This article originally appeared in Forbes
“It’s not easy finding new and interesting ways to illustrate the growth of income inequality over the past few decades,” says Mother Jones blogger Kevin Drum. You think that’s hard? Try getting the commentariat to acknowledge that the leading source of inequality trends cannot say what everyone thinks it does.
Let’s stipulate that income inequality is at staggering levels in the U.S., and that income concentration at the top has probably risen (probably). But is it true that the rich have “devoured the American economy”, that they have “conquered” it, or that “the rich aren’t just grabbing a bigger slice of the income pie—they’re taking all of it”? These characterizations have in common their source in a chart from a paper by Bard College economist Pavlina Tcherneva.
The chart, which I’ve reproduced below, shows that over time, more and more of the income gains in economic expansions go to the top ten percent. In the current ongoing expansion and the last one, all of the gains have gone to the top. (Note that the bars do not indicate the percentage change in income for these groups—they give the share of total gains that went to each. Within an expansion, the two bars add up to 100 percent.)1
Tcherneva’s chart went viral last week, receiving the same uncritical embrace that met Thomas Piketty’s Capital in the Twenty-First Century earlier in the year. It was hailed as “astonishing,” “stunning,” and “the most important chart about the American economy you’ll see this year.”
But as a depiction of income inequality trends, the Tcherneva chart obscures more than it reveals. The shortcomings of the chart—and more broadly, its source data compiled by Piketty and his fellow economist Emmanuel Saez—do not stem from errors made by Piketty and Saez (or Tcherneva), but from the pervasive misinterpretations about the practical relevance of the data. We’ll get to that gradually, but first, because everything needs to be illustrated “in one chart” these days, here is the takeaway for this piece—the characterization of how gains have been shared that is most relevant to our policy debates:
Problem 1. We Care About Households and Individuals, Not “Tax Units”
The Piketty and Saez estimates come from IRS tax return data. By relying on tax returns where most before them had used survey data, Piketty and Saez overcame shortcomings of the latter that had obscured the degree of income concentration and its rise over time. The most widely-used data source for income statistics—the Current Population Survey (or “CPS”)—badly understates income concentration and badly understates its increase.
But the tax return data badly understate the disposable incomes of households below the top, badly understate the market incomes of the working-age population below the top, and badly understate the rise in those incomes. I’ve gone on about this before, a lot, but let me try to tailor the argument to this particular case. (No, I mean a lot.)
The Piketty and Saez data series yield the average “tax return gross income” (more on that soon enough) of the bottom 90 percent and top 10 percent of “tax units.” Tax units are basically tax returns, with a small number of would-be tax returns added in to account for people who make so little they don’t have to file. A tax unit can be an individual or a married couple. It is not a household, except in cases where a single person or a married couple lives alone.
Tax units are poorer than households on average. A divorced mother with two teenagers who have summer jobs is one household with three tax units (two of them very “poor” indeed). A married couple with a dependent college student living in a dorm who has a work-study job is one household in the CPS but two tax units in the Piketty-Saez figures (the unit on campus also typically “low income”). A cohabiting unmarried couple each making the same salary comprise two tax units that are individually half as rich as the household they share. Four Millennial roommates create four poorer tax units in one household.
The Piketty-Saez data indicates the average income of the bottom 90 percent of tax units was $31,284 in 2010. In analyses that I will discuss momentarily that attempt to use the same income concept as Piketty and Saez, the mean for the bottom 90 percent of households was $39,397 in 2010 (putting it in 2012 inflation-adjusted dollars using the same adjustment as in the Piketty-Saez data).
To see how the distinction between tax units and households can be problematic for trend analyses, consider a thought experiment where there are 100 households in which 180 tax units live. In 80 households there are two tax units who each receive $25,000, while 20 households have a single tax unit who receives $1 million. One year later, every tax unit has seen income growth of five percent. If tax units are analyzed, the top ten percent (the top 18 tax units) receive 75 percent of the income growth by the measure Tcherneva uses. That is true even though everyone saw the same percentage increase in income, because there was already inequality when the increase occurred. Meanwhile, if households are analyzed rather than tax units, the top ten percent (the top 10 households) receive just 42 percent of income gains. How can that be? Because there was more initial inequality across tax units than there was across households.
Is this theoretical exercise relevant to the Piketty and Saez data? The next chart compares the trend that Tcherneva displayed using their data with the trend one gets using household data from the Congressional Budget Office. CBO assigns each tax unit in the same IRS data used by Piketty and Saez to a similar unit created in the CPS. It then adds the CPS unit’s transfer income to the tax unit from the IRS data before aggregating the tax units back into CPS households. Anyone sufficiently familiar with the data (a small club, to be sure) can create an income measure fairly consistent with the “tax return gross income” concept used by Piketty and Saez (again, more spinach on that to come), so that their tax-unit-based figures may be validly compared to household-based figures.2 One also has to readjust the CBO figures for inflation to make them consistent with the inflation adjustment used by Piketty and Saez.
The CBO data only run from 1979 to 2010, so I exclude the 2009-12 period that Tcherneva shows from this chart. I show the original Tcherneva estimates for the expansions from 1949 to 1979, which are based on tax units, and replace her estimates after 1979 with the household-based CBO figures. The dashed line between these two sets of estimates is intended as a reminder that they may not be directly comparable; we do not know what the 1949 to 1979 period would look like for households.
Nevertheless, the two sets of estimates line up remarkably well. The rise in the top ten percent’s share of income gains is actually steadier than in Tcherneva’s original chart, which is suggestive that perhaps the two sets of estimates are comparable. The 1982-90 expansion looks like it experienced less lopsided growth when the CBO data are used, but otherwise the two sources produce similar conclusions about the trend in inequality. That will be less true in the next set of charts.
Problem 2. We Care About Business Cycles, Not Expansions
When a group experiences gains in an expansion, how we evaluate that should depend on whether it subsequently experiences losses in a recession. When all is said and done, we want to know whether people are better off at similar points in the business cycle and whether some see greater income growth than others over such periods. The fact that the bottom 90 percent received such a larger share of gains in the 1950s expansions than it did in the past few ones looks depressing, but if they also suffered disproportionate losses in the recessions that followed, then we are looking at the wrong data points.
Looking at business cycles (grouping post-peak recessions with subsequent expansions, or analyzing trends “peak to peak”), Tcherneva’s chart looks very different. This chart, like Tcherneva’s original, is entirely from the Piketty-Saez data on tax units:
OK, lots to discuss here. First, the income gains of the bottom 90 percent look less impressive in the 1950s and 1960s when you take into account that they experienced a greater share of losses than the top 10 percent during those decades. The gains enjoyed by the top ten percent in the 1980s and 2000s are even bigger than before (note the scale of the chart has changed). The bottom 90 percent experienced more than half of the income losses between 2007 and 2012.
But the Piketty and Saez data shows the bottom 90 percent losing income not only from 2007 to 2012, but from 1979 to 1990 and from 2000 to 2007. This is, simply, inaccurate as a meaningful depiction of what happened to this group. We will get to that in due course, but this is a good time to check your priors—do you really think that the bottom 90 percent of Americans is poorer today than in 1979, as the Piketty and Saez data indicate?
In the rest of the charts, I will stick with business cycles and I will leave out the post-2007 (ongoing business cycle). I do this because both the bottom 90 percent and top 10 percent experienced losses from 2007 to 2010 and because the estimates for the share of income gains going to each group shift wildly depending on the measure used. That is because the overall decline is often not very large, which means the denominator in the “share of losses going to the top” fraction is small relative to the numerator, resulting in unstable estimates. Primarily, however, I omit years after 2007 because the current business cycle is not complete (and may not be complete for some time).
If we replace the post-1979 estimates with the CBO household-based figures, we get a different picture:
I’ve changed the scale for comparison with the charts to follow, but it should be clear that the share of gains enjoyed by the top is much lower than it was for tax units in the 1980s and from 2000 to 2007. The bottom 90 percent’s income rose in both of these periods, in contrast to the Piketty and Saez results. Using households rather than tax units recovers the Tcherneva finding of steadily rising shares of income gains going to the top. But the steady increase only starts in the 1960s rather than going back to the 1950s, if the earlier tax unit estimates are to be believed.
If there is a reason to worry about inequality of tax units rather than inequality of households, I can’t think of it. It’s certainly reasonable to want to look at inequality across individuals rather than households, but tax unit trends cannot be interpreted as showing what has happened to individual inequality either. A married couple is a tax unit, and the decline in marriage and increase in divorce mean that over time more and more tax units are individuals. Interpreting tax unit inequality trends as trends for individuals will miss the fact that more tax units were couples rather than individuals in the past. At the very least, if one wanted to look at tax-unit inequality, it would be important to exclude dependent children from the data.
But really, you shouldn’t focus too much on these charts either. We’ve got bigger problems to address.
Problem 3: The Baby Boomers
This is not a Gen-Xer’s whine about the shortcomings of his parents. The aging of the baby boomers causes endless interpretive errors when looking at inequality trends.
Let’s go back to the Piketty-Saez “tax return gross income” measure. Essentially, Piketty and Saez look at the adjusted gross income line on tax returns and add back the adjustments (the “A” in AGI). Gross income is a pre-tax, pre-transfer income measure. It doesn’t include income from government programs like unemployment insurance or—cough, cough—Social Security, and it doesn’t deduct income, payroll, or other taxes (or add in refundable credits for filers with no income tax liability).
There is nothing wrong, per se, with looking at inequality of pre-tax and -transfer income. However—and this is maybe the most overlooked problem with interpretations of the Piketty-Saez data—inequality trends that do not account for redistribution must be contextualized by addressing the fact that the U.S. has near-universal public pension and retiree health insurance systems. Social Security constitutes more than half the income of more than half of elderly Americans. The population of retirees and semi-retirees has grown as lifespans have increased, as our senior entitlements have become more generous, and as the baby boomers have aged.
That means that within the bottom 90 percent, the growing retiree population exerts a downward pull on market incomes because retirees rely heavily on Social Security and Medicare. If the retiree population were not growing, we would have stronger market income growth.3 Indeed, rising prosperity creates more retirees by allowing people to stop working earlier. Among the non-retiree population, market income growth is more robust than the Piketty-Saez data suggest.
In short, if retirees are grouped in with everyone else, the results give us limited information about how incomes are growing or how income inequality is changing. They tell us little about the health of the American economy, the adequacy of our economic policies, or the state of our politics.
There are two ways to address this interpretive issue. First, we can concede that if we want to look at the population as a whole, lumping the aged with the young, we have to add government transfers to our pre-tax and -transfer income measure. If we’re going to do that, we ought to also subtract the taxes people pay to (sort-of) fund those programs. The CBO data allow us to add in income from all of the major federal cash and noncash transfer programs and to subtract out federal income and payroll taxes.
Normally I would include the value of employer- and government-sponsored health insurance in my estimates of post-tax and -transfer income. To exclude them means that people are indifferent between having health insurance paid for by someone else or not having it. To be sure, just how to value health insurance as “income” is an enormous and not-settled puzzle, but assigning a value of $0 is surely not right. Here I sidestep this controversy by continuing to exclude the value of employer-sponsored health insurance and of Medicare and Medicaid. (I’ve also refrained in all of these charts from using the inflation adjustment that CBO and I prefer over Piketty and Saez’s.) What’s the new chart look like?
Again, I can only make this modifications to the income measure from 1979 to 2007 because the CBO data do not go further back. The Tcherneva estimates for the earlier periods, which I continue to show in the chart, might look different if we could look at the post-tax and -transfer income of households—or even of tax units—before 1979. But since the 1953-to-1979 period did not see rising income concentration, it is unclear that the bars would look any different with ideal data.
Compared with income trends using pre-tax and -transfer estimates, the post-tax and -transfer ones shift again. The top ten percent’s share of income gains is higher in the 1980s and substantially lower in the 1990s and pre-Great Recession 2000s. While the gains to the top are higher after 1979 than before, there is no clear increase after 1979.
The typical objection to accounting for taxes and transfers in income is that if government redistribution is doing all the work of income growth for the bottom and the middle class, that reflects poorly on the American economy. We want strong pre-tax and -transfer income growth. There is something to this argument. We might worry, for instance, that pre-tax and -transfer income inequality could indicate unequal opportunity.4
But in most contexts, it is more relevant to look at post-tax and -transfer income inequality. Many observers have an odd habit of pointing to pre-tax and –transfer income to demonstrate high levels of inequality that require redistribution. By that practice, one can imagine a world where we redistributed to eliminate disposable income inequality entirely, but where market income concentration would still be cited as evidence in favor of redistribution.
At any rate, as discussed, we shouldn’t look at trends in pre-tax and -transfer inequality unless we exclude retirees from the data. When observers argue that government redistribution is responsible for most of the rise in middle-class incomes—that the American economy is broken because growth in pre-tax and -transfer income below the top 10 percent is meager—they are misinterpreting the Piketty-Saez data.
What would Tcherneva’s chart look like if retirees were excluded? Ideally, we would look at the bottom 90 percent and top 10 percent of non-retiree households ranked by their pre-tax and -transfer incomes. The best we can do with the existing CBO data—brace yourself—is to consider the pre-tax and -transfer income of households headed by someone under age 65, ranked by pre-tax but post-transfer household income (after adjusting for household size). The people in the bottom 90 percent of pre-tax, post-transfer size-adjusted income are not identical to the people in the bottom 90 percent of unadjusted pre-tax and -transfer income, but the difference is likely to be unimportant. Trends in the income of both “bottom 90 percents” are likely to be extremely similar. (And to be clear, it is only the ranking that involves size-adjusted income—the averages in the CBO charts use unadjusted incomes.)5
Here’s the chart for the nonelderly using pre-tax and -transfer income:
Well, that looks an awful lot like the chart that includes the elderly (go check, I’ll wait). The only difference is that the 2000-07 period looks considerably worse. (The estimates look even worse than in the chart because the increase in overall income was small enough that when the increase at the top is divided by it, the resulting fraction equals 410 percent. This is an unfortunate feature of this share-of-gains measure.)
But are we done? Almost, intrepid reader….
Problem 4. Measuring the Capital Incomes of the Rich Is Really, Really Hard
It just is. Two big problems stand out. First, the way that the IRS data captures capital gains is problematic. Let’s say that someone sells Apple stock this year that they purchased twenty years ago. That person will realize capital gains and report them on her tax return. She will “receive” a lot of income in 2014. Consider a second person who bought Apple stock the same year but holds onto it. He receives no income from capital gains in 2014. But the way to account for the flow of purchasing power that Apple has bestowed on these two people is to recognize that both became richer year by year for two decades. Apart from the tax savings and transaction costs, it is no different than if they had sold their stock annually and then reinvested all of it (including the gains realized) back into Apple, year after year. Capital gains are properly considered income regardless of whether or not they are realized through the sale of an asset.
By counting capital gains accrued over many years as income in a single year, the Piketty/Saez data includes far too many very high capital incomes. Because realization of gains is often timed well relative to the performance of stock, housing, and other asset markets and because these markets have boomed in recent decades (prior to busting), the problem has worsened over time.
But it’s worse than that, because non-taxable capital gains are not counted as income in the IRS data even when realized. Non-taxable assets are particularly important to the bottom 90 percent—think home sales, which have capital gains exclusions so large that few people below the bottom 90 percent are liable for taxes. The other big form of saving below the top? Retirement vehicles, which are also tax preferred and which also are invisible in the IRS data until retirees start drawing down from them.
These problems argue for excluding taxable realized capital gains from the CBO pre-tax and -transfer income measure, which it is possible to do. But there are also substantial problems with what CBO calls “business income.” This includes income from self-employment and closely held businesses. Changes in tax policy over the last 35 years have created incentives for business owners to incorporate under Subchapter S of the relevant part of the tax code rather than Subchapter C. Subchapter C corporations pay corporate income tax, and their income shows up on corporate tax returns, not on individual income tax returns. This income is invisible in the Piketty and Saez data until owners sell their stock and realize capital gains. Subchapter S corporations pay out income to owners as dividends each year which are reported on individual income tax returns.
While in the end, all of the income from both Subchapter C and Subchapter S corporations may show up on tax returns, the shift to Subchapter S corporations has at the very least affected the timing of when this business income shows up on individual tax returns. There are visible breaks in the Piketty and Saez inequality series that correspond to changes in tax policy that made Subchapter S incorporation more attractive. Business income rose from 7 percent of pre-tax and -transfer income in the top one percent in 1986 to 16 percent in 1988, in part due to the Tax Reform Act of 1986 having lowered individual income tax rates below corporate ones, thereby making S corporations advantageous.
Given these issues and others, a strong case can be made for looking at trends in earnings inequality rather than at pre-tax and -transfer income more broadly. It’s likely that properly measured, capital incomes are less equally distributed than labor incomes (wealth is much less equally distributed). But that doesn’t mean that inequality of capital incomes has grown faster than inequality of earnings. There is at least some evidence for the propositions that capital income concentration and wealth concentration have increased less than earnings concentration has. If income concentration has been growing more generally, there is no reason to think that the trend in earnings concentration would understate that growth.
The penultimate chart below combines a series from the Piketty and Saez data with one from CBO. The Piketty-Saez estimates, through 1979, look at the wage and salary income of tax units; the CBO ones I have created as the wage and salary income of households with nonelderly heads. Once again, the estimates look very different from the Tcherneva chart, and they look very different from the chart on pre-tax and -transfer income of nonelderly households.
The 1950s and 1960s saw a high and largely stable share of wage and salary gains going to the bottom 90 percent. This pattern sharply reverses in the 1970s, and the share of wage and salary income going to the top 10 percent increases further in the 1980s. However, the 1980s were the high-water mark for lop-sided growth. In the 1990s, the share going to the top was lower than in the 1970s, and it fell again between 2000 and 2007, when the bottom 90 percent received more than half the gains.
I’ve stitched together Piketty-Saez and CBO series throughout this essay, and here I have attached a pre-1979 series for tax units with wages and salaries to a post-1979 series for households with a nonelderly head (some of whom don’t have wage or salary income). Maybe I’ve been selling you a bill of goods with these analyses. Anyone is welcome to download the Excel spreadsheet I’ve used for these estimates (though it was not created for easy public use and includes unlabeled sheets copied from Piketty/Saez and CBO spreadsheets—feel free to email me after working through it with questions). But you might save yourself some time. Because I can show you that last chart using only the Piketty-Saez data for all of the business cycles, so you can compare the two versions. Here it is—the money chart from the top of this essay again:
1: Technically, the chart shows the share of inflation-adjusted income gains that go to the top 10% and bottom 90% holding population size constant. In particular, the measure gives different estimates depending on the inflation adjustment used. Arguably, no inflation adjustment should be used, in which case the measure would give the share of nominal gains going to both groups, and the extent to which those gains were eroded by increases in the cost of living is a secondary issue. The measure also gives different estimates if population size is not held constant, and arguably, the extent to which the population grows is also a secondary issue after the share of gains is estimated.
2: The CBO averages I created for this chart start with CBO’s definition of “market income” and subtract out the value of employer-sponsored health care, the employer’s share of payroll taxes, and the corporate and excise taxes that CBO allocates to households as income. To do so, one uses tab 6 in the CBO spreadsheet to which I linked, which ranks people by their pre-tax, post-transfer incomes (adjusted for household size), then shows mean incomes by different sources for several quantiles of this person-level distribution. The means, however, are for households, not people, within a quantile, and they are not size-adjusted. The number of households in quantiles of the same size differs (while the number of people does not), but because CBO includes the number of households in each quantile, one can back into the total income of households within quantiles, aggregate the totals across quantiles, and then compute the mean income of households for the aggregated group. It is easy to subtract the means for the income components that are in CBO market income but not Piketty/Saez’s pre-tax and -transfer income before combining quantiles to get the means in each year for the top 10 percent and bottom 90 percent. The resulting estimates are not precisely what we want because the means are for households in the bottom 90 percent or top 10 percent of individuals’ size-adjusted household pre-tax/post-transfer income rather than for households in the bottom 90 percent (top 10 percent) of households’ unadjusted household pre-tax income. However, this difference is unlikely to be important; the “bottom 90 percent” will largely be the same households by either ranking, and the same is true for the “top 10 percent”.
3: And it is highly misleading to say that retirees “depend” on Social Security. In the absence of Social Security (and Medicare), Americans would save more and work longer, both of which would show up as more pre-tax and -transfer income among the bottom 90 percent of older adults and retirees than what we see today.
4: The uncertainty of that “could” ought to be acknowledged. A society can choose to compensate different positions very unequally yet have equal access to the positions.
5: Technically, the estimate for the bottom 90 percent is the non-adjusted pre-tax and -transfer income of the average household that has size-adjusted pre-tax-post-transfer income that falls in the bottom 90 percent of people ranked by their size-adjusted pre-tax-post-transfer household income. See why that’s not in the body?
Scott Winship is the Walter B. Wriston Fellow at the Manhattan Institute for Policy Research. You can follow him on Twitter here.
Interested in real economic insights? Want to stay ahead of the competition? Each weekday morning, e21 delivers a short email that includes e21 exclusive commentaries and the latest market news and updates from Washington. Sign up for the e21 Morning eBrief.