Richard Burkhauser on the Middle Class
What this covers
Richard Burkhauser of Cornell University joins Russ Roberts to examine how methodological choices about income measurement can produce radically different narratives about middle-class prosperity. Burkhauser shows that using identical data from the Current Population Survey over 1979–2007, measured median income growth ranges from 3.2 percent to 36.7 percent depending on how income is defined and which unit—tax filer or household—serves as the basis for analysis. The central claim is that narrower definitions used by some researchers, including Emmanuel Saez, capture only market income of tax units, while broader definitions that incorporate transfers, taxes, and in-kind benefits like employer health insurance yield a substantially different picture of whether the middle class has stagnated.
The discussion unfolds through concrete examples of how these definitional choices compound. A jump from 3.2 percent growth to 15.2 percent occurs simply by shifting from tax unit to household measurement, reflecting changes in household composition and filing patterns. Further gains appear when taxes and transfers are included, since the median household's tax burden actually fell over the period. Burkhauser introduces the technical apparatus of equivalence scales) to account for household size, and addresses Census Bureau changes in top-coding that contaminate inequality trends if unaccounted for. The conversation encompasses work incentive effects of transfer programs, skill-biased technological change as a driver of wage dispersion, and the role of employer health insurance—covering roughly 75 percent of workers—in total compensation. Burkhauser and Roberts also examine whether the measured decline of the middle class withstands scrutiny through distribution overlay analysis, finding that most households that left the middle income bracket moved upward rather than down.
Burkhauser argues that claims of middle-class stagnation rest on narrow methodological choices (tax units, market income only), and that under more appropriate assumptions—household sharing units, transfers, taxes, and in-kind benefits like health insurance—median income grew roughly 37% from 1979-2007 rather than stagnating.
- The same CPS data yields median income growth from 3.2% to 36.7% depending on definitional choices
- Piketty and Saez measure market income of tax units, a valid but different question than household economic well-being
- Transfers and in-kind compensation have grown materially since 1979 and excluding them distorts the picture
Median well-being rose far more than narrow income metrics show when adjusted for household composition, transfers, and fringe benefits.
- About 75% of US workers receive employer-provided health insurance (their own or via a working spouse's plan), making it a major component of compensation, yet it is not included in standard measures of wage earnings or income, which distorts comparisons of well-being over time.
“approximately 75% of all the workers in the United States either have their own employer providing them with compensation in health insurance, or they are the spouses of someone who is working and they are on their plan. So, this is a major part of the compensation program, and it is not included in measures of wage earnings or in income.”
- Measuring individual rather than shared income produces absurd results (e.g., a stay-at-home spouse and children counted as poor), so economic well-being must be measured at the level of the sharing unit—household, family, or tax unit—because people pool and share income within those units.
“if you simply looked at the personal income of people who live in households, you'd get very strange results... you'd end up if you used the individual for your analysis showing that the husband was making a lot of money and was in perfectly good shape... The wife would have no income of her own and therefore would be counted as having, as being poor.”
- Because larger households economize (shared heat, bulk purchases, shared bedrooms), per-capita division overstates need; dividing income by the square root of household size (the 0.5 power) is a reasonable adjustment, meaning a family of four needs exactly twice the income of one person to reach the same well-being—approximating how US poverty lines scale.
“the formula people usually use is take the income, divide it by the number of people, and then do it to a power... they move it to the value of 0.5... a family of four needs exactly twice as much as a family of one to be at the same level of economic well-being.”
- Using the same CPS data over 1979-2007, median income growth rises from 3.2% (tax unit, pre-tax pre-transfer market income) to 15.2% (household unit), to 23.6% (household size-adjusted pre-tax post-transfer), to 29.3% (adding taxes), to 36.7% (adding the value of employer health insurance, Medicare and Medicaid)—a more than tenfold difference driven purely by definitional assumptions.
“The median income of these tax units has only increased by 3.2% over the entire 30 years.”
- Median income growth rises when including taxes (from 23.6% to 29.3%) because the median household paid a smaller share of its income in taxes over the period—a counterintuitive result where subtracting taxes raises measured growth because the tax burden at the median declined.
“How can you get median income growing faster when you are subtracting things out? And the answer is that we've actually been paying, at the median anyway, a smaller share of our income into taxes; and that's why you get the growth is 29.3% rather than 23.6%.”
Census and tax measurement changes, not real income shifts, account for apparent inequality increases.
- Between 1992 and 1993 the Census Bureau changed its ability to capture exotic top-1% incomes, and in 1995-96 switched from top-coding to cell-means; researchers who don't adjust for these definitional changes falsely conclude inequality jumped, when the apparent jump merely reflects improved or altered measurement.
“if people look at the CPS data and don't recognize that in 1993 we suddenly were able to better capture income, they will get the false notion that income inequality increased between 1992 and 1993. It didn't. It just was in our ability to capture that income.”
- Contrary to prior belief that the CPS cannot capture the tails of the income distribution, Burkhauser's team showed—using internal CPS data to get around top-coding—that the CPS can closely replicate Piketty and Saez's IRS-based levels and trends, validating the CPS for top-income inequality analysis.
“We can actually get very close to the levels and trends that they get in the IRS data using the CPS data. That's what took a long time.”
- The 1986 tax reform made the top personal income tax rate lower than the corporate rate for the first time, causing high earners to record income as personal market income rather than retain it as distributed profits in partnerships and businesses, producing a recorded uptick in the top 1% share that reflects relabeling, not real income gains.
“in the 1986 reforms, for the first time made the highest personal income tax rate lower than the corporate rate. And when that happened, not surprisingly, there was a dramatic increase in the amount of market income recorded on people's taxes.”
- The jump from 3.2% (tax unit) to 15.2% (household) growth occurs because the number of tax units per household rose over the 30-year period—more people living together while filing separately—plus Piketty and Saez's assumption that any unmarried person over 20 is a separate tax unit inflates the low-income left tail and pulls down the median.
“that happens because the number of tax units per household has been rising over that 30-year period. There are a lot more people living together and sharing everything except a marriage certificate.”
Popular inequality narratives misrepresent actual distributional outcomes by selecting inappropriate metrics or ignoring transfer systems.
- Two sets of smart researchers (Burkhauser's team vs Piketty and Saez) reached wildly different conclusions about the middle class not because of bad data or math errors, but because they were fundamentally asking different questions while using nearly identical words to describe their findings.
“It wasn't the data. It wasn't that we were making mathematical errors. It was that we were fundamentally asking different questions, but using almost the same words to describe our findings.”
- The median person has higher income than the median household (about $60,000 vs $50,000) because households with more income tend to contain more people, so ranking all persons by their household income pulls the median upward relative to ranking households.
“the median person has higher income than the median household does... It's because households with more income also have more people... median income is about $50,000; but the median person in those households has income of about $60,000.”
- When you overlay the 1979 and 1989 household income distributions, the middle of the distribution did shrink and 'disappear,' but about 90% of those who left the middle did so by becoming richer, not poorer—so the popular narrative of a vanishing middle class misrepresents what actually happened.
“we actually could see that you ended up with a very small portion of the tail that got poorer, about 10%, and of the mass that you pushed down from the middle, moved to the left and became a little poorer. But 90% became richer.”
- Claiming the US has returned to a 1920s 'Gilded Era' because the top 1% market-income share is comparable ignores that there was no Social Security system then; today's elderly with little market income are sustained by transfers, so comparing only market income across these eras completely misuses the data to imply equivalent well-being inequality.
“In the 1920s, there was no Social Security System. If you weren't earning market income, you were in bad shape. We do now have a Social Security System, and all the people over the age of 65 who have no market income are not starving in the streets. But that transfer income is not included in that comparison.”
- Single mothers are the one group whose market income dramatically increased since 1996, because the shift from AFDC (which discouraged work) to TANF (time-limited benefits plus earned income tax credits that subsidize work) changed incentives to reward employment rather than leisure.
“the one group whose market income has dramatically increased since 1996 are single mothers... we went from the Aid to Families with Dependent Children (AFDC) program, which completely discouraged work, to the new TANF program... if you work, we will, even if it's in a low wage job, we'll provide you with earned income tax credits which will supplement and subsidize work rather than leisure.”
Rising market inequality reflects skill-biased technological change and globalization, not predatory top-earner behavior.
- Rising market inequality is driven primarily by skill-biased technological change and globalization: the return to education has risen even as more people get educated, while unskilled workers now compete with unskilled labor worldwide, so the appropriate response is investment in education rather than blaming a 'greedy top 1%.'
“we're having technological change which is skill-biased, and if you have skills you actually are doing pretty well. The return on education has been rising even though the number of people going to school and getting higher educations has been rising. But if you don't have skills today you are really competing with the unskilled labor around the world.”
- Markets function like mirrors that reveal how the real world is actually working; you can either deny what they show or adjust your practices to do better within that reality.
“I think markets work--markets are like mirrors; they show you the way the real world is working. And you can either deny that or you can adjust your practices so you are able to do better in that world.”
- It is appropriate to transfer income to low-skill, low-income, and poor people, but it should be done in a way that preserves productive incentives—not by simply raising marginal tax rates on the top 1%, which would impair productivity and growth ('the golden goose').
“I'm absolutely in favor of transferring income to those folks; but I want to do it in a way that doesn't kill the golden goose that allows us to be productive--and that is by allowing people to get a fair return on their investments.”