Section 8. They do not, however, include as income any value for owner-occupied homes. Cash transfers include cash assistance from TANF (and AFDC, its predecessor program), unemployment insurance, and Supplemental Security Income (SSI). Taxes include tax liabilities and tax credits, such as the EITC and CTC. Table A1 of the online appendix shows trends in benefit allocations for each of these programs from 1967 to 2015.

Finally, we emphasize that we are measuring only the direct effect of transfers on the income distribution. We do not account for behavioral responses or indirect effects that might result if the transfer programs were changed in some way. This is a limitation of all pre-post analyses of income transfers and should be kept in mind when interpreting the results.

## Results

### Descriptive Trends in Inequality and Income Growth

Figure 1 shows indexed trends in the Gini coefficient, the p90-p50 ratio, and the p50-p10 ratio for each year from 1967 to 2015, both before and after taxes and noncash transfers are accounted for. The first panel shows trends in the Gini coefficient, the second panel displays trends in the p90-p50 ratio (upper-tail inequality), and the third panel displays trends in the p50-p10 ratio (lower-tail inequality).

With either income definition, both the Gini and the p90-p50 ratio increase from 1980 onward. From 1967 to 2015, both measures of inequality increased by roughly 20% when accounting for all taxes and transfers. Changes in these two indicators align almost perfectly (r = .97 for the Gini and the p90-p50 ratio), again highlighting the Gini coefficient’s sensitivity to the rise of high incomes. The post-tax, post-transfer Gini aligns less closely with changes in the post-tax/transfer p50-p10 ratio (r = .60). Despite the steep increase in the Gini, the final panel of Fig. 1 shows that lower-tail inequality actually declined from 1967 to 2015. As we discussed in the Introduction, the Gini can mask nonmonotonic variation in income growth in the underlying distribution. Put differently, the Gini tends to conceal trends in the p50-p10 ratio and particularly income growth at the lower end of the distribution.

A second important takeaway from Fig. 1 is the role of taxes and noncash transfers in reducing inequality, and the p50-p10 ratio in particular. For each of the three indicators, we see that inequality measured with pre-tax money income increased by approximately 30% from 1967 onward. For the p50-p10 ratio, however, taxes and noncash transfers fully offset the increase in pre-tax and transfer inequality. For the Gini coefficient and the p90-p50 ratio, in contrast, taxes and transfers played a much smaller role in limiting the rise in inequality.

Finally, Fig. 1 shows that the 1980s were a critical turning point in the evolution of the inequality in the United States (at least within five decades covered here). From 1967 to the start of the 1980s, each of the three measures of inequality declined after accounting for taxes and transfers. After 1980, however, the Gini and upper-tail inequality steadily increased. Since 1990, the p50-p10 ratio has ebbed and flowed, ending up near its 1967 level in 2015.

Is the 10th percentile unique in some way, or has the bottom half of the income distribution also experienced growth on par with the median? To answer this question, we present in Fig. 2 the growth incidence curves for the 5th to 95th percentiles of the income distribution from 1967 to 2015. The black line in Fig. 2 reflects cumulative income growth for each percentile from 1967 to 2015 using only pre-tax money income. The gray line represents income growth at the same percentiles when measured with post-tax resources. Looking at the median (50th percentile) of the distribution, for example, we see that real income growth measured with pre-tax money was about 80% from 1967 to 2015 and about 75% when taxes and in-kind transfers are included. The drop in growth rates from pre-tax to post-tax/post-transfer incomes occurs as families in the middle of the distribution benefit less from government transfer yet still must pay both income and payroll taxes. This is consistent with other research documenting how families in the middle of the income distribution benefit little from government transfer programs and face high marginal tax rates on income (Institute for Research on Poverty 2019).

Looking more broadly at the distribution, we see that tax credits and in-kind transfers have enhanced relative income growth between the 5th and 40th percentiles. This is particularly true for the 5th through 10th percentiles. For example, post-tax resources at the 10th percentile increased nearly 80% from 1967 to 2015, which is slightly more than the rise at the 50th percentile. Using the pre-tax money income measure, in contrast, the 10th percentile has grown at only one-half the rate of the median (40% vs. 80%). Interestingly, the 20th through 45th percentiles experienced slightly lower income growth than the 50th percentile. These findings suggest that the American welfare state has increased the relative resources available to those at the very bottom but has done less to boost relative incomes between the 10th and 40th percentiles. Figure 2 also indicates that the observed decline in the p50-p10 ratio from 1967 to 2015 does not reflect broad income growth across the bottom of the distribution. Switching to a p50-p15 ratio, for example, would show an increase in lower-tail inequality rather than a decrease.

Although Fig. 2 documents cumulative changes in income growth from 1967 to 2015, it is also informative to examine the trends by 10-year intervals to identify which decades were beneficial for the income growth at different parts of the distribution. Figure 3 shows the average annual growth rate by income percentile for five separate periods, as labeled in the figure. For simplicity, Fig. 3 shows only changes in income measured using post-tax resources.

Figure 3 shows that the bottom half of the distribution experiences its fastest income growth during the 1967–1977 period. This was also the only period in which the 10th percentile experienced income growth at the same rate as the 90th percentile. The 10th percentile also grew at a faster rate than the 50th percentile in three of the five periods examined (the two exceptions are 1977–1987 and 1997–2007). Income growth appears to have been most progressive between 1967 and 1977, and least progressive between 1977 and 1987. Only from 2007 to 2015 did the median experience no real income growth. There are numerous reasons why changes might differ by decade, and we are mostly agnostic about the likely causes, preferring instead to document patterns over time using a more robust measure of household resources. However, potential causes include changes in the labor market (e.g., technological change, minimum wage policy, and unionization rates), changes to transfer policies (e.g., the nationalization of the Food Stamp Program in the 1970s and the large expansion of the EITC in the 1990s), and changes in demographics (e.g., the rise of single-headed families, increasing education among the adult population, and so forth).

### The Direct Effect of Taxes and Transfers on Income Growth

Figure 2 makes clear that changes in taxes and transfers have had a strong effect on income growth near the bottom of the income distribution, but it does not indicate which types of transfers are primarily responsible for these changes. Following Eq. (2) in the prior section, Fig. 4 shows the contributions of our four groups of taxes and transfers to income gains at different percentiles of the income distribution.

Above the 35th percentile and higher, changes in receipt of food/nutrition support, cash transfers, and housing support from 1967 to 2015 had no observable direct effect on income growth. Increases in the value of food and nutrition support, in particular, boost income growth only toward the bottom of the distribution. At the 5th percentile, for example, the rise of food and nutrition support (SNAP, in particular) contributed about 25 percentage points to income growth, although this effect fades quickly as we move up the income distribution.3 Above the 25th percentile, changes in food and nutrition support do not appear to have an observable effect on changes in income. Increases in housing assistance also play a large role in income growth near the very bottom, but they fade above the 20th percentile of the distribution.

Changes in cash assistance stand out as the only means-tested transfer that weakened over time. After the introduction of TANF in the mid-1990s, cash assistance caseloads declined steadily as access to the program’s benefits waned (Stanley et al. 2016). The decline in the cash transfers captures this effect. When cash transfers are dropped from the income definition in all years, resource growth is about 30 percentage points higher at the 5th percentile than when they are included. Put differently, cash transfers contributed much to the bottom of the income distribution in 1967 but far less by 2015. Figures 2 and 3 both suggest, however, that the rise of housing support, food and nutrition assistance, and refundable tax credits has more than offset the decline of cash assistance for households at the bottom of the income distribution. In the online appendix (Fig. S7), we break down the contribution of cash transfers into the group’s constituent parts: AFDC/TANF, SSI, and unemployment insurance. The results demonstrate that the increase in SSI expenditures contributed favorably to income growth in the lower parts of the distribution but that declines in AFDC/TANF and unemployment insurance have contributed negatively.

Meanwhile, taxes on income have become more progressive over time. A rise in refundable tax credits, primarily through the EITC, boosted the growth of after-tax income among those below the 35th percentile. The rising value of tax credits also contributed directly to a 15 percentage point increase in income growth at the 10th percentile—more than cash transfers or food/nutrition support at this percentile. Meanwhile, rising tax liabilities dampened income growth above the 35th percentile. Note that much of the apparent increase in the redistributive effect of taxation is due to the large increases in pre-tax earnings, as documented in Fig. 1.

So far, we have observed that taxes and transfers are largely responsible for driving the increase in post-tax resources among households between the 5th and 35th percentiles of the distribution. Increases in food and nutrition support, housing assistance, and refundable tax credits appear to have had the strongest positive effects toward the bottom of the distribution. However, these latter programs also generally targeted households with children. In looking at the income growth of individuals in households with versus without children, we see different trends in income growth and inequality.

Figure 5 presents the cumulative income growth rates for these two household types. We immediately see that changes in the income distribution between 1967 and 2015 look different for households with and without children. For these analyses, we remove older-age adults (ages 65 and older) from the estimates to present a like-to-like comparison of households with and without children. This is important given that the resources of retirement-age adults are quite different from those of working-age adults. The left side of Fig. 5 shows that transfers to households with children have substantially increased income growth in the bottom quartile of the distribution. Among households without children, however, the differences between pre- and post-tax income are much smaller. Moreover, the p50-p10 ratio actually increased over time for childless households. Thus, the decline in lower-tail inequality among the total population is primarily attributable to the rise in transfers targeted at households with children. For further evidence of this, Fig. 6 displays the direct effect of changes in tax and transfer programs on changes in cumulative income growth by percentile for the two different household types.

As shown in Fig. 2, households with children and resources near the bottom of the distribution benefited substantially from growth in the SNAP and EITC programs. SNAP provided food support for a growing share of low-income families, and the EITC provided more cash support to families in which the adult worked in a low-wage job. At the same time, however, changes in cash transfers via AFDC/TANF contributed to declines in income growth among households with children at the bottom of the distribution. In the online appendix (Figs. S1–S6), we present similar results for jobless households, households with part-time workers, and older households. These alternative analyses confirm that the decline of cash support has been particularly detrimental for households with jobless and part-time workers, with much less effect on households with full-time workers. Consistent with findings from Moffitt and Pauley (2018), we find that changes to the tax and transfer system have had much smaller effects on childless families with incomes in the bottom half of the distribution. Moreover, Fig. S4 (online appendix) shows that when we compare jobless households with working households, the jobless households have gained very little from expansions in refundable tax credits, which tend to be conditional on employment. Jobless households have also lost much more from the decline of cash transfers.

Households in the first decile benefited particularly from increases in SSI and unemployment insurance (UI) receipt between 1967 and 2015. The welfare state otherwise had a small redistributive effect on these families. Again, the story of changes in the income distribution—and especially the direct effect of taxes and transfers—varies dramatically between households with and without children. Lower-tail inequality has declined over time for individuals in households with children but has risen for childless households.

## Discussion and Conclusion

In this study, we set out to offer a clearer picture of changes in the U.S. income distribution since 1967. We use several measures of trends in income inequality to understand how taxes and transfers directly shaped income growth across the distribution and across household type. Our findings lead to three main conclusions.

First, trends in the Gini coefficient tell an incomplete story with respect to the uneven changes in America’s income distribution. Trends in the Gini align almost perfectly with trends in the p90-p50 ratio but very poorly with trends in the p50-p10 ratio. This is not surprising. As others have shown, the Gini is particularly sensitive to income changes at the top of the distribution. However, our findings also emphasize the importance of multiple indicators to understand changes in the income distribution. We show that the American welfare state has worked particularly well to lift incomes near the 10th percentile of the distribution, to the point where lower-tail inequality actually declined from 1967 to 2015 after taxes and transfers were accounted for. This is consistent with other research showing that these benefits have become more important for the working poor, particularly since the 1990s when expansion of Social Security, the EITC, and SNAP generally offset the decline of ADFC and TANF, shifting support from the poorest families to the working poor, elderly, and disabled (Hardy et al. 2018; Moffitt and Scholz 2010).

Our second main finding is that changes to the U.S. tax and transfer system have greatly strengthened income growth at the bottom of the distribution. Omitting taxes and noncash transfers when measuring a household’s resources, as is the common practice in U.S. Census Bureau reports, leaves an incomplete understanding of inequality and changes in the income distribution. Specifically, we find that food and nutrition assistance (such as benefits from the SNAP program) and refundable tax credits (such as the EITC) have increasingly benefited individuals with low pre-tax income. This increase in transfers has contributed to nearly all the income growth at the 5th percentile of the distribution from 1967 to 2015 and roughly one-half the rise of the 10th percentile. Moreover, taxes and transfers largely account for the 5th through 15th percentiles experiencing net income growth at the same rate as the 50th percentile from 1967 to 2015. We emphasize again that these are relative growth rates (the percentage change in each percentile from 1967 to 2015) rather than measures of absolute income growth. Although the American tax and transfer system is still weaker than that of other advanced economies, it appears to have strengthened considerably over time in its ability to raise the incomes of low earners.

That said, our third key finding reveals a large gap in the welfare state’s treatment of households with children versus those without. Taxes and transfers tend to offer less for individuals in households without children (and working-age adults without children, in particular) relative to households with children. As a result, we see rising inequality across the income distribution of individuals in households without children but declining lower-tail inequality among households with children. As Figs. 5 and 6 demonstrate, there is much less difference in pre-tax and transfers versus post-tax and transfer resources for individuals in households without children.

There are several limitations to our study. First, even our expanded definition of resources does not provide a complete picture of inequality trends. For instance, we consider only metrics based on income and ignore consumption, which could change the picture. Future research should use hardship metrics from the Survey of Income and Program Participation—measuring households’ abilities to meet expenses for mortgages, rent, utilities, or medical care—to investigate whether trends in household hardship run parallel to our findings.

In terms of income measures, even our expanded definition does not include several important components. First, income and payroll taxes are not the only taxes that affect households. Sales taxes, property taxes, corporate profits taxes, and excise taxes also matter (Newman and O’Brien 2011). Second, our figures do not account for employer- or government-provided health insurance (Burkhauser and Simon 2008), which clearly impacts the well-being of families at different points in the distribution over time. Third, the figures presented here include the value of housing subsidies for low-income families but not the value of mortgage subsidies for homeowners or the use value of owner-occupied housing. Arriving at a full definition of post-tax resources that includes all these components is beyond the scope of this article (and probably beyond the scope of available data), but it is important to acknowledge that accounting for these omissions might change our revised picture of trends in economic inequality in other new ways.

## Acknowledgments

Liana Fox contributed to this article in her personal capacity. The views expressed in this research, including those related to statistical, methodological, technical, or operational issues, are solely those of the authors and do not necessarily reflect the official positions or policies of the U.S. Census Bureau. The authors also wish to thank Irwin Garfinkel, Neeraj Kaushal, and Jane Waldfogel for their contributions to creating the data underlying this study, and also the anonymous reviewers of the study for their valuable feedback. Funding from the Annie E. Casey Foundation and The JPB Foundation is gratefully acknowledged, though all opinions and errors are the authors’ alone.

## Authors’ Contributions

Wimer, Fox, Fenton, and Jencks led initial writing and data analysis. Parolin led subsequent writing and data analysis.

## Data Availability

All data are available through the public versions of the U.S. Current Population Survey. Full replication code is available upon request to the authors.

## Compliance With Ethical Standards

### Conflict of Interest

The authors report no conflict of interest.

### Ethics and Consent

The authors report no concerns relating to ethics or consent.

## Notes

1

Pre-tax money income includes cash transfers from programs such as Temporary Assistance for Needy Families (TANF) as well as pension support from Social Security. It includes income from child support and private transfers across households but does not include child support paid or remittances to family members outside the country. It should not be confused with market income (earnings from paid labor only), which comparative studies often apply in pre-post analyses.

2

There are numerous alternate ways to adjust for household size, as detailed by Buhmann et al. (1988). When household income is adjusted for household size, we assume some economy of scale whereby household members share fixed costs such that each additional member lowers the household’s cost per capita. To estimate this, household income is typically divided by household size raised to some scale elasticity between 0 and 1 (Household income/household sizee, where e equals the scale elasticity). When elasticity is 0, there is no adjustment for household size, and thus each household member requires the same dollar amount to afford the same level of consumption. When elasticity is 1, there are no economies of scale, and so each member has equivalent costs regardless of household size. We choose a commonly used approach that sets elasticity between the two extremes at .5 (Johnson et al. 2005).

3

Given evidence of underreporting of SNAP benefits in the CPS ASEC, the actual effects of SNAP on income growth at the bottom of the distribution are likely larger than they appear in Fig. 4.

## References

Allison, P. D. (
1978
).
Measures of inequality
.
American Sociological Review
,
43
,
865
880
.
Atkinson, A. B. (
1970
).
On the measurement of inequality
.
Journal of Economic Theory
,
2
,
244
263
. 10.1016/0022-0531(70)90039-6
Atkinson, A. B., & Jenkins, S. P. (
2020
).
A different perspective on the evolution of UK income inequality
.
Review of Income and Wealth
,
66
,
253
266
.
Atkinson, A. B., Rainwater, L., & Smeeding, T. M. (
1995
).
Income distribution in OECD countries: Evidence from the Luxembourg Income Study
.
Paris, France
:
OECD
.
Blank, R. M., & Greenberg, M. H. (
2008
).
Improving the measurement of poverty
(Policy Brief No. 2008-17).
Washington, DC
:
Brookings Institution
.
Bofinger, P., & Scheuermeyer, P. (
2019
).
Income distribution and aggregate saving: A non-monotonic relationship
.
Review of Income and Wealth
,
65
,
872
907
. 10.1111/roiw.12376
Buhmann, B., Rainwater, L., Schmaus, G., & Smeeding, T. M. (
1988
).
Equivalence scales, well-being, inequality, and poverty: Sensitivity estimates across ten countries using the Luxembourg Income Study (LIS) database
.
Review of Income and Wealth
,
34
,
115
142
. 10.1111/j.1475-4991.1988.tb00564.x
Burkhauser, R. V., & Simon, K. I. (
2008
).
Who gets what from employer pay or play mandates?
.
Risk Management and Insurance Review
,
11
,
75
102
. 10.1111/j.1540-6296.2008.00131.x
Congressional Budget Office
. (
2018
).
The distribution of household income, 2015
DeNavas-Walt, C., & Proctor, B. (
2015
).
Income and poverty in the United States: 2014
(Current Population Reports, P60-252).
Washington, DC
:
U.S. Census Bureau
.
Duncan, G. J., Kalil, A., & Ziol-Guest, K. M. (
2013, April
).
Increasing inequality in parent incomes and children’s completed schooling: Correlation or causation?
Paper presented at the annual meeting of the Population Association of America,
New Orleans, LA
.
Feenberg, D., & Coutts, E. (
1993
).
The TAXSIM model
.
Journal of Policy Analysis and Management
,
12
,
189
194
. 10.2307/3325474
Filauro, S., & Parolin, Z. (
2019
).
Unequal unions? A comparative decomposition of income inequality in the United States and European Union
.
Journal of European Social Policy
,
29
,
545
563
. 10.1177/0958928718807332
Förster, M. F., & Vleminckx, K. (
2004
).
International comparisons of income inequality and poverty: Findings from the Luxembourg Income Study
.
Socio-Economic Review
,
2
,
191
212
. 10.1093/soceco/2.2.191
Fox, L., Wimer, C., Garfinkel, I., Kaushal, N., & Waldfogel, J. (
2015
).
Waging war on poverty: Poverty trends using a historical supplemental poverty measure
.
Journal of Policy Analysis and Management
,
34
,
567
592
. 10.1002/pam.21833
Gornick, J. C., & Jäntti, M. (
2016
).
Poverty
.
Pathways Magazine: The poverty & inequality report
(Special issue),
16
24
.
Gottschalk, P., & Smeeding, T. M. (
1997
).
Cross-national comparisons of earnings and income inequality
.
Journal of Economic Literature
,
35
,
633
687
.
Guillaud, E., Olckers, M., & Zemmour, M. (
2020
).
Four levers of redistribution: The impact of tax and transfer systems on inequality reduction
.
Review of Income and Wealth
,
66
,
444
466
. 10.1111/roiw.12408
Hardy, B., Smeeding, T., & Ziliak, J. P. (
2018
).
The changing safety net for low-income parents and their children: Structural or cyclical changes in income support policy?
.
Demography
,
55
,
189
221
. 10.1007/s13524-017-0642-7
Iceland, J. (
2013
).
Poverty in America: A handbook
. (3rd ed.).
Berkeley
:
University of California Press
. 10.1525/9780520956797
Institute for Research on Poverty
. (
2019
).
Understanding benefit cliffs and marginal tax rates
(IRP Fast Focus Research/Policy Brief No. 43–2019).
:
Institute for Research on Poverty
.
Johnson, D., Smeeding, T. M., & Torrey, B. B. (
2005
).
United States inequality through the prisms of income and consumption
.
Monthly Labor Review
,
128
(
4
),
11
24
.
Jouini, N., Lustig, N., Moummi, A., & Shimeles, A. (
2018
).
Fiscal policy, income redistribution, and poverty reduction: Evidence from Tunisia
.
Review of Income and Wealth
,
64
(
Suppl. 1
),
S225
S248
.
Kaushal, N., Magnuson, K., & Waldfogel, J. (
2011
).
How is family income related to investments in children’s learning?
. In G. J. Duncan, & R. J. Murnane (Eds.),
Whither opportunity? Rising inequality schools, and children’s life chances
(pp.
187
206
).
New York, NY
:
Russell Sage Foundation
.
Kornrich, S., & Furstenberg, F. (
2013
).
Investing in children: Changes in parental spending on children, 1972–2007
.
Demography
,
50
,
1
23
. 10.1007/s13524-012-0146-4
Luxembourg Income Study (LIS)
. (
2016
).
Disposable household income
[Data set].
Esch-Belval, Luxembourg
:
LIS Cross-National Data Center
.
Meyer, B. D., & Mittag, N. (
2015
).
Using linked survey and administrative data to better measure income: Implications for poverty, program effectiveness and holes in the safety net
(NBER Working Paper No. 21676).
Cambridge, MA
:
National Bureau of Economic Research
.
Meyer, B. D., & Sullivan, J. X. (
2013
).
Winning the war: Poverty from the Great Society to the Great Recession
(NBER Working Paper No. 18718).
Cambridge, MA
:
National Bureau of Economic Research
.
Milanovic, B. (
2016
).
Global inequality: A new approach for the age of globalization
.
Cambridge, MA
:
Harvard University Press
. 10.4159/9780674969797
Moffitt, R. A., & Pauley, G. (
2018
).
Trends in the distribution of social safety net support after the Great Recession
(Working paper).
Stanford, CA
:
Stanford Center on Poverty and Inequality
Moffitt, R., & Scholz, J. K. (
2010
).
Trends in the level and distribution of income support
.
Tax Policy and the Economy
,
24
,
111
152
. 10.1086/649830
National Research Council
. (
1995
).
Measuring poverty: A new approach
.
Washington, DC
:
.
Newman, K. S., & O’Brien, R. L. (
2011
).
Taxing the poor: Doing damage to the truly disadvantaged
.
Berkeley
:
University of California Press
. 10.1525/9780520948938
Nolan, B., Roser, M., & Thewissen, S. (
2019
).
GDP per capita versus median household income: What gives rise to the divergence over time and how does this vary across OECD countries?
.
Review of Income and Wealth
,
65
,
465
494
. 10.1111/roiw.12362
Organisation for Economic Co-operation and Development (OECD)
. (
2011
).
Divided we stand: Why inequality keeps rising
.
Paris, France
:
Organisation for Economic Co-operation and Development
.
Organisation for Economic Co-operation and Development (OECD)
. (
2015
).
In it together: Why less inequality benefits all
.
Paris, France
:
Organisation for Economic Co-operation and Development
Parolin, Z. (
2019
).
The effect of benefit underreporting on estimates of poverty in the United States
.
Social Indicators Research
,
144
,
869
898
. 10.1007/s11205-018-02053-0
Parolin, Z., & Brady, D. (
2019
).
Extreme child poverty and the role of social policy in the United States
.
Journal of Poverty and Social Justice
,
27
,
3
22
. 10.1332/175982718X15451316991601
Piketty, T. (
2014
).
Capital in the twenty-first century
.
Cambridge, MA
:
The Belknap Press of Harvard University Press
.
Ravallion, M., & Chen, S. (
2001
).
Measuring pro-poor growth
(English) (World Bank Policy Research Working Paper No. WPS2666).
Washington, DC
:
World Bank
.
Reardon, S. F. (
2011
).
The widening academic achievement gap between the rich and the poor: New evidence and possible explanations
. In G. J. Duncan, & R. J. Murnane (Eds.),
Whither opportunity? Rising inequality, schools, and children’s life chances
(pp.
91
116
).
New York, NY
:
Russell Sage Foundation
.
Saez, E., & Zucman, G. (
2016
).
Wealth inequality in the United States since 1913: Evidence from capitalized income tax data
.
Quarterly Journal of Economics
,
131
,
519
578
. 10.1093/qje/qjw004
Stanley, M., Floyd, I., & Hill, M. (
2016
).
TANF cash benefits have fallen by more than 20 percent in most states and continue to erode
(Technical report).
Washington, DC
:
Center on Budget & Policy Priorities
.
Wimer, C., Fox, L., Garfinkel, I., Kaushal, N., & Waldfogel, J. (
2016
).
Progress on poverty? New estimates of historical trends using an anchored supplemental poverty measure
.
Demography
,
53
,
1207
1218
. 10.1007/s13524-016-0485-7

## Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.