In recent weeks, a major debate has gripped the field of economics, one with significant policy stakes. In one sense, the debate is about the nature and extent of inequality in the US: are the rich really taking more of the pie than ever before? Has their share of income been growing very fast, or gradually? Is inequality the defining challenge of our time, or simply one among many?

In another, perhaps more accurate, sense, the debate is about how to interpret IRS audit studies and assign retirement income across time.

Lots of disputes in the social sciences are like this: touching on big, societally important questions, but ultimately turning on disagreements over highly technical issues. And so it is with the great inequality battle of 2023.

On one side are Thomas Piketty, Emmanuel Saez, and Gabriel Zucman, three French-born economists who’ve pioneered the use of tax data in the US and elsewhere to track how the income of rich individuals has grown over time. They are among the most famous and successful people in their field; Piketty’s 2013 book Capital in the Twenty-First Century was a surprise worldwide bestseller, and Saez and Zucman each won the John Bates Clark Medal, an honor that rivals the Nobel in its prestige within the profession.

On the other side are Gerald Auten and David Splinter. Both are well-respected if less publicly known government economists, the former at the Treasury Department and the latter at the Joint Committee on Taxation. The two camps have been arguing over inequality for at least half a decade now, but with the Journal of Political Economy, one of the field’s most prestigious journals, agreeing to publish Auten and Splinter’s latest paper, the debate has come to a head.

The backstory: Piketty, Saez, and Zucman have, in a series of papers going back to 2003, documented a large and ongoing increase in the share of income in the US going to the top 1 percent, and even larger increases in the share going to the top 0.1 or 0.01 or even 0.001 percent. Inequality, according to their data, is rising, and it’s rising faster the higher up the income scale you go. Their data helped popularize the notion that the 1 percent is pulling away from the 99 percent beneath them; it’s hard for me to imagine the phrase “we are the 99 percent” becoming a mainstream political slogan without this research preceding it.

Importantly, Auten and Splinter agree that inequality has increased. But they are less certain that the top 1 percent’s share, or that of the top 10 or 0.1 percent, has increased. (Notably, this is a different question from whether inequality overall has grown, because income gaps can also show up within the bottom 90 percent.) The different perspectives are summed up in the latter chart, comparing how the two groups estimate the top 1 percent’s share of income as increasing over time:

Piketty, Saez, and Zucman estimate that the top 1 percent earned 9.1 percent of national income in 1960, rising dramatically to 15.1 percent by 2019. Auten and Splinter, by contrast, show a very small increase: 8.1 percent in 1960 to 8.8 in 2019. The share is now actually lower, they find, than it was in the mid-1960s.

Auten and Splinter reach similar conclusions about the top 10 and top 0.1 percent: per their estimates, the latter grew from 29.2 to 29.7 percent from 1960 to 2019, the latter from 2.5 to 3 percent.

How can two research teams, both looking directly at US tax data, reach such different conclusions? It’s not that one group are craven “inequality deniers,” to borrow an ugly term that Piketty has slurred Auten and Splinter with, or that the other group is “thoroughly discredited,” as billionaire libertarian economist Cliff Asness has described Piketty, Saez, and Zucman. The disagreement is, at root, about how to deal with the limitations of tax data, by far the best source of information on who earns what in America.

Neither side has a monopoly on truth. On some issues, Auten and Splinter made judgment calls that seem more reasonable to me; on others, Piketty, Saez, and Zucman do; on still more, it’s wildly unclear which assumptions are most appropriate to make.

At the same time, it’s clear — including in Auten and Splinter’s data — that inequality has increased in the US in the past half-century, despite some headlines suggesting they’ve explained away the increase in inequality altogether. The question, instead, is how big a problem that increase is — and the answer to that has less to do with economic models than it does with politics.

A long, romantic story of IRS tax data

To understand the current debate, you have to first go back to 2003. Beyoncé had just gone solo, we all were learning what “yellowcake” uranium was, and Piketty and Saez — then two relatively junior scholars — published their first dataset on top incomes in the US in the Quarterly Journal of Economics.

This was a big deal. While various researchers in the 1990s, such as Larry Katz, Kevin Murphy, and David Autor, had used survey evidence to show that inequality was increasing, Piketty and Saez used IRS information on hundreds of millions of taxpayers — a data set that was both far wider and more reliable than survey responses.

For all its advances, the IRS data had limitations, many of which Piketty and Saez acknowledged in their first paper. Some forms of income are not taxed: the average employer spends over $15,000 per year on each employee they offer health insurance, and none of that money is subject to personal income taxes. Nor are most government benefits, like food stamps or Medicaid, or retirement income from a Roth account, which is untaxed. None of that will show up as income on a tax return, but it’s still very real income that people enjoy.

These limitations led to two decades of efforts from Piketty, Saez, and collaborators like their protégé Zucman to develop more comprehensive datasets that could give a full picture of income inequality in the US (and around the world). This culminated in what the team calls “distributional national accounts,” an attempt to break down countries’ full GDP and determine where each dollar of it goes: how much to the rich, the poor, the top 1 or 0.1 percent, etc. This way, nothing gets left out: every dollar of national income, from health benefits to government programs to money retained by corporations, gets accounted for and assigned to an actual person.

But how, exactly, to assign all that income is not obvious. It’s reasonably clear who gets income from reported wages or health benefits (the worker getting them!) but in 2019 there was also some $604.3 billion in “underreported” income in the US which showed up in GDP, but not in tax returns. The figures are similar for other years. To whom should we assign that (often illegally) unreported income?

Or take government spending. Food stamps are income to the person receiving them. But how do you assign the benefits of, say, Yellowstone National Park? Of a fighter jet? What about government deficits? They’re part of GDP, and represent a loss of income to someone in the future, either through higher taxes or spending cuts; how do you include that?

These kinds of questions, about how to assign income whose beneficiary is not clear in the data, are the ones around which the Piketty/Saez/Zucman (PSZ) versus Auten/Splinter (AS) debate revolves.

Death by a thousand imputations

The difference between the PSZ and AS data is not explainable by just one disagreement. Several different adjustments play a role, as summarized by this table:

Auten and Splinter find that the rise in the top 1 percent’s share of income from 1962 to 2019 was 4.3 points smaller than the rise in Piketty/Saez/Zucman. No one factor explains more than 1.6 points of this disagreement, and even those are offset by a couple of areas (like dealing with owner-occupied housing and Social Security) where Auten and Splinter make decisions that show a larger increase in the top 1 percent’s share than PSZ.

You can read Auten and Splinter’s latest paper for a rundown on all these differences, but for now I’ll focus on the two largest: their treatment of income underreporting, and their treatment of retirement income.

Underreporting

Piketty, Saez, and Zucman assume that each type of unreported income is distributed identically to that type of reported income; in practice, business income is the type that accounts for the bulk of underreporting, as it’s much easier to fudge than, say, wages or dividends. On the surface, the PSZ approach seems reasonable: if the top 1 percent gets 15 percent of reported business income, say, they also get 15 percent of unreported business income.

Auten and Splinter counter that we’re defining the top 1 percent here as the group reporting the most income to the IRS. Does it really make sense that rich people who are squirreling away tons of money and not paying taxes on it are at the same time reporting tons of other income to the IRS, such that they show up as rich in these numbers?

Or are we looking for people who might report very little income to the IRS, and shielding the vast majority of it? This latter group isn’t, in IRS-reported terms, the “top 1 percent.” It could even be the bottom 20 percent. Donald Trump, for instance, reported negative income for several years in the 2010s due to business losses. He would not show up as in the top 1 percent in the tax data — is it reasonable to assume that because he’s at the bottom of the income distribution, he’s not hiding much money? (It is not.)

So Auten and Splinter’s approach is to rely on the results of random IRS audits, which can estimate the level of tax avoidance discovered by auditors by taxpayers with different levels of reported income. They then use this audit data to estimate underreporting rates across the income scale, and allocate unreported income to taxpayers accordingly. This leads to less of the underreported income going to people with the highest reported incomes than the method that Piketty, Saez, and Zucman use.

Saez and Zucman, in a response paper, argue that Auten and Splinter are not using the audit data appropriately, and assign less unreported income to the rich than the audit data suggests; in a reply to that reply, Splinter argues that he and Auten use the data perfectly well and that their ultimate conclusions match the audit results. Still another argument, from Brookings Institution researchers William Gale, John Sabelhaus, and Samuel Thorpe, is that the inability of the IRS to detect sophisticated tax evasion should lead audit studies to underestimate how much the rich are evading their taxes, which would be a point in Piketty/Saez/Zucman’s favor. (Splinter, naturally, has a reply to this as well.)

No approach here is perfect, but to the best of my understanding, Auten and Splinter’s approach makes more sense than assuming that people who report the most income also evade the most. But this is an area where subsequent research, especially from an IRS that just got a burst of new funding and can better tackle trickier tax evasion situations, could be very productive.

Roll over, 401k!

The second biggest contributor to the gap between the teams is their treatment of untaxed retirement income. Auten and Splinter estimate that the top 1 percent gets about 6 percent of income from retirement accounts; PSZ’s estimate is more than twice that. This disparity is mostly because the latter group includes nontaxable money coming out of retirement accounts in their calculations. In some cases, that makes sense: if you have a Roth IRA, then the money you take out once you’ve reached the minimum age will be tax-free, as you’ve already paid taxes when you contributed the money.

But Auten and Splinter argue that the vast majority of untaxed retirement account distributions are just rollovers: money being transferred from one retirement account to another. When I left the Washington Post for Vox, I moved my 401(k) money from the Post into an IRA account. This shows up in the data as income earned the year of that transfer, but it wasn’t income at all; it was just wealth being shuffled from one account to another. It should be excluded from income data.

To their credit, Saez and Zucman have conceded this point and adopted an adjustment where they treat, at most, 10 percent of retirement income in any given year as non-taxable. Splinter has replied that this adjustment doesn’t go nearly far enough, and the share should be much lower than 10 percent, which would imply less retirement money going to the top 1 percent.

Inequality has grown — the question is how much

The inequality dispute has not bubbled up to the general public because the public has strong views on the proper treatment of 401(k) rollovers in academic economics studies. It’s become hotly disputed because of the perceived stakes for our society: Is inequality growing or not? Is it growing a lot or not?

I would submit, without denigrating in any way the tremendous intellectual effort each team has put into this project, that the gap between the two, in terms of the “big picture” they paint about American society, is not too large.

Both teams show that income inequality has increased in the US in the past half-century. While Auten and Splinter show at most modest increases in the top 1 percent’s share of income after taxes, they do find substantial increases before taxes. Moreover, that’s only one way to measure inequality. The most comprehensive measure is something called the Gini coefficient, which attempts to summarize the whole scale of inequality across the income spectrum, not just at the very top. An increase in the coefficient means that inequality has risen.

Before taxes and government safety net programs, Auten and Splinter estimate that the Gini coefficient for the US grew 25 percent since 1962 and 23 percent since 1979. After taxes and transfer programs, the increases are 10 percent since 1962 and 16 percent since 1979. More progressive tax and spending policy reversed some of the increase, but the increase is real and significant.

To give a sense of the scale here, the after-tax/transfer figure grew from 0.355 in 1979 to 0.417 in 2019, a 0.062 increase. If that number means nothing to you (it means nothing to me!) then consider that this increase is akin to the gap between the US and countries like Australia, Spain, and Switzerland today. That is, if this increase in inequality hadn’t happened, the US would be close to many other rich countries as opposed to the rather unequal outlier it is today.

You can also see signs of a general increase in inequality across some other measures. Take wealth, for instance. A recent rigorous attempt by economists Matt Smith, Owen Zidar, and Eric Zwick to track wealth inequality (that is, the gap in net worth, rather than income, between rich and power) in the US over recent decades shows a pronounced increase:

So it’s not reasonable to conclude that Auten and Splinter are engaged in “inequality denial.” They are quite literally documenting that it has increased, if not to the degree that Piketty and company claim.

What, then, are the stakes? One is how social science is conducted — while at times the disputants have accepted critiques from the other side, too often the debate has degraded into namecalling and impugning of motives. But another, more substantive stake is about where inequality ranks among society’s ills. There’s no disputing that inequality across a number of dimensions has increased in recent decades. But that’s a different claim from saying that this is a New Gilded Age, an “age of inequality” where inequality is the “defining challenge of our time.”

What Auten and Splinter’s data suggests is that inequality is one challenge among many, which may have important implications for policy, often in surprising ways. It may suggest, for instance, that a policy that makes fossil fuel-powered electricity more expensive, worsening inequality by increasing energy costs more for low-income people, is more worthwhile: inequality is not at such a crisis point where we need to avoid undertaking policies that fight global warming out of concern about inequality.

Insofar as I come out of the debate with any firm conviction, it is that holding firm convictions about the state of inequality is probably a mistake. Measuring this stuff is hard, much harder than it appears at first glance. Simple approaches, like just looking at reported taxable income, can be misleading, as can focusing on how much Jeff Bezos or Elon Musk made in the stock market. It’s easy to make errors or leave aspects of the work incomplete. We have to exist in a state of frustrated uncertainty.