In a recent paper published by the Cato Institute, I made the seemingly heretical observation that inequality in incomes, wages, consumption, and wealth among the U.S. population as a whole does not appear to have increased significantly since 1988. My Cato paper was condensed and revised from a paper I presented at the Western Economics Association (WEA) last July. That WEA paper, in turn, was condensed and revised from Chapter 5 in my book Income and Wealth, written early last year as part of a series of college textbooks for Greenwood Press.

Income and Wealth did not say that U.S. inequality probably stopped rising after 1988. That thesis arrived much later, and tentatively, in two pages of the Cato paper. My book’s tutorial is about avoiding common statistical pitfalls involved with measuring the growth of average wages and incomes over time, the division of income between labor and capital, the concepts of mobility and lifetime incomes, the concentration of wealth ownership, etc. Early chapters explain why inequality might be expected to increase over time, because I too had initially assumed the “professional consensus” asserting an ongoing increase in inequality was an adequate substitute for facts.

I have been bemused by the emotional, non-factual responses at various blogs to my inability to discover any evidence of a “significant and sustained” rise of inequality since 1988. By “significant” I mean to exclude the .003 increase in the Gini index for pretax money income between 2001 and 2005 that Brookings’ Gary Burtless mentioned. Oxford University economist Anthony Atkinson suggests the Gini coefficient must rise by 3 percentage points in order to be considered a significant new trend. By “sustained,” I mean to exclude the tax-induced spike in capital gains realizations in 1986, the 1993 break in Census Bureau data, and the Internet stock euphoria of 1997-2000.

Surveying the Evidence

My heretical claim did not originate with me. My Cato paper cited the following evidence:

1. Card and DiNardo found that wage inequality did not increase from 1988 to 2000.
2. Johnson, Smeeding and Torrey found that a Gini coefficient for a broad measure of consumption, from the Survey of Consumer Finances, fell slightly from 0.283 in 1986 to 0.280 in 2002.
3. Kopczuk and Saez found that the top shares of wealth were stable during the 1990s. The top 1 percent’s share of wealth was also stable from 1995 to 2004, according to more recent estimates from Arthur Kennickell.

Unless these scholars have erred, that takes care of wages, wealth, and consumption. And it does so without even mentioning the Consumer Expenditure Survey, which I am supposedly enamored of, according to Gary Burtless.

What about disposable (after-tax) income? Last November, a study by Burkhauser, Oshio and Rovba found that, “For the United States … real mean household size-adjusted after-tax income increased by 10.93 percent over the 1980s and by 7.27 percent over the 1990s while median after-tax income increased by 5.95 percent and 7.10 percent respectively over these periods… [But] income inequality rose substantially over the business cycle of the 1980s whether measured by the 90/10 ratio (23.67 percent) or by the Gini coefficient (14.17 percent). In contrast, income inequality fell over the 1990s business cycle [1989 to 2000] whether measured by the 90/10 ratio (-6.82 percent) or the Gini coefficient (-2.24 percent).” From 1989 to 2000, they added, “the confluence of significant economic growth and work-based welfare reforms dramatically improved the employment and economic well-being of single women with children relative to the rest of the population and more generally did so for lower-skilled workers.”[1]

The aforementioned 90/10 ratios, incidentally, compared incomes between the highest and lowest deciles of full-time workers. The ratio was 6.9% in 1987-90 and 6.7% in 2003-2004, which lends support to the Card-DiNardo finding that wage inequality did not increase even between polar extremes. I have discovered great resistance to this message, particularly from conservative economists. One reason appears to be a common misunderstanding about the assumed link between incomes and work.

CEA Chairman Edward P. Lazear says, “There is little doubt that there has been a 25-year trend of a growing gap, sometimes called income inequality, between the wages of the skilled and the unskilled.”

Despite the absence of doubt, Lazear’s assertion is clearly inconsistent with the research of Burkhauser, Oshio and Rovba, Card and DiNardo and others. But I am far more troubled by his conceptual muddling of skill gaps and inequality. An accompanying graph does not clearly show median hourly earnings for college grads growing faster than those with less schooling from 1989 to 1996 or from 2001 to 2004. But even if there was such a sustained trend, a growing gap between wages of college graduates and high-school dropouts would necessarily translate into increased income inequality only if: (1) everyone worked full-time all year, and (2) all income came from work, and (3) college graduates were not a rising share of the labor force and native high-school dropouts a declining share.

The number of people who worked full-time all year in 2005 amounted to only 3.2 million in the poorest fifth of households but 16.7 million in the top quintile. There are many more singles in the lower income groups (including students and widows), and many more mature two-earner couples at the top. The top quintile – every couple with a pretax income above $91,705 – accounted for 29.1 percent of all full-time, year-round workers, which is the largest single reason they received 40% of disposable income in 2004 [xls]. The top quintile also had more college grads, but that only affects market income if they work.[2]

Economists who try to explain most difference in household income by hourly wages are ignoring the huge differences in the number of hours worked per household. They are also ignoring transfer payments, most of which (except the EITC) are effectively limited to those who work little or not at all.

Other economists attempt to dismiss the Card-DiNardo conclusion by referring to the top 1 percent’s share of W2 income. First, let me make a general objection. We will never discover anything about the distribution of income among all or even most Americans by looking at only 1 percent of the tax-paying population. We gather no information about what is happening to living standards of the poor or the (typically undefined) “middle class” by examining only the upper tail of the income distribution. Pretending to describe the income distribution by using only the share of income going to the top 1% makes no more sense than doing the same with the incomes of the bottom 1%. Except under extreme zero-sum reasoning (arguing that an extra billion for Steve Jobs means a billion less for others), the top 1% fetish is essentially irrelevant.

Even if you insist on focusing on the top 1%, it is particularly misleading to compare – as Piketty and Saez do in their widely cited study – income reported on tax returns in two years (1980 and 2004) that fall before and after the 1986 Tax Reform Act. In the Piketty-Saez estimates of W2 labor income, the top centile’s share suddenly increased from 7.3% in 1986 to 9.4% in 1988 and then averaged 9.1% from 1988 through 1996. That is quite consistent with the reasonable estimates of elasticity of taxable income cited in my December 14th Wall Street Journal op-ed, and not with Piketty and Saez’s subsequent disavowal of Saez’s own estimates. Using the Social Security data that Burtless cites, Schwabish finds [pdf] “the share of earnings at the top of the distribution … has fallen precipitously” from 2000 to 2003. My Cato paper showed that is also true of CEO pay. Both rebounded in 2004 because more than 11% of SCF respondents reported receiving stock options in 2001 [pdf] — reported on W2’s when exercised, and first exercisable in 2004. That, too, is a predictable response to lower tax rates, and executives accounted for only a fourth of option grants by 1999 [pdf].

Statistical Mirages

My Cato paper compared the 20.7% real increase in top decile pay from 1979 to 2004 with the 21% increase in bottom quintile pay, using median income from the Survey of Consumer Finances. Not interesting? Critics prefer to focus on defects in Census Bureau data which, Burtless suggests, I am either unaware of or reluctant to reveal.

For example, my Cato paper compared the Census Bureau estimate of the income share of the top 5% with that of Piketty and Saez. The Census figure rose from 18% in 1986 to 20.9% in 2004, mostly because of a data break in 1993. The comparable Piketty-Saez figure jumped from 22.6% in 1986 to 27% in 1988, which I attribute to the 1986 Tax Reform and so did Piketty and Saez until recently. Their measure of the top 5 percent’s share of income (as they define it) hit 31.2% by 2004. Paul Krugman and others were a bit too quick to blame sample size and top coding of Census data. Even when we exclude all income above $5 million from the Piketty-Saez estimate of the top 5 percent’s share, that would narrow the gap by only 0.9% leaving nine percentage points unexplained. The fact that Census includes cash transfer payments in total income is also not nearly enough to begin to fill the gap.

My Cato paper also cited the Census Bureau’s Gini coefficient for their broad (14th) definition of disposable income, which subtracts income and payroll taxes, adds cash and in-kind transfer payments, and (unfortunately) also adds capital gains that show up on tax returns. This is the measure Burtless approves of yet does not disclose.

The Gini coefficient for disposable income can be rounded to 0.38 or 0.39 for all but one year between 1984 and 1992, meaning it was essentially unchanged. It briefly jumped to 0.41 in 1986 for just one year, but that was clearly due to a rush to sell assets before the capital gains tax went up. The index was identical in 1985, 1988 and 1992, at 0.385.

In 1993, when the Census survey methods were computerized and “top-coding” limits hugely increased, the Gini suddenly leaps to 0.40 in 1993 where it remained in 2004, following a brief rise to .41 when capital gains boomed in 1999-2000. In order to pretend to see any sustained upward trend in the Gini for disposable income from 1984 to 2004, one would have to argue that it happened in one year – 1993. That’s where the Census Bureau critics begin to stumble.

Burtless writes:

The Census Bureau questionnaire does not provide accurate or consistent assessments of the incomes of the top 2% or 2½ % of income recipients. One reason is that respondents’ answers are top-coded, or at least they were in the not-too-distant past. Another is that the sample of high-income recipients is too small to give an accurate or consistent estimate of the incomes of the very top income recipients, say, those with incomes above $750,000 a year. This means the Census Bureau probably gives us an underestimate of true income inequality every single year, no matter which concept of income we choose to measure.

Burtless claims the CBO misses a lot of income among those with more than $750,000 a year, which he seems to equate with the top 2% or 2 1/2%. But $750,000 is only slightly shy of the$1.1 million threshold defining the top one-tenth of one percent in Piketty and Saez. It is miles above the $266,800 threshold defining the top 1% in CBO’s definition of income which includes, absurdly 59.4% of corporate profits.

Sampling errors are random, so even if one believes 1080 is too small a sample for the top 2%, that cannot tell us whether such incomes will be overestimated or underestimated. It certainly does not imply “an underestimate of true income inequality every single year.” At worst, the estimate of top 2% incomes would be too large some of the time and too small at other times, which is presumably what Mr. Burtless means by “consistent.”

The core of the CBO and Piketty-Saez estimates relies on a sample of adjusted gross income from the Statistics of Income (SOI) division of the IRS. That sample is twice as large as the sample of 54,000 in the Current Population Survey (CPS) Annual Social and Economic Supplement. But the Census Bureau surveys everyone, while the SOI survey just covers those who file a tax return. The SOI sample oversamples incomes above $5 million, but excludes a few trillion dollars of lesser incomes mainly because of tax avoidance (the $1 trillion AGI gap) and legal nonfilers. The net result it to exaggerate the ratio of highest incomes to total income.

There is no top-coding of total income, contrary to what Piketty and Saez and Paul Krugman have implied, but there is top-coding of specific types of income in the public use data to protect respondents’ privacy. When the Census Bureau calculates the income share of the top 5%, they use internal data that is free of top-coding, but there remain what Census officials call “internal processing limits” on, for example, the maximum salary recorded for any one job.

Edward J. Welniak, Chief of the Income Statistics Branch of Census Bureau, explains: In 1979, the questionnaire allowed the recording of up to $99,999 for 23 income sources. In 1985, the limit for recording earnings from longest job increased to $299,999. The final recording limit increase occurred in 1993 when each of the four earned income sources allowed the recording of amounts to $9,999,999.[3] The post-1993 limits obviously pose no problem for total incomes above $750,000, much less for average incomes of the top 1% or top 5%. Welniak found that in 1999 there were no more than 26 cases excluded out of a sample of 54,000. He also found, as have Burkhauser and others, that the increases in internal processing limits have caused the increase in inequality over time to be overstated, not understated.

Welniak notes that public use data caused “overstatement of income inequality growth over the 1967-2001 period,” because the tighter restrictions in the past make it look as though inequality was lower than it really was, while the easing of top-coding limits in the 1990s looks like increased inequality (rather than just increased visibility of high-income data). In Welniak’s words, “The larger growth in income inequality using public-use data is the result of: 1) topcoded income in 1967 which reduced measured income inequality and 2) increased high income through the plugging of mean topcoded values beginning in 1996.”

The pregnant remark from Burtless about respondents’ answers being top-coded “in the not-too-distant past” is a major reason why I argue that Census Bureau Gini coefficients, or top income shares, do not show that inequality has increased since the late 1980s. Welniak calculates that the 1985 change in methodology caused “a slight increase” in measured inequality, when none really occurred. Since the Gini coefficients from 1979 to 1984 were understated, that means the apparent increase in Gini coefficients in the late eighties was overstated. In 1993, Welniak notes that “CPS ASEC introduced computer-assisted personal interviewing and increased the recording levels for earnings to $1 million as well as increasing the recoding levels for other income sources.”

When Burtless refers to 0.469 as “the highest Gini coefficient ever recorded,” he is technically correct that it is higher than 0.466 in 2001, but he is not correct to compare it with pre-1993 data. As the Economic Policy Institute has explained, “a change in survey methodology in 1993 led to a sharp rise in measured inequality.”[4] The dramatic 1993 increases in the amounts of income recorded, and the substitution of computers for pencils, caused a spurious jump in inequality measures.

Before 1993, in that not-so-distant past, some limits on the amount of certain types of income caused Gini coefficients to be understated, underestimating actual inequality and therefore creating a spurious increase when comparing years before and after 1993.

It is baffling why Paul Krugman, Piketty and Saez and Burtless all thought sample size and “top-coding” was some sort of refutation of my observation that income inequality has not increased since the late eighties. Sample size could go either way, and changes in top-coding points in the exact opposite direction of what they are trying to imply.

Conclusion

For reasons that were partly explained in my book and further explained in a forthcoming academic paper, not one of the four data sets that attempt to estimate income from a sample of tax returns are credible for comparing income share across periods of changing tax rates. CBO data is arguably the worst of the four, and Piketty and Saez is the best. Even W2 data can’t be compared across different tax regimes because (1) the timing of nonqualified stock options exercises is highly sensitive to stock prices and because (2) switching to restricted stock — which does not appear on the W2 — is highly sensitive to the tax rates on dividends and capital gains. Ask Bill Gates or Steve Jobs.

If there were any better data showing a significant and sustained increase in the inequality of disposable income, consumption, wages, or wealth since 1988, I suspect someone would have shared it with us by now.

Notes

[1] Richard V. Burkhauser, Takashi Oshio and Ludmila Rovba, “How the Distribution of After-Tax Income Changed Over the 1990s Business Cycle: A Comparison of the United States, Great Britain, Germany and Japan,” November 2006. http://www.mrrc.isr.umich.edu/publications/index_abstract.cfm?ptid=1&pid=463

[2] U.S. Census Bureau, Current Population Survey, 2004 and 2005 Annual Social and Economic Supplements.

[3] Edward J. Welniak, “Measuring Household Income Inequality Using the CPS,” James Dalton, J. & Kilss B., eds., Special Studies in Federal Tax Statistics: 2003, Select Papers Given at the Annual Meeting of the American Statistical Association. http://www.irs.gov/pub/irs-soi/03preprt.pdf

[4] Lawrence Mishel, et. al., The State of Working America 2004/2005 (Washington: Economic Policy Institute, January 2005), p. 67, http://www.epinet.org/content.cfm/books_swa2004.

Alan Reynolds poses a straightforward question: “Has American inequality really increased?” Based on my reading of the evidence, including his recent paper, article, and op-ed pieces, my answer is “Yes, inequality has increased.” I would guess, based on the tone of his writing, this is not the take-away conclusion he hopes to hear.

Reynolds is skeptical there is any clear evidence showing inequality has risen since the late 1980s. He offers a number of reasons for skepticism. Most of them boil down to this: The many data series that show income inequality has risen are not worthy of our trust, whereas the series that show very little trend should be accepted at face value.

Like many students of the income distribution, I take seriously some of Reynolds’s criticisms of the data on income disparities. No single data source is perfect, and a couple of them have serious flaws. An unwary user can draw misleading conclusions if the data problems are ignored. Reynolds points to some serious problems, and in many cases fair-minded experts will agree with him.

The problem is, he is strongly critical of data series that do not support his views, while he is usually silent about equal or more serious problems with data sets that show little change in inequality. Rather than do the hard work needed to measure the effect of particular data problems, he cherry-picks evidence to attack researchers whose results he finds displeasing.

Like many people with conservative inclinations he is enamored of consumption data in the BLS’s Consumer Expenditure Survey (CES). This is probably because it shows no overall change in inequality since about 1986. What Reynolds doesn’t mention is that the quality of the consumption data has deteriorated badly since the mid-1980s. In 1985, the CES uncovered 80% of the consumption that is recorded in the U.S. National Income and Product Accounts. By the year 2000, the percentage had fallen to 61%. If the trend in total consumption is not reliably reflected in the survey, it is hard to see why we should accept its estimate of the trend in the distribution of consumption. (Contrary to Reynolds claim, the CES, not the Survey of Consumer Finances (SCF), is the information source used by Johnson, Smeeding, and Torrey to derive estimates of the trend in consumption inequality.[1] The SCF is a survey of wealth holdings, and thus it provides no direct evidence on household consumption.) So far as I know, no statistical series that tries to approximate total income has suffered such a terrible decline in quality as the data from the consumption survey. You’ll look long and in vain for any mention of this problem in Reynolds’s paper.

Reading his analysis, one is struck by how much it resembles a lawyer’s brief rather than an even-handed weighing of evidence. My reading of the evidence is reasonably straightforward, and it seems consistent with most of the income data available to us. Income inequality was higher at the end of the 1980s than it was in the beginning of that decade, and it was higher in 2005 than it was in 1989. Reynolds is certainly right when he says inequality did not increase “continuously” between 1979 and the present. It fell in some years, and remained approximately stable in others. On the whole, however, income inequality rose in the 1980s, and it also increased after 1989.

There’s an important difference between the rise in inequality in the earlier period and in the more recent period, however. Between 1979 and 1989, the percentage gap between the incomes of the middle-class and the poor got bigger, and the percentage gap between the rich and the middle class also got bigger. Inequality widened up and down the U.S. income distribution. Starting at some point in the early or mid-1990s, the proportional gap between low-income and middle-income Americans stopped rising and in fact probably shrank somewhat. After the early 1990s, the main way in which inequality widened is that the incomes of very well off Americans increased much faster than those of both the middle class and the poor.

Reynolds devotes most of his criticisms to the use of income tax data to measure trends in the income distribution. People who don’t know much about income distribution statistics might think this is because the income tax statistics are the country’s main source of information about distributional trends. They aren’t. Ever since the income distribution became a hot topic in the 1980s, the main source of information on which people rely comes from an annual Census Bureau survey of American households. The reason most people think inequality has risen since the late 1980s is because the household survey suggests it has.

Economists’ favorite indicator of inequality is the Gini coefficient. A higher Gini means there’s more inequality; a smaller Gini means there’s less. In 1989 the Census Bureau reported a Gini coefficient of household money income equal to 0.431. In 2005, the Bureau said the Gini coefficient was 0.469, which incidentally is the highest Gini coefficient ever recorded. By this measure, American inequality was about 9% higher in 2005 than it was in 1989. You can forgive most reporters and ordinary citizens for interpreting this Census statistic to mean that inequality has gone up.

There are problems with the Census Bureau’s main income statistics. One big problem is that the Bureau’s standard measure of income excludes in-kind benefits and capital gains, and it ignores the effects of income and payroll taxes. The Bureau recognizes these problems, and it publishes several alternative measures of inequality which use different definitions of income. If you look at some of the most comprehensive definitions of income, it turns out that inequality increased less, possibly much less, after 1989 than indicated by the Census Bureau’s headline number.

By the same token, however, inequality increased much faster under those alternative definitions during the 1980s than it did as measured by the Census Bureau’s headline number. So the Bureau’s headline number understated the growth in inequality during the 1980s and overstated the rise after 1989. Characteristically, Reynolds mentions the overstatement but fails to mention the understatement.

Reynolds could have written a very short paper in which he urged readers to ignore the Census Bureau’s headline measure of inequality and turn instead to a more obscure measure of inequality that is only published with a lag after the poverty and income statistics are first released. But that would have been a very short and uninteresting paper. It would also have been a very misleading one.

For a number of reasons, the Census Bureau survey does not provide accurate or consistent assessments of the incomes of the top 2% of income recipients. One reason is that respondents’ answers are top-coded, and the top-coding procedures have varied from time to time. Another is that the sample of high-income recipients is too small to give an accurate or consistent estimate of the incomes of the very top income recipients, say, those with incomes above $750,000 a year. Still another is that some income items that are important to the wealthy are under-reported in the Census survey. This means the Census Bureau probably gives us an underestimate of the gap between rich and middle-class families every year, no matter which concept of income we choose to measure.

What is worse, the underestimate will get bigger if top income recipients have incomes that grow faster than the incomes of people further down in the income distribution. That is precisely what most experts think has occurred since the late 1980s. It is also the interpretation of the data that Reynolds passionately wishes to reject.

If we cannot rely on the Census surveys to tell us what has happened to incomes at the top of the distribution, where might we turn? The scholars criticized by Reynolds have turned to income tax records. Another potential source of information is the Social Security Administration’s tabulations of the W-2 files, though this data series only covers wage earnings. The advantage of these administrative records is that they are supplied by many taxpayers. The records are so numerous that we can develop very accurate estimates of the incomes of people at the very top of the distribution.

Reynolds’ problem is that the data from both income tax returns and the W-2 records tell a simple and similar story. The relative incomes and wages of very top income recipients have been increasing much faster than the incomes and wages of people further down in the distribution. This was true in the 1980s, and it has also been true since 1990. Between 1990 and 2005 the median annual earnings of a full-time, year-round American worker increased about 4½%. What happened to the wages of top wage earners according to the W-2 records? At the 98th percentile, real earnings rose 33%; at the 99th percentile they rose 37%; at the 99.99th percentile, they rose 82%. Most people hearing those numbers would conclude – rightly, in my view – that wage inequality has gone up since 1990. That’s because 33%, 37%, and 82% are all bigger numbers than 4.5%.

Reynolds criticizes researchers who tabulate income reported on IRS 1040 forms without making any adjustment for the changing incentives to report income to the tax authorities. He notes, for example, that the 1986 Tax Reform Act made it advantageous to report capital income on 1040 forms rather than to partially shelter it by retaining earnings inside of a corporation. Tax reform in 1986 certainly increased the amount of income that top income recipients directly reported on their 1040s.

Reynolds’ problem here is that analysts he criticizes, like those at the CBO, recognize many of the problems of relying solely on income tax records, and they have tried to address them. CBO calculates total tax burdens faced by Americans, taking into account both the personal and the corporate taxes that they pay, directly on their own tax returns and indirectly through the corporations in which they have an ownership share. CBO may not do a perfect job, but at least it attempts to measure tax burdens and net incomes in an even-handed and consistent way. In addition, the CBO has developed a comprehensive measure of household income, including taxed and untaxed income, including in-kind benefits.

Reynolds is eager to remind readers that assets held in IRAs, 401(k)s, and pension plans generate current capital income that is not reported on income tax forms. (However, an unknown percentage of this income is reported on the Census Bureau’s household income survey.) If we corrected the omission, it would certainly add to the incomes of many middle-class households. Reynolds seems to believe this will reduce the measured trend toward greater inequality. He forgets that the assets in these plans are even more unequally distributed than ordinary income. If we included all the capital income that these plans produce, standard estimates of inequality would almost certainly rise, not fall. What matters for measuring the trend in inequality is the distribution of changes in untaxed capital income across the income distribution. Characteristically, Reynolds does nothing to learn how these changes have actually been distributed. He simply assumes (or hopes readers will conclude) that the additions to capital income will reduce the trend toward inequality. My suspicion is the opposite, but I would not try to persuade readers of this opinion before doing a careful analysis of the distribution of pension assets across households.

On an after-tax basis, the CBO’s estimates show that the average income of the top 1% of income recipients was 13.7 times the average income of the middle one-fifth of families in 1988-1990. By 2002-2004, this income ratio had risen to 15.9. Because 15.9 is a bigger number than 13.7, it seems reasonable to conclude inequality has gone up since the late 1980s. I don’t think Alan Reynolds has given us any persuasive reason to think this conclusion is wrong.

Reynolds is harshly critical of the tabulations and conclusions of analysts who find inequality has increased. It’s hard to see how these criticisms can have a big impact on our interpretation of the W-2 records. These clearly show a rise in wage inequality at the very top. In 1990 the ratio of the wage received by an earner at the 99.99th percentile to the median wage was 46:1. In 2005 that same ratio was 81:1. Yes, part of this increase was because of stock options, bonuses, and other nifty elements of the modern compensation package, but so what? In the old days, highly compensated wage earners did not receive these kinds of benefits or received much smaller helpings of them. Their total compensation was nearer to that of middle-income wage earners.

You can argue, as Reynolds does, that much of the increase in incomes at the top is due to turbo-charged stock prices and other special circumstances. Using the same line of reasoning you could also argue that, adjusting for the weather and the season, no homeowner in New Orleans ended up with a wet basement in August 2005. It might be true, but it’s not much comfort to the residents who had to flee a flooded home.

Ordinary citizens who think inequality has gone up are not making sophisticated adjustments for stock prices or other factors that vary from time to time. They’re reading news stories that tell how Robert Nardelli received compensation of $40–$45 million a year while failing to serve effectively as CEO of Home Depot. They’re asking why Hank McKinnell received an even more generous compensation package for doing an even worse job at Pfizer. Yes, these are horror stories; they are not data. But when careful economists go sifting through SEC filings, they find that the data match the horror stories. Top corporate officers’ pay rose faster than most people’s wages in the 1980s, and the growth differential was even bigger after 1990 than it was in the 1980s. If you think the typical U.S. worker got pay increases of 9½% a year after adjusting for inflation, as top corporate officers did in the 1990s, you move in different circles than the rest of us.

Notes

[1] David S. Johnson, Timothy M. Smeeding, and Barbara Boyle Torrey, “Economic Inequality Through the Prisms of Income and Consumption,” Monthly Labor Review, April 2005, http://www.bls.gov/opub/mlr/2005/04/art2full.pdf [pdf].

Is Alan Reynolds correct that widening inequality is a myth, or is he standing against a tide of increasingly persuasive evidence to the contrary?

If we were to find, as Reynolds contends, that differences in income and wealth are not as wide as we thought, that would be welcome news. Finding that lower-income individuals are better off than initially reported is preferable to finding they have lower income and wealth levels than we knew about. Unfortunately, a closer look at the evidence does not support Reynolds’ contention that data and measurement problems have produced a misleading picture of inequality in recent decades.

Before looking at some of the specific measurement issues Reynolds identifies, it’s useful to step back and take a broad overview of the research on inequality. There is considerable evidence on the inequality issue, much more than is discussed in Reynolds’ lead essay — it would be impossible to review it all here. But what we know about changes in inequality is summarized well in the speech given by Federal Reserve Chairman Ben Bernanke last week.[1] Chairman Bernanke says:

Although average economic well-being has increased considerably over time, the degree of inequality in economic outcomes has increased as well. Importantly, rising inequality is not a recent development but has been evident for at least three decades, if not longer. The data on the real weekly earnings of full-time wage and salary workers illustrate this pattern…. The long-term trend toward greater inequality … is also evident in broader measures of financial well-being, such as real household income….

He gives quite a bit more detail in the speech. It is worth noting that the view that inequality is increasing is shared widely across political party lines.

Let’s turn next to some of the more specific data issues. In his essay, Reynolds cites two issues repeatedly, the census top-coding issue and the fact that Gini coefficients do not show a rise in inequality.

Let’s take the Gini coefficient issue first. In a paper “Currents and Undercurrents: Changes in the Distribution of Wealth, 1989–2004“[pdf] from January 2006, a paper Reynolds mentions but does not give a full account of, Arthur B. Kennickell, the Senior Economist and Project Director of the Survey of Consumer Finances (i.e. someone who fully understands the data issues), says:

The Gini coefficient shows significant increases in the concentration of wealth in 2004 relative to 1989, 1992, and 1995; but the estimates from the 1998 and 2001 surveys are not significantly different from that from the 2004 survey. On the other hand, estimates of the total amount of wealth held by different subgroups of the wealth distribution show that the share of the least wealthy 50 percent of families fell significantly from … 1992–2001 … to about 2.5 percent of total family wealth in 2004…. Graphical analysis indicates that over the 1989–2004 period, there were statistically significant gains across the wealth distribution and that the level of gains was largest by far for the top few percent of the distribution.

I suppose it would be possible to cherry-pick a few statements from the paper to make a case, but the Gini coefficient evidence is by no means the refutation of rising inequality that Reynolds would have us believe. The write-up to the paper makes this clear.

What about the Census top-coding issue, does Reynolds have a point there? Last month, after Reynolds raised the top-coding issue, Paul Krugman reexamined this point and asked specifically whether top-coding would still matter. Krugman concludes:

The bottom line: top-coding really, truly does matter –- and yes, Virginia, income inequality is still rising.

The technique he uses is rather technical and involves fitting and integrating a Pareto distribution for top incomes and calculating the top-coding effect, but it is clear that the top-coding issue remains important despite Reynolds’ objections.

Let me turn to another place where Reynolds has raised measurement issues regarding inequality data. Recently, in a Wall Street Journal commentary, Alan Reynolds and David Henderson say CBO data on income inequality, which are widely used in inequality research, are misleading because they fail to properly account for interest and dividends earned on deferred income IRA and 401(k) type accounts. According to the commentary, since tax-deferred earnings are not reported, the distribution of interest income from these assets is imputed from reported interest on other assets and this skews the measured distribution of income toward inequality.

Is this an issue? Yes. Should we correct for it if we can? Of course, more precise data are always best. Will correcting the data affect the overall picture? No. In fact, it may even work in the other direction (Gary Burtless makes the same point in his reply).

To look at this, I used the Survey of Consumer Finances data shown in Tables 1 and 5 in “Recent Changes in U.S. Family Finances: Evidence from the 2001 and 2004 Survey of Consumer Finances,” by Brian K. Bucks, Arthur B. Kennickell, and Kevin B. Moore of the Federal Reserve Board’s Division of Research and Statistics.

This report shows that the median value of retirement assets for the bottom 90% of the income distribution was approximately $13,000 in 2001 and $15,400 in 2004, while the median value of retirement assets of the top 10% was approximately $138,500 in 2001 and $182,700 in 2004 (all values are expressed in 2004 dollars). If we assume a 5% rate of return, then the median lower 90% household would have earned around $770 in interest income. Presumably the CBO allocates some interest income to the lower 90% group. After all, they do have financial assets and report interest income to the government. However, even if we assume that no interest income at all is allocated to the bottom 90%, this correction would only raise their incomes by around $770, not enough to matter, especially when compared to the rise in incomes in the top 10% from sources other than the potential mismeasurement of capital income. The basic point is that most Americans have so little capital income that exactly how you count it is not an important issue.

One final point on this. When you examine the distribution of financial assets in the Survey, you see that the distribution of retirement assets rises more steeply with income than do other categories of financial assets such as stocks and bonds. Thus, if interest income is distributed by the CBO according to the distribution of reported interest income, this works in the opposite direction from what Reynolds claims since too little rather than too much interest income will be allocated to upper-income taxpayers.

I chose these examples because they are indicative of Reynolds’ analysis generally. When we examine the points he makes, we find that they are either too inconsequential to change the inequality picture, that they are an incomplete presentation of the evidence, or rebutted by other work. As conservative economist Bruce Bartlett says:

Even accounting for the factors Reynolds cites, there are too many different sources all showing a rise in income inequality… No matter how you slice it, the distribution of income has become more unequal over the last 20 years or so.

As another example, consider Reynolds’ criticisms of Siketty and Paez’s work in his lead essay and in this editorial from the WSJ. In response to Reynolds’ criticisms, Thomas Piketty and Emmanuel Saez have posted a detailed rebuttal[pdf] on Emmanuel Saez’s web site. Here are a few selections:

In his … article, … Alan Reynolds casts doubts on … our results showing that the share of income going to the top 1% families has doubled from 8% in 1980 to 16% in 2004. In this response, we want to outline why his critiques do not invalidate our findings and contain serious misunderstandings on our academic work. …

Alan Reynolds points out that reported incomes may not reflect true incomes because of tax evasion or tax avoidance. This is a legitimate concern and we, along with a number of colleagues, have actually spent substantial time investigating this issue. Alan Reynolds has picked some of the facts in order to provide a very skewed view. …

Even the small point on 401(k)s is conceptually mistaken… In sum, our work has shown the top 1% income share has increased dramatically in recent decades… The reduction in taxes at the top since 2001 has mechanically exacerbated the discrepancy in disposable income…

Moving to a final issue, one that Gary Burtless raises as well, there are potential measurement issues that work in both directions –- some adjustments to income and wealth work against inequality and some work for it -– and a fair presentation of the evidence would note both sides. But that is not what we have. Only those adjustments that favor the proposition that rising inequality is a myth are presented favorably by Reynolds. For example, we didn’t hear about this:

There is evidence, however, that because of the way the G.D.P. is calculated, the actual shift [in inequality] is much more pronounced. “We know that income inequality is quite substantial,” said Harry J. Holzer, a labor economist at Georgetown University, “and this new evidence suggests that it is worse than we thought.”

The Bureau of Economic Analysis, which issues the G.D.P. reports each quarter, is on the case. So are two prominent economists at the Federal Reserve. … If these [adjustments] … were incorporated into G.D.P. …, labor’s share of national income would decline from a fairly steady 65 percent in the 1950′s, 60′s and 70′s to less than 60 percent today.

The issue is whether R&D should be counted as an investment or an expense. If it is counted as an investment, and there are good reasons to do so, the picture changes dramatically. This change alone would dwarf the kinds of adjustments Reynolds discusses.

As I said at the start, I would be very pleased to find out that growing inequality is not a problem. However, despite the attempts of Alan Reynolds and a few others to argue otherwise, the preponderance of evidence and of professional opinion clearly indicates that inequality has been rising since 1988, and that the trend toward widening inequality has been present for much longer than that. The question is what, if anything, to do about it. If we can get past the attempts to cloud the issue, perhaps we can proceed to more important discussions.

Notes

[1] Janet Yellen, president of the San Francisco Fed, also has a very nice summary of the inequality evidence. See “Economic Inequality in the United States.”

One doesn’t have to be a devotee of Kurosawa or even Jacques Derrida to understand that folks with access to the same facts can and will provide alternative interpretations of them. So what may seem like a simple question of fact is often far from it. Here I provide the perspective from which I consider whether income inequality has risen since 1989; my interpretation of what the March Current Population Survey (CPS) can tell us about it; and answer the deeper question—why does it matter?

As someone interested in measuring how economic well-being changes over time, I often focus on average United States household income and its distribution using public-use CPS data. Recently, after a lengthy process, I gained access to the restricted-use internal CPS data the Census Bureau used to estimate the household income Gini values Gary Burtless referenced in his comments. From my perspective Burtless’ simple comparison of the difference in these two values overstates the trend in income inequality after 1989 and fails to convince me that the income equality increase, once it is properly measured, matters.

Figure 1 using CPS data shows how median income changed in the United States between 1967 and 2005 in constant 2005 dollars. While median income rose substantially over this period, years of economic gain were followed by years of economic loss. Hence one could, with an appropriate choice of years, either demonstrate that median income increased, decreased, or stayed the same. To separate longer-run trends from differences in the business cycle, I believe it’s best to compare similar years—e.g. peak to peak or trough to trough. Doing so, one can characterize the income consequences of the 1980s business cycle comparing 1979-1989 and the 1990s business cycle comparing 1989-2000 or alternatively by comparing trough years 1983-1993 and 1993-2004. Either way, United States median household income improved significantly over both the 1980s and 1990s. This is good news that does matter—long-term economic growth has occurred over the past two business cycles improving the economic well-being of the average American.

Describing what happened to the income distribution of households is more complicated. In Burkhauser, Oshio, and Rovba (forthcoming) discussed by Reynolds, rather than using a single Gini coefficient to do so, we provide pictures of the entire distribution over the peaks years 1979-1989-2000 in the United States and compare them with peaks years in the 1990s business cycles in Great Britain, Germany, and Japan. Laying the 1989 distribution over the 1979 distribution (Figure 2), the middle of the distribution (the middle class) dramatically declined over the 1980s. This is consistent with the Gini increase discussed by Reynolds and Burtless. But the main point is that the vast majority of the “disappearing” middle moved to the right tail with a small minority moving to the left. That is, while inequality certainly increased in the United States in the 1980s, it did so primarily because the disappearing middle became disproportionately richer.

The news is much better for the 1990s. The 2000 distribution is to the right of the 1989 distribution. That is, in 2000 the person found at every point in the income distribution was better off than the person at that same point in the distribution in 1989—first order stochastic dominance. This is a major achievement for our society. But it is especially so when compared with Germany and Japan (Figure 3 and Figure 4) whose distributional changes over the 1990s look much like ours in the 1980s—a decline in the middle, with most people become richer but some becoming poorer. Great Britain matched our achievement in the 1990s. (Figure 5).

These pictures of reported household income minus estimated income and social security taxes do not differ from those unadjusted for taxes. (They would be closest in spirit to the CBO after-tax estimates Burtless reports.) But they provide a more nuanced view of what is behind the Gini changes discussed by Reynolds and Burtless.

We also estimate Gini coefficients and find a small decline in after-tax inequality between 1989 and 2000 and no change in before-tax income inequality over this period. So how does this square with the increasing Gini values Census produced using the internal CPS data reported by Burtless?

Like the vast majority of researchers, who would rather estimate their own Gini coefficients than depend on the Census, our results come from the public-use CPS. As can be seen in Figure 6 based on new work with Shuaizhang Feng and Stephen Jenkins, if we use public-use data unadjusted for top-coding, we also find a large rise in pre-tax income inequality 1989-2000. But this is in part because of substantial increases in the top codes and in the use of cell means since 1995 and in part because of other methodological changes in the CPS in 1993. Like Burtless, when he does his own work using the CPS, and others, we correct for this by consistently top-coding the public-use data—choosing the year between 1975 and 2004 in which an income source top code hits at the lowest point in the distribution and consistently top-coding all other years of that income source at this same low point in the distribution.

In Burkhauser, Butler, Feng, and Houtenville (2004), we show that while this method systematically understates levels of labor earnings inequality, it captures the trends in both the unadjusted Census internal and external Gini values controlling for the spikes in these data in 1992-1993 and 1994-1995, respectively. Our results are similar to those who simply “trim” the top 2 or 3 percent of the public-use data. Hence our results consistently report what has been happening to household income and its distribution for the bottom 97 or 98 percent of the distribution. I believe both Burtless and Reynolds would at least agree that the public-use CPS data when appropriately adjusted for these problems accurately captures trends for this part of the population.

What we have carefully documented since gaining access to the internal CPS data is that even these data have problems with inconsistent censoring at the top of the distribution. Burtless reports Census Gini values that are part of a series from the internal CPS data that is not corrected for censoring. As Reynolds discusses and as we independently document in Burkhauser, Feng, and Jenkins (2006), over the period 1975-2004, the internal CPS data, like the public-use data, do not systematically capture the upper end of the income distribution. Like the public-use CPS, internal censoring, including top-coding, occurs on each individual source of income rather than on overall income, and we find that the share of persons living in households with one or more income sources top-coded varies from 0.1 to 0.8 percent.

When we adjust the internal income data using our consistent top-coding procedure we find a more modest increase in our Gini values over the period after 1989 than is found using the uncorrected internal data. For 1989-2000 it is 4.67 rather than 6.82 percent. But even this is likely to be too high a rise in both because, even adjusting for top-coding and censoring, we still find a spike in the 1992-1993 internal CPS data that, while lower than the spike in the unadjusted internal data, is still implausibly high. In fact, if you only follow Gini trends from 1993-2004, all the years of internal data available to us since the 1992-1993 spike, Gini values only increase by 1.45 and 2.43 percent respectively. My bottom line is that the rise in inequality captured by internal Census data, once it is adjusted for censoring, tells much the same story for the bottom 99 percent of the income distribution as the adjusted public-use CPS data tells. Over the 1990s business cycle the entire distribution moved to the right with little or no change in income inequality. Since 1989 household income inequality has risen very little and much less than in the previous decade. This is very good news that matters.

I agree with Reynolds and Burtless that CPS data are less valuable for looking at the top 1 or 2 percent of the income distribution. Unfortunately, other data sets are also hard-pressed to do so and there is much greater uncertainty over how much the income of the top 1 percent changed since 1989 than there is for the rest of us.

But does this really matter? Our economy is not a zero sum game. My gain does not mean your loss or vice-versa. I know of no evidence that increases in the incomes of the top 1 percent of our population are the root cause of the challenges faced by those at the other end of the distribution. Over the last full business cycle, the incomes of rich and poor moved in the same direction. I suspect that for every Robert Nardelli or Hank McKinnell there are far more people like Tiger Woods, Steve Jobs, Oprah Winfrey, or Bill Gates, whose skills, vision, and effort allowed them to burst out of the pack of us ordinary 99 percent. But in doing so, the value of the goods and services they create for us greatly exceeds the earnings they receive. That is the consequence of our market economy and that does matter.

References

Burkhauser, Richard V., J. S. Butler, Shuaizhang Feng, and Andrew J. Houtenville. (2004) “Long Term Trends in Earnings Inequality: What the CPS Can Tell Us,” Economic Letters, 82 (2) (February): 295-299.

Burkhauser, Richard V., Shuaizhang Feng, and Stephen P. Jenkins (2006). “Using the P90/P10 Ratio to Measure Inequality Trends with the Current Population Survey: A View from Inside the Census Bureau Vaults, Cornell University Working Paper (November).

Burkhauser, Richard V., Takashi Oshio, and Ludmila Rovba (forthcoming). “Winners and Losers over the 1990s Business Cycles in Germany, Great Britain, Japan, and the United States,” Schmollers Jahrbuch: Journal of Applied Social Science, 127 (1).

Welniak, Edward J. 2003. “Measuring Household Income Inequality Using the CPS,” James Dalton and Beth Kilss (Eds.), Special Studies in Federal Tax Statistics 2003, Statistics of Income Directorate, Inland Revenue Service, Washington DC.

Introduction

Inequality is a fascinating subject, one that provokes discussion and makes it hard to settle the apparently simple question of whether income inequality in the US has increased since 1988. In his essay Alan Reynolds presents evidence in favor of his thesis that income inequality has not increased, Gary Burtless and Mark Thoma present evidence in favor of the opposite view that inequality has significantly increased, while Richard Burkhauser concludes that there has been an increase, but a modest one. In our reply we will bring additional facts to this discussion. Our main point, however, is to argue that to focus only on the evolution of current income inequality is insufficient if one is interested in the evolution of the distribution of living standards in the U.S.

Why only look at current income inequality?

The material sources of a household’s well-being are the flow of consumption and possibly the flow of its leisure enjoyed over its lifetime. Thus if one is ultimately interested in the distribution of well-being across U.S. households, the object of study ought to be the joint distribution of lifetime consumption and leisure across them.

Measuring inequality in lifetime consumption and leisure is an impossible task as it would require data on consumption for many households, each followed for a very long period of time. Such data unfortunately do not exist. What is then a feasible and appropriate way of measuring inequality in lifetime consumption?

If people would earn constant income throughout their lives, current income would be a very good approximation of lifetime consumption. In the data, though, current income fluctuates substantially over the life cycle and thus it is a potentially poor indicator of lifetime resources. Fortunately, we can use simple theories of consumption decisions of households over time (pioneered by Nobel prize winners Milton Friedman and Franco Modigliani), which show that if households can borrow and lend on financial markets, then there is a strong link between the lifetime resources of a household (sometimes also called its permanent income) and its current consumption. Current income shocks that are not fully permanent (such as seasonal shocks, job losses, bonus payments) strongly affect current income (and thus have a strong impact on the distribution of current income), but only have a moderate impact on permanent income and thus consumption.

While current consumption is not always perfectly connected to permanent income (this connection for example is weakened in the presence of borrowing constraints), it is, under many circumstances, a better proxy for it than current income. Under those circumstances documenting trends in consumption inequality is more informative about the inequality in lifetime resources.

Let’s illustrate the point further with a simple example. Think of a world with two households, the White and the Reds, whose monthly income has two parts: a fixed part and a fluctuating bonus. Suppose that the Whites have a fixed part of $100 and a bonus that alternates between $10 and $30, while the Reds have a fixed part of $50 and the same bonus structure. Both households realize that their bonuses fluctuate but they both dislike fluctuating consumption. As a consequence they will save part of their high bonus and run down their bank account when their bonus is low. This will lead the White to consume around $120 and the Reds to consume around $60.

Now suppose that bonuses become more volatile and go from a $10-$30 structure to a $0-$40 structure. What happens to current income inequality? The higher volatility of bonuses makes it more likely to observe larger differences in income and thus current income inequality will go up. Should we worry about this increase in inequality? Not really, since the lifetime distribution of resources between White and Red has not changed (as reflected in the unchanged consumption inequality). If anything inequality will have a positive effect on incentives as both will work harder to get the high bonus.

Suppose instead that the bonus structure is eliminated and White receives a permanent raise of $5 while Red takes a $5 permanent cut. What happens to income inequality? Since the very volatile bonuses have disappeared, current income inequality is likely to be unchanged or even diminish. Should we be happy about it? Not really, because the decline in current inequality masks an increase in inequality in lifetime resources and thus in well-being. Note that since the cut and the increase are permanent they will be fully reflected in consumption, and thus tracking consumption inequality will fully reveal the increase in lifetime inequality, with its adverse distributional consequences.

The example above, although very stylized, suggests that focusing only on inequality in current income can lead to very misleading conclusions regarding the welfare effects of inequality. Looking also at consumption inequality can help us avoid these conclusions. The example also stresses the role of financial markets (the bank account in our stylized example) as the key link between inequality in lifetime resources and inequality in current consumption. It suggests that in these days of very developed financial markets, consumption, and not current income, should be the cornerstone of empirical and theoretical inequality studies.

Income and Consumption Inequality in the US

One important reason for the popularity of using income inequality for measuring the distribution of living standards in the U.S. is the availability of several data sets collecting household income data. These differ in their sample size, quality, and ability to capture a representative sample of the U.S. population, as spelled out in the Reynolds essay. However, comprehensive and detailed household-level consumption data is available for the U.S. for the period under question (which, following Reynolds, we take to be 1988 to 2005). While other authors involved in the public debate have questioned the quality of the Consumer Expenditure Survey (CES)[1], which is administered by the U.S. Bureau of Labor Statistics, a number of prominent scholars have now used this data set to document trends in consumption inequality over the last 25 years.[2]

One bonus of using the CES is that it also contains information on income, and hours worked. Thus we can look at inequality from different angles for the same set of households. One drawback is that the CES is a relatively small survey (between 5,000 and 10,000 households per quarter) and thus does not contain very precise information about the top 1% of the population. In Figure 1 (from Krueger and Perri, 2006) we report the evolution of several measures of income (the green lines) and consumption inequality (the blue lines), all computed from the CES sample for the 1980-2003 period.[3] The bottom two panels, for example, report the 90/10 ratio and the 50/10 ratio for income and consumption. The 90/10 and 50/10 indicators have the desirable properties that are not affected by changes in top-coding procedures and are easy to interpret: an income 90/10 ratio equal of 5 suggest that the per capita income in the household at the top 10% of the income distribution is 5 times the income of the household at the top of the bottom 10% of the income distribution. The key message from our figure is that the increase in income inequality in the U.S. has been much more pronounced than the corresponding increase in consumption inequality.

Let us come back to the original question posed by Reynolds. Has US current income inequality increased over the period 1989-2003? Looking at CEX data suggests that yes, it did. As also suggested by Gary Burtless, one main source of the increase has been the increase of incomes at the top part of the income distribution relative to incomes at the bottom and middle of the distribution. The 90/10 ratio increases from around 5 in 1989 to around 6 in 2003.

The consumption data suggest, though, that the consequences of this increase have not caused an increase in the dispersion of the distribution of lifetime resources; if it did it would have showed in increased consumption inequality. Consumption inequality, however, has remained substantially stable.

Is the stability of the consumption distribution simply caused by a massive and increasing measurement error and/or misreporting of consumption? We can’t rule out this possibility, but there is some additional evidence that makes us think this is not the case. An increasing income inequality coupled with a stable consumption inequality implies that, according to economic theory, the larger income fluctuations are now smoothed through stronger use of credit markets. Therefore we should expect to observe a significant increase in the volume of household credit in the US over the last 15 years. Many credit indicators point exactly in that direction. This seems to indicate that the stability of the consumption distribution is indeed informative about the trend of inequality in lifetime resources.

Conclusion

One conclusion we would not like the readers to take home is the generic one that we should not worry about income inequality. Rather we would like to convince them that understanding the welfare effects of changes in measured inequality, and possibly the appropriate policy measures to deal with it, is a complex task that involves more than reporting the distribution of current resources. Ideally one should understand and measure the distribution of lifetime resources. In order to understand how lifetime resources translate into observable indicators, and what these indicators are, it is crucial to have a thorough understanding of how and to what extent households can transfer resources through time and across states of the world using financial markets. Our own previous work has highlighted the importance of using consumption as an indicator, but recent exciting work is being done by leading researchers in the economics community stressing the role of inequality and dispersion in other variables, too, such as labor effort or wealth, and assessing their impact on incentives, the allocation of resources, and the distribution of welfare.

References and Footnotes

Krueger Dirk and Fabrizio Perri, “Does Income Inequality lead to Consumption Inequality?,” Review of Economic Studies, March 2006.

[1] Gary Burtless correctly mentions that when one measures total consumption in the CES and compares it to total consumption as measured in the National Income and Product Accounts (NIPA) the fraction of CES to NIPA consumption declines substantially from 1985 to 2000. While this raises some concern about the quality of the CES data (NIPA consumption data may not be perfect either, though), it is very hard to assess if and how it biases estimates of inequality measures and their trend over time. In order to partially address this problem we include in our measure of consumption several categories of durables, for which the problem mentioned above is much less severe. In addition, in the most recent waves of the CES (2000-2005) the discrepancy between growth in NIPA consumption and aggregate CES consumption has shrunken.

[2] To reply to Gary Burtless’ suggestion that many people with conservative inclinations [are] enamored of consumption data in the BLS’s Consumer Expenditure Survey (CES), we want to point out that a) We used the CES because 30 years of economic theory teach us that looking at consumption is a much more compelling way to understand the effects of inequality and the CES is the only dataset that allows us to do so in the US for a long enough period of time; so if we were enamored of the CES, it is only because it is the only girl in town b) We like to think of ourselves as social scientists, and as such we keep our political inclinations, whatever they are, very separate from our work.

[3] Details about the construction of the data are available in the paper, which is available here[pdf] or here[pdf].

Gary Burtless rightly emphasizes that “economists’ favorite indicator of inequality is the Gini coefficient,” but “the Bureau’s standard measure of income excludes in-kind benefits and capital gains, and it ignores the effects of income and payroll taxes.” That is why I presented a chart of 25 years of Gini coefficients[pdf] for the Census Bureau’s measure of “disposable income” — which adds transfer payments and taxable capital gains, but subtracts income and payroll taxes.

Suppose that instead of providing data for 25 years, I had mentioned only two years as Burtless, Bernanke, Piketty and Saez and other keep doing. I could then say the Gini coefficient for disposable income fell from 0.41 in 1986 to 0.40 in 2004, proving inequality has fallen for more than 20 years. Or I could say the Gini coefficient rose from .39 in 1985 to .40 in 2004, proving inequality has increased. We cannot find out what happened when by showing data for only two years and then drawing an imaginary line between them.

A look at the past 25 years of Gini coefficients shows no significant and sustained increase in inequality of disposable income since 1988, or even 1985. Thoma claims, “The Gini coefficient evidence is by no means the refutation of rising inequality that Reynolds would have us believe.” Amazingly, he refers only to a Gini coefficient for wealth — as if income doesn’t matter. After ignoring all income statistics, he accuses me of “an incomplete presentation of the evidence” and “attempts to cloud the issue.”

Thoma asks, “What about the Census top-coding issue, does Reynolds have a point there?” My point, which he never mentions, is that the increase in (misnamed) “top-coding” in 1993 made it appear as if inequality had jumped to a higher plateau when all that happened is that the survey did a better job of recording higher incomes. I respect the Economic Policy Institute for properly reporting that “a change in survey methodology in 1993 led to a sharp rise in measured inequality,” and I naively expect others to be equally honest. It is for this reason, as Burkhauser explains, that “Burtless’ simple comparison of the difference in these two values overstates the trend in income inequality after 1989.” Burtless refers only to the “headline” Gini coefficient which he criticized and only for two years, 1989 and 2005.

Ben Bernanke

Thoma quotes Ben Bernanke saying “rising inequality … has been evident for at least three decades.” But Bernanke also presented data for only two years, 1979 and 2004. In the seriously flawed income data he mentioned, all of the increase in inequality (aside from the survey change in 1993) occurred between 1979 and 1985. The data Bernanke cited for the share of income received by households in the top and bottom fifths were not, as he claimed, “after taxes have been paid and government transfers have been received.” The figures he cited were not for disposable income, as he implied, but for “post-social insurance income” — which shows what income distribution would look like if there were no poverty programs and no taxes. His measure includes taxable capital gains but not the taxes on those gains or any other taxes, and it explicitly excludes all means-tested transfer payments such as the EITC, TANF, WIC, Medicaid, housing allowances and food stamps.

Bernanke began by mis-defining income inequality as differences in the wages of college grads and dropouts. Yet Burtless found that wage inequality explains “only about one-third of the increase [from 1967 to 1999] in income disparities.”[1]

Quoting such famous people appears to provide comfort to those unwilling to even look at my graphs. But a cozy consensus of “professional opinion” about bad data is no substitute for good data.

Paul Krugman

Thoma links to Paul Krugman’s discussion of a topic illustrated in my Figure 4 — the $1 trillion gap between the Piketty-Saez and Census Bureau estimates of the top 5 percent’s income share. The Census data, I wrote, say the income share “rose from 18% in 1986 to 20.9% in 2004, mostly because of a data break in 1993. The comparable Piketty-Saez figure jumped from 22.6% in 1986 to 27% in 1988 … [and] hit 31.2% by 2004.” Krugman claims the difference of 10.3 percentage points in 2004 is because Census supposedly misses extremely high incomes. Even if Census missed every dollar above $5 million (where the tax data shine), that would not come close to explaining the gap. If we “exclude all income above $5 million from the Piketty-Saez estimate of the top 5 percent’s share,” I calculated, that would narrow the gap by less than one percentage point.

Krugman concealed the huge gap and thereby trivialized the issue into a matter of differences in rates of change since 1994. He wrote that, “The Census data say that the income share of the top 5% rose only slightly, from 21.2% to 22.2%, between 1994 and 2005. The Piketty-Saez data, which only go up to 2004, show a 3.7% rise. Our little exercise with earnings data suggests that the missed income due to reporting limits rose by about 2 percentage points over the same period.”

The Piketty-Saez estimate of the top 5 percent’s share rose by 8.6 percentage points from 1986 to 2004. Figure 4 shows that half of that happened between 1986 and 1988, when the business portion of top 5% incomes from 8.8% in 1986 to 15.5%. Krugman’s Pareto interpolation seems a painfully elaborate ruse to avoid discussing (1) the trillion dollar gap between the two series, and (2) the fact that, as Burtless put it, “tax reform in 1986 certainly increased the amount of income that top income recipients directly reported on their 1040s.”

The CBO

When Thoma comments on my Wall Street Journal piece with David Henderson, he changes the subject from corporate profits to interest income. He imagines we wrote that “since tax-deferred earnings are not reported, the distribution of interest income from these assets is imputed from reported interest on other assets and this skews the measured distribution of income toward inequality.” Amazingly, the words “corporate profits” appear nowhere in his convoluted analysis of something we never wrote, even though the misallocation of 59.4% of corporate profits to the top 1% was the focal point of our graph and article. The CBO added 39% of corporate profits to top 1% incomes in 1989 and 59% in 2004, thus fabricating a wholly artificial increase in the top 1 percent’s share.

“CBO may not do a perfect job,” says Burtless, “but at least it attempts to measure tax burdens and net incomes in an even-handed and consistent way.” Attempting something is not the same as achieving it. The CBO’s attempt to include tax-exempt interest before it began being reported in 1987, to include only the taxable portion of capital gains (outside of IRAs), and to allocate corporate profits by using an indefensible technique renders CBO data much worse than useless for estimating top centile shares of income over time.

Piketty and Saez

I plan to write a detailed comment on the Piketty and Saez reply to my December Wall Street Journal article elsewhere. For now, I will mainly focus on sections from their reply that others have mentioned.

Piketty and Saez did not suggest that I have misquoted them or that any of my statistics are wrong. If their reply to me is correct, then much of what they have written in the past (and I have quoted extensively) must have been wrong.

They now assert, for example, that there is an “emerging consensus” that the elasticity of taxable income (ETI) is just a transitory blip. On the contrary, the most recent (1999 to 2004) estimates for permanent ETI were 0.57 from Auten and Carroll, 0.40 from Gruber and Saez, and 0.53 from Kopczuk and 0.62 from Saez himself. Piketty and Saez claim that Goolsbee’s 2000 paper about executive pay (which double-counts stock options when granted and exercised) trumps Saez’s 2004 paper. Yet Eissa and Giertz report that, “for executives, we find a permanent earned income elasticity for the early 1990s of 0.8 (with no anticipation effect).”

By the time the Piketty and Saez reply reached The Wall Street Journal’s letter section on January 11, they had prudently omitted their previous comment that my “small point on 401(k)s is conceptually mistaken.” My Figure 3[pdf] shows it is not a small point http://www.cato.org/event.php?eventid=3441. The Reynolds-Henderson piece proves it is not mistaken.

Playing the tiresome two-year game, Piketty and Saez say “the share of income going to the top 1% families has doubled from 8% in 1980 to 16% in 2004.” But their data are for tax units, not families. Two married people earning $50,000 apiece report twice as much income per tax unit as an unmarried couple with the same income. Half of the 1980-2004 increase in top centile shares happened in just two years, 1986-87, and all of the apparent increase since 1986 is fully accounted for in the Cato paper and even in Figure 1 and Figure 2.

Although their data exclude taxes and transfers, Piketty and Saez boldly assert that, “the reduction in taxes at the top since 2001 has mechanically exacerbated the discrepancy in disposable income.” CBO estimates show the opposite of what Piketty and Saez assert.

Brad DeLong recently singled out the most substantive comment in the Piketty and Saez reply. They suggest that if many businesses had switched from filing under the corporate tax to the individual tax (a “scenario” documented by Saez), then we would have seen more business income in the top 1% but also smaller capital gains. So what?

My Figure 2[pdf] shows that the share of top incomes from capital gains did indeed fall dramatically in 1986-88 when the share from business soared, confirming the Piketty-Saez theory. But the dollars gained from business income were much larger than dollars lost from capital gains, so the net effect pushed the top income shares way up. The capital gains tax soared as the business share stopped rising after the individual income tax was increased in 1993, but cutting the capital gains tax in 1997 was a major reason. When tax rates were simultaneously reduced on business income, capital gains and dividends, the top 1 percent’s income from those sources climbed as the salary portion declined. Yet Piketty and Saez focus on the dwindling salary share. The only way to see what happened is to reveal all the data, rather than just two years. The Piketty-Saez time series on sources of top income shares reveal that their data is seriously distorted by taxpayer responses to changing tax rates.

Gary Burtless

Burtless and Thoma make a reasonable request for a fair and even-handed weighing of the defects of distribution studies based on samples of tax returns, the Consumer Expenditures Survey (CES), and Census Bureau’s Current Population Survey. I make the same request for my work on CBO and Piketty-Saez data, and for my figures on changes in median income by fractile from the Fed’s Survey of Consumer Finances in Figure 7.

I do not believe an even-handed approach would have been possible before the publication of Income and Wealth. Until then, scarcely anyone had seriously questioned income distribution estimates that were based on tax returns. I have now presented considerable evidence that taxpayers change what they report as income, and how they report it, in response to changes in absolute income tax rates (elasticity) and relative tax rates (income shifting). To dismiss all that evidence as “inconsequential,” as Thoma does, is neither reasonable nor persuasive.

Unlike taxpayer behavior, the topic of consumption inequality has never been a major focus of my work; it takes up only 3 of the 231 pages in my book (pp.162-64). I have written hundreds of articles since 1972 about income, or about wealth. But I recall writing only one that used the CES. I have mainly left this important topic to experts, such as Krueger and Perri.

Burtless corrects my careless memory lapse, when I wrongly suggested that Johnston, Torrey and Smeeding did not rely on the Consumer Expenditure Survey. He finds it a sign of unfair bias on my part that I failed to mention that, in his words, “in 1985, the CES uncovered 80% of the consumption that is recorded in the U.S. National Income and Product Accounts. By the year 2000, the percentage had fallen to 61%.” I didn’t mention those figures because I find them misleading.

A team of five BLS economists recently found that when comparing comparable categories of items, “that CE aggregate expenditures are 86 percent of PCE aggregate expenditures for 1992, drop to 85 percent in 1997, and fall further to 81 percent in 2002.”[2] To push that 81% down to about 60%, as Burtless does, requires comparing categories that are not at all comparable.

The CES asks what consumers spend. PCE also includes what governments and nonprofit organizations spend on products and services used by consumers. The main reason the gap has widened between PCE and CES is Medicare and Medicaid, plus other third party expenditures on education, social welfare, religion, research and war. In 1997, the CES was only 17% as large as the PCE for medical care – a $724 billion gap. The CES was only 51% as large as the PCE for education and research, only 27% as large for legal services, and only 13% as large for Social Welfare. Unlike PCE, the CES does not include clothing and food for the troops in Iraq, or research grants and scholarships, or purchases of food and clothing by churches and the Salvation Army. The fact that only the NIPA’s PCE series includes such rapidly-expanding spending by governments and nonprofits is not evidence the CES suffers from declining quality.

Another Red Herring

Citing a newspaper article, Thoma opines that treating R&D expenses as an investment would “dwarf the kinds of adjustments Reynolds discusses.” The article claims that “when R&D is counted as profit, the employee compensation share of national income drops by more than one percentage point,” and says this would somehow shrink labor’s share from 65% in the sixties (no R&D back then?) to “less than 60 percent today.” But investment is not profit and one percentage point is not five. Labor’s share of national income (aside from self-employment) was 62.5% from 1960 to 1960 and 65.5% from 2001 to 2005.[3] Besides, changes in NIPA bookkeeping conventions have no effect on anyone’s income.

What does this swelling sea of red herrings have to do with my simple request for some credible evidence — meaning, not based on tax returns — that shows rising inequality from 1988 to 2000 and/or from 2000 to date?

Anyone seriously interested in the level or change in relative living standards must look at either consumption or a mix of disposable income for the whole population, — not just for 1 percent or for a few dozen CEOs in the study Burtless cites, and not for just for two years separated by such major breaks as the 1986 tax reform and the 1993 change in top-coding.

Nobody has yet found fault in the evidence I presented showing no significant and sustained change in the inequality of disposable income or consumption since 1988. If we can get past attempts to change the subject, perhaps we might begin to examine the facts.

References

[1] Gary Burtless, “Has Widening Inequality Promoted or Retarded U.S. Growth,” Canadian Public Policy, Vol XXIX, 2003, p. S189.

[2] Thesia I. Garner, et. al., “The CE and the PCE: a comparison,” Monthly Labor Review, September 2006.

[3] Alan Reynolds, “Statistical Politics Update.” The Washington Times, October 22, 2006. (Webbed here.)

Alan Reynolds wants to disprove the widely accepted view that American income inequality widened after 1988. In his attempt to make this case he offers some evidence that is relevant, much more that is irrelevant, and still more that cannot be evaluated without careful and open-minded analysis of the data. Unlike many of the analysts he criticizes, including economists in the Congressional Budget Office and Professors Thomas Piketty and Emmanuel Saez, he has never actually done any of the hard analysis that would allow us to assess the importance of claimed shortcomings in the data. I leave it for other readers to decide whether Reynolds has the temperament to treat good researchers’ analysis and results with an open mind, especially when their findings conflict with his fond hope that income inequality stopped rising almost two decades ago.

In my earlier comment on his claims, I noted that the 25-year trend toward higher inequality changed sometime in the early 1990s. Between 1979 and the early 1990s the percentage income difference between poor and middle class families increased, as did the percentage income gap between middle class and very affluent families. After the early 1990s there is considerable evidence that the percentage gap between low- and middle-income families stabilized or actually began to shrink. If we use the most comprehensive definitions of income, ones that take account the effects of cash and near-cash government transfers as well as income and payroll taxes, you can make a convincing case that inequality in the bottom half of the income distribution actually declined. The strong economic expansion after 1994 produced sizeable income gains for low-income working families. A big expansion of the Earned Income Credit helped boost the after-tax incomes of many breadwinners who earn low wages. Richard Burkhauser is one among several good analysts who have demonstrated how income gains in the 1990s expansion were more broadly distributed than gains in the 1980s expansion.

Economists at the Congressional Budget Office, Professors Piketty and Saez, and economists Ian Dew-Becker and Robert J. Gordon, are among the researchers who have demonstrated that income gains at the very top of the income distribution followed a different path from the one observed in the middle and at the bottom of the distribution after 1988. Income and wage disparities at the very top of the income distribution did not shrink, as was the case in lower parts of the distribution. Income disparities at the top got bigger. For reasons mentioned in my earlier comment, income trends at the very top of the distribution are hard to measure in standard household income surveys conducted by the Census Bureau. I don’t think Reynolds has made a persuasive case against this view. Nonetheless, the Census Bureau’s main household survey also shows unmistakable evidence of growing inequality at the top of the distribution. In 1988 – Reynolds’ preferred base year – a wage and salary worker who earned the median hourly wage received $13.20 an hour (measured in constant 2005 dollars). By 2005 the median worker’s wage increased to $14.29 an hour, an increase of 8.3%. Over the same span of years, a worker earning the 95th-percentile wage experienced a 20.3% gain in pay.[1] This pattern of bigger earnings gains at the very top is mirrored in hourly pay trends for both male and female workers. It is also reflected in the annual earnings reports collected once a year by the Census Bureau in its Current Population Survey (CPS). The closer we get to the very top of the earnings distribution, the bigger the earnings and income gains enjoyed by workers and their families.

My own calculations, based on the CPS files and using the most comprehensive definition of income available to us, suggest that median household-size-adjusted disposable income increased 13% between 1988 and 2004. (The gains are calculated using constant dollars.) At the 75th percentile, real income increased 16%; at the 90th percentile, it increased 21%; and at the 95th percentile, it increased 27%. Inequality in the bottom 95% of the income distribution did not increase as much as these figures suggest, because income recipients at the bottom of the distribution enjoyed proportionately faster income gains than income recipients in the middle. Inequality in the bottom half of the income distribution fell, while inequality in the top half of the distribution increased. The bottom line, however, is that the Census Bureau’s CPS files show considerable evidence that wage and income gains were bigger at the top than in the middle.

Furthermore, the same pattern of faster income gains at the top of the earnings distribution is confirmed in both the IRS income tax data and the Social Security payroll tax data. In my earlier comment, I demonstrated how real earnings gains in the Social Security wage data showed progressively bigger percentage gains in wage and salary earnings as we move up the annual earnings distribution. A couple of additional statistics reinforce this point. In 1990 the Social Security Administration found there were about 15,600 wage earners who had annual earnings above $1 million. By 1998 the number with more than $1 million had increased to 46,700 wage earners. In 2005 it was 82,100.

In 1990 the Social Security Administration found 739 earners with an annual wage above $5 million. By 2005 this number had risen to 6,746, more than nine times the number earning this amount 15 years earlier. In 2005 2,133 workers earned at least $10 million per year.[2] This is about three times as many people as the number who reported earning $5 million back in 1990.

Mr. Reynolds can spin the statistics any way he wants, but these numbers clearly show a dramatic increase in the labor incomes of America’s top earners. By most people’s reckoning, the numbers also show a noticeable increase in income inequality. The income gap between the very rich and the middle class got bigger.

Some commentators ask the question “So what?” By this they mean “Why should we care if the rich are getting richer much faster than everyone else?” It’s a good question, but it is not the question posed by Mr. Reynolds in his Cato policy analysis piece (“Has U.S. Income Inequality Really Increased?”).

Nonetheless, “So what?” is a question worth considering. One answer is that distributional statistics help us to understand how the benefits from prosperity are distributed across a population. Since Mr. Reynolds prefers to use 1988 as his base year for analysis, let’s think about how the benefits of U.S. prosperity have been divided since that year. According to the most recent reports from the Bureau of Economic Analysis, per capita GDP measured in constant chain-weighted prices increased about 31% between 1988 and 2004. This gain translates into an annual rate of increase in U.S. average income of 1.7%. That rate of improvement, which is higher than the one experienced by many other rich countries, is often seen as a vindication of modern American capitalism. In other countries, where they may speak French or German, growth rates have been slower, perhaps because these countries have an inexplicable fondness for over-generous unemployment benefits and rigid worker protection laws (or so it is said).

The vindication of American economic institutions does not look so convincing if it turns out that only a few U.S. citizens enjoyed income gains as high as 1.7% a year. In a paragraph above, I mentioned that household-size-adjusted disposable income for the median American increased 13% between 1988 and 2004. At the 75th percentile, real income increased 16%, and at the 95th percentile, it increased 27%. These numbers translate into annual growth rates of 0.7% for the median American, 0.9% a year at the 75th percentile, and 1.5% a year at the 95th percentile. If we could accurately calculate the rate of income improvement at the 99th percentile or the 99.99th percentile, our calculations would almost certainly show rates of gain that are much faster than 1.5%.

There are many reasons that measured income growth in the middle of the distribution was slower than the growth of GDP per capita. For example, some income and consumption gains enjoyed by Americans are not measured by the Census Bureau’s CPS survey. (Health care consumption, for example, is largely unmeasured by the survey.) One reason the median American has seen comparatively slow income growth, however, is that a disproportionately large percentage of measurable income gains have been enjoyed by people who are well up in the distribution.

If you happen to occupy a position at the top of the distribution, you might very well ask the question “So what?” But you should not defend American institutions by saying they have contributed to income growth in the United States that is faster than the growth in other, less favored countries. At the very least, you should first check to see whether the relative performance of median income growth corresponds with that of average income growth. If half or more of American families see their incomes grow 0.5% a year or less, it can be little comfort for them to know that average income growth in the United States exceeds growth in, say, Germany or France. For a middle-income American, the more relevant question is “How have middle-class Americans fared in comparison with middle-class Germans or Frenchmen?” On that score it is less clear whether the American economy has outperformed the economies of Western Europe.

I do not claim that median income has grown faster in Europe than the United States. I don’t know enough about the facts to hazard a guess. People who disparage income distribution statistics should recognize, however, that the statistics have practical consequences for evaluating economic performance. From the perspective of an ordinary citizen, rapid growth in average income is not a strong argument for American institutions unless the growth is reflected in a discernable improvement in the person’s material well-being. When incomes grow more unequal, the benefits of economic growth will become less obvious to poor and middle class citizens.

Notes

[1] Readers can confirm these estimates by checking the hourly earnings tabulations of the Economic Policy Institute, posted at http://www.epi.org/datazone/06/wagecuts_all.pdf. The EPI examines hourly wage reports of respondents to the Census Bureau’s monthly Current Population Survey.

[2] The Social Security Administration’s tabulations of the W-2 earnings data for 2005 are reported here. The Social Security Administration uses current dollars to measure the income thresholds in its tables. Therefore, the numbers reported in the text do not make an adjustment for inflation. Even making an inflation adjustment has little effect the basic finding that income gains at the very top of the earnings distribution were much faster than the gains in the middle and bottom of the distribution.

I had hoped to move on to new issues, but that will have to wait as I want to respond to some of what Alan Reynolds says in his reply essay.

In his reply, Reynolds provides a very good example of what Gary Burtless points to when he says:

The problem is, he is strongly critical of data series that do not support his views… Rather than do the hard work needed to measure the effect of particular data problems, he cherry-picks evidence to attack researchers whose results he finds displeasing. … Reading his analysis, one is struck by how much it resembles a lawyer’s brief rather than an even-handed weighing of evidence.

I made a similar point:

[T]here are potential measurement issues that work in both directions … and a fair presentation of the evidence would note both sides. But that is not what we have. Only those adjustments that favor the proposition that rising inequality is a myth are presented…

Reynolds’ reply shows how he reacts when presented with evidence that does not conform to the inequality story he has been trying to sell. In my reply to Reynolds’ lead essay, I noted research that finds data adjustments that work against Reynolds assertions on inequality. Reynolds’ response to that evidence? He says:

Citing a newspaper article, Thoma opines that treating R&D expenses as an investment would “dwarf the kinds of adjustments Reynolds discusses.”

He then goes on to dismiss the work. In doing so, Reynolds wants us to believe it is just a “newspaper article” that I am citing, nothing more than a “red herring,” as he calls it later, so we should not give it much credence. But this is an example of how he attempts to mislead readers.

Yes, I did link to an article in the New York Times: a well-reported article about the work of a team of economists at the Bureau of Economic Analysis, and “two prominent economists at the Federal Reserve.” The article does a nice job of summarizing their work. Behind the “newspaper article” is a body of academic research supporting the results I cited. But Reynolds doesn’t want you to know that. For example, in his reply he doesn’t cite or even mention any of the economists or their research, instead he says, “The article claims that …” as though the reporter is the one who had done the work and is making the claims.

He also mixes up the two studies in his rebuttal. Reynolds says:

The article claims that “when R&D is counted as profit, the employee compensation share of national income drops by more than one percentage point,” and says this would somehow shrink labor’s share from 65% in the sixties … to “less than 60 percent today.” But investment is not profit and one percentage point is not five.

The more than 1% figure is based upon the work of the Bureau, while the 5% figure is based upon the work of the economists at the Fed. Reynolds acts as though the reporter implies one causes the other when it’s very clear from the article that these are separate pieces of work. For example:

The Bureau of Economic Analysis … is on the case. So are two prominent economists at the Federal Reserve. … When R & D is counted as profit, the employee compensation share of national income drops by more than one percentage point. … Sumiye Okubo, an associate director of the [Bureau of Economic Analysis], … is directing the experimental project.

The two Fed economists — Carol A. Corrado and Daniel E. Sichel — along with an outside collaborator, Charles R. Hulten, a University of Maryland economist, go much further than Ms. Okubo and her team in arguing that the G.D.P. data should be revised. …

In a recent research paper, “Intangible Capital and Economic Growth,” they … say that this treatment should be extended to a host of other investments… If these …[adjustments] were incorporated into G.D.P. …, labor’s share of national income would decline from a fairly steady 65 percent in the 1950′s, 60′s and 70′s to less than 60 percent today. The long decline doesn’t show up in the standard G.D.P. accounts, which ascribe nearly 65 percent of national income to labor.

Nothing Reynolds says in his confused reply rebuts this work.

I do agree with Reynolds in the following sense. There is a big difference between peer-reviewed work in academic journals and newspaper articles — though, as illustrated above, accurate newspaper reports can be very helpful. And to take this a step further, there is an even bigger difference between the peer-reviewed academic work that forms the core of the inequality results cited by professional economists and work appearing on the opinion pages of newspapers. While good work does appear on those pages from time to time, there is nothing to ensure that the claims made in a typical opinion piece are backed by evidence that can withstand professional scrutiny.

That’s why I was surprised to see Reynolds, after implicitly dismissing the work of academic economists as a mere “newspaper article,” say in a different part of his reply that “I have written hundreds of articles since 1972 about income, or about wealth” in an attempt to bolster his own credentials. But how many of those appeared on the opinion pages of newspapers (and predominantly one or two newspapers at that), and how many are peer-reviewed articles appearing in top flight academic journals, or in any academic journal for that matter? When you read Alan Reynolds’ assertions, remember that they have not been subjected to the rigorous process required to get a paper published in a top academic journal, they have only passed through an editorial board at a newspaper. He’s right to suggest that we be wary of such work.

Here’s another example of a misleading presentation by Reynolds. In discussing his claim that the treatment of tax deferred income distorted the inequality statistics, I noted that the important point is that most Americans have so little capital income exactly how you count it is not an important issue.

What is Reynolds’ response to this? First, he makes a big deal out of the words used to describe the income – he didn’t like the use of the word capital income and wanted “corporate profits” to be used instead, even though the issue is the income of the top 1% of Americans. That’s fine, call it what you want; it doesn’t change the point.

Second, the statistics he cites are misleading. They are the total allocated to the top group in each year, not the amount of the mismeasurement he is asserting exists. That is, the 39% to 59% increase he cites says nothing by itself. That is simply the amount of corporate profits allocated to the top income group. That figure says nothing at all about whether the growth in capital income reflects the actual change in asset holdings for this group. Though he tries to make a case that it doesn’t reflect underlying asset holdings, when you look at this, you see there has been substantial growth in asset holdings for this group, and therefore the misallocation does not appear to be very large if it is there at all.

But the main thing to notice is how Reynolds uses misleading statistics in an attempt to support his case. Saying, as he does, that “The CBO added 39% of corporate profits to top 1% incomes in 1989 and 59% in 2004, thus fabricating a wholly artificial increase in the top 1 percent’s share” implies that all of the increase represents a misallocation when that’s not the case at all. It is the change from 39% to 59% relative to the change in asset holding across groups that matters, but those numbers are not given. If they were, they would be small.

Now let me turn, quickly, to a few more of Reynolds’ responses. In my reply to Reynolds’ lead essay, I noted that not all Gini coefficient results are consistent with his assertions and I pointed to some work that supports different conclusions. He says:

Amazingly, [Thoma] refers only to a Gini coefficient for wealth — as if income doesn’t matter. After ignoring all income statistics, he accuses me of “an incomplete presentation of the evidence” and “attempts to cloud the issue.”

The point was that contrary evidence exists on both the wealth and income inequality issues, both of which are discussed in his essay, and that he does not tell us about contrary findings. This provided an example of such evidence, it was not an attempt to provide a systematic survey of all work in the area. Exactly how adding additional information about research on inequality clouds the issue, except to the extent that it clouds the issue because it doesn’t agree with the conclusion he wants to sell, is not clear.

Another point concerns Bernanke’s speech on inequality. Reynolds’ objection is that Bernanke did not use the measure of inequality he wants him to use (the one that supports his conclusion), but instead uses another measure, which shows growing inequality. Of course, in Reynolds’ eyes, the data Bernanke used is “bad data” that is “no substitute for good data.” But we know that the finding of rising inequality is quite robust across inequality measures, so this is not much of a refutation. And on Reynolds’ “good data” — the data that support his conclusion — it’s useful to recall what Gary Burtless said about some of the data Reynolds uses:

The problem is [Reynolds] is usually silent about equal or more serious problems with data sets that show little change in inequality. … What Reynolds doesn’t mention is that the quality of the consumption data has deteriorated badly since the mid-1980s. … So far as I know, no statistical series that tries to approximate total income has suffered such a terrible decline in quality as the data from the consumption survey. You’ll look long and in vain for any mention of this problem in Reynolds’s paper.

If you only look at evidence on one side of the issue, cherry pick results, start in specific years (and insist everyone else follow suit), use the “right” measures of income or wealth, ignore data problems that work against your results, and so on, and so on, you might be able to argue, if everything falls in your favor, that inequality is no worse since 1988. But that does not fairly characterize the overall evidence.

This debate reminds me of the debate over global warming, though using the word “debate” implies there is more disagreement than there really is. There are three questions in the global warming debate. The first question is whether global warming exists. The second question is, if it does exist, what is causing it. The third question is what to do about it. In order to avoid the consequences involved with the third step, doing something about it, there are many who try to cloud the issue and keep the first question alive and kicking for as long as possible, or claim the cause is from natural forces that we can do nothing about.

The inequality debate appears to be unfolding similarly with those who would like to avoid policies to address inequality, policies such as more progressive taxation, hoping to keep the first question open as long as possible or claiming that the rise in inequality is the inevitable result of natural market forces and we should not interfere.

There is a role for skeptics, but there is also a time to accept that the preponderance of evidence points in one direction and to begin to think about and implement corrective measures. I believe an important question is how we respond to inequality – will it be through progressive taxation, minimum wage legislation, changes in the structure of health care, investments in education and retraining programs, wage insurance and so on, or will we do nothing?

The question of what to do is linked to the causes of rising inequality. Has inequality been rising because of tax policy, the decline in unions, the rising skill premium, global competition and changing technology, a falling minimum wage in real terms, or for other reasons? How much does each factor contribute? Is the income of those who have experienced the largest gains based upon economic fundamentals, i.e. does their pay reflect their contribution to production, or does the pay of, say, CEOs depend upon market failures that allow departures from competitive market outcomes?

There are lots and lots of important questions to be answered involving both equity and efficiency (many of which do not require rising inequality since 1988, just its existence) and, as I said in my first essay, it will be too bad if attempts to cloud the issue divert us from discussing how best to respond to income and wealth inequality.

It strikes me that there is often an inverse relationship between the heights of an essayist’s rhetorical flourishes and the depth of the scientific evidence they marshal to support it. I give Mark Thoma high marks for raising the rhetorical stakes in this debate to the point where those who remain more skeptical of the evidence for substantial increases in income inequality since 1989 are relegated to the status of contrarian ideologues unwilling to sign onto Mr. Gore’s “Inconvenient Truth.” But isn’t this a bit much, even for a true believer?

Rather than raise the rhetorical level it might have been more interesting for Thoma to acknowledge the new evidence I provided in Figure 6. It demonstrates how changes in top coding and censoring both in the public use and in the internal Current Population Survey (CPS) data significantly inflate inequality trends, especially in the 1990s. That is, they are much higher relative to trends based on that same data which has been consistently top coded; in each year, each income source is top coded at the same point in its distribution.

This standard practice in the labor earnings and income inequality literature shows that, at least as far as the CPS data are concerned, there has been little increase in income inequality in the bottom 99 percent of the income distribution since 1989 as measured by standard Gini coefficients. And much of that increase is likely due to changes in US Census data collection processes that show up in the 1993 internal CPS data and in the 1996 public use CPS data.

But the news is even better: unlike the business cycle of the 1980s, where a small but significant share of the income distribution got worse (see Figure 2) over the 1990s business cycle (1989-2000), the entire distribution moved to the right. A person at every point in the distribution in 2000 was better off in real terms than his counterpart at that point in the distribution in 1989.

That is something that did not happen in Germany or Japan, two countries with much bigger and more progressive income tax systems and a willingness to interfere with markets to reduce income inequality greater than that in the United States. Despite their progressive efforts, income inequality increased more in both these countries over their 1990s business cycles than in the United States. This fact should at least give Thoma, and others who are so ready to make things better though redistributive policy, reason to pause and to consider whether their proposed changes would really do more good than harm.

As for Reynolds and Burtless, I continue to suspect that there is fundamentally little difference between my findings and their findings with respect to the bottom 99 percent of the income distribution. And I continue to suspect that our differences with respect to the way inequality within that population changed over the last 25 years is less than our conversation to date would suggest. But it will require a painstaking comparison of footnotes to be sure. But that should not be surprising: this kind of detailed work is what careful research done on imperfect data sets requires.

Burtless puts great emphasis on a break point between 1979-1994 (the start of the strong economic expansion) and 1994-2004 in his discussion. But my reading of Figure 6 (see especially the consistently top coded internal Gini values which I believe are the most appropriate to consider) suggests that inequality rose dramatically from 1979-1983 and much less thereafter. This is not a trivial observation. 1979-1983 was a difficult time in the United States. It began with a couple of years of double-digit inflation and ended with the most serious recession since the Great Depression.

In my view this was the price our economy paid for a decade of failed Keynesian policies and stagflation. However, this dark period of macroeconomic outcomes also marked the end of federal government macro policies based on Keynes’s economic principles. The Reagan Administration put into place a series of macroeconomic policies practiced by all subsequent presidents, both Republican and Democrat: the Federal Reserve Board Chairmen they have appointed have focused primarily on an inflation target, and the federal government has increasingly allowed free markets to work their magic.

What has happened to income inequality since then? My reading of Figure 6 is that there was little change in income inequality from 1983 to 1992. There is then a major spike in the data in 1992-1993, which to some degree is caused by changes in US Census data collection procedures. From 1993 to 2004 there is no change in consistently top coded internal CPS data. Hence for the bottom 99 percent of the income distribution — except for the largely unexplained spike in 1992-1993 — there has been remarkably little increase in income inequality.

One can certainly argue that there has been no decreases in income inequality either. Hence, since 1983 the United States has had a substantially higher, but relatively constant, level of inequality than we experienced in previous decades. But during the past 20 years (at least for the bottom 99 percent of us) people at all points on the income distribution have experienced increases in economic well-being with little additional increase in income inequality.

So what’s all the shouting about? In my view it is about what the CPS and especially consistently top-coded CPS data can not directly tell us. What about that other 1 percent? Here Reynolds and Burtless strongly disagree. Reynolds argues that even including that 1 percent in the mix there has been no great increase in income inequality since 1988 and Burtless offers counter proof based on other data sets. I am not yet sure who is right between them, but I am very sure that Thoma is wrong. It is still reasonable to be skeptical of both arguments.

It strikes me that there is often an inverse relationship between the heights of an essayist’s rhetorical flourishes and the depth of the scientific evidence they marshal to support it. I give Mark Thoma high marks for raising the rhetorical stakes in this debate to the point where those who remain more skeptical of the evidence for substantial increases in income inequality since 1989 are relegated to the status of contrarian ideologues unwilling to sign onto Mr. Gore’s “Inconvenient Truth.” But isn’t this a bit much, even for a true believer?

Rather than raise the rhetorical level it might have been more interesting for Thoma to acknowledge the new evidence I provided in Figure 6. It demonstrates how changes in top coding and censoring both in the public use and in the internal Current Population Survey (CPS) data significantly inflate inequality trends, especially in the 1990s. That is, they are much higher relative to trends based on that same data which has been consistently top coded; in each year, each income source is top coded at the same point in its distribution.

This standard practice in the labor earnings and income inequality literature shows that, at least as far as the CPS data are concerned, there has been little increase in income inequality in the bottom 99 percent of the income distribution since 1989 as measured by standard Gini coefficients. And much of that increase is likely due to changes in US Census data collection processes that show up in the 1993 internal CPS data and in the 1996 public use CPS data.

But the news is even better: unlike the business cycle of the 1980s, where a small but significant share of the income distribution got worse (see Figure 2) over the 1990s business cycle (1989-2000), the entire distribution moved to the right. A person at every point in the distribution in 2000 was better off in real terms than his counterpart at that point in the distribution in 1989.

That is something that did not happen in Germany or Japan, two countries with much bigger and more progressive income tax systems and a willingness to interfere with markets to reduce income inequality greater than that in the United States. Despite their progressive efforts, income inequality increased more in both these countries over their 1990s business cycles than in the United States. This fact should at least give Thoma, and others who are so ready to make things better though redistributive policy, reason to pause and to consider whether their proposed changes would really do more good than harm.

As for Reynolds and Burtless, I continue to suspect that there is fundamentally little difference between my findings and their findings with respect to the bottom 99 percent of the income distribution. And I continue to suspect that our differences with respect to the way inequality within that population changed over the last 25 years is less than our conversation to date would suggest. But it will require a painstaking comparison of footnotes to be sure. But that should not be surprising: this kind of detailed work is what careful research done on imperfect data sets requires.

Burtless puts great emphasis on a break point between 1979-1994 (the start of the strong economic expansion) and 1994-2004 in his discussion. But my reading of Figure 6 (see especially the consistently top coded internal Gini values which I believe are the most appropriate to consider) suggests that inequality rose dramatically from 1979-1983 and much less thereafter. This is not a trivial observation. 1979-1983 was a difficult time in the United States. It began with a couple of years of double-digit inflation and ended with the most serious recession since the Great Depression.

In my view this was the price our economy paid for a decade of failed Keynesian policies and stagflation. However, this dark period of macroeconomic outcomes also marked the end of federal government macro policies based on Keynes’s economic principles. The Reagan Administration put into place a series of macroeconomic policies practiced by all subsequent presidents, both Republican and Democrat: the Federal Reserve Board Chairmen they have appointed have focused primarily on an inflation target, and the federal government has increasingly allowed free markets to work their magic.

What has happened to income inequality since then? My reading of Figure 6 is that there was little change in income inequality from 1983 to 1992. There is then a major spike in the data in 1992-1993, which to some degree is caused by changes in US Census data collection procedures. From 1993 to 2004 there is no change in consistently top coded internal CPS data. Hence for the bottom 99 percent of the income distribution — except for the largely unexplained spike in 1992-1993 — there has been remarkably little increase in income inequality.

One can certainly argue that there has been no decreases in income inequality either. Hence, since 1983 the United States has had a substantially higher, but relatively constant, level of inequality than we experienced in previous decades. But during the past 20 years (at least for the bottom 99 percent of us) people at all points on the income distribution have experienced increases in economic well-being with little additional increase in income inequality.

So what’s all the shouting about? In my view it is about what the CPS and especially consistently top-coded CPS data can not directly tell us. What about that other 1 percent? Here Reynolds and Burtless strongly disagree. Reynolds argues that even including that 1 percent in the mix there has been no great increase in income inequality since 1988 and Burtless offers counter proof based on other data sets. I am not yet sure who is right between them, but I am very sure that Thoma is wrong. It is still reasonable to be skeptical of both arguments.

I opened the discussion by presenting evidence suggesting little change since 1986-88 in the inequality in disposable income or consumption “among the U.S. population as a whole” (as opposed to, say, the 99.99th percentile doing better than the 99th percentile). One unique piece of that evidence — comparing growth of median income by quintile and decile from the Fed’s in-depth Survey of Consumer Finances — has attracted no comment. Instead, two commentators who might be expected to rely on Gini coefficients have resorted to comparing mean income or wages between top and bottom deciles, before taxes, and doing so for only two years.

Nobody, least of all Krueger and Perri, questioned my suggestion that consumption inequality has remained largely unchanged for many years, although Burtless did raise some concerns about data quality which I tried to resolve.

In the 2005 Consumer Expenditure Survey the bottom fifth accounted for 8.2% of total consumer spending, compared with 39% for the top fifth. But there were only 1.7 persons and 0.5 workers per “consumer unit” in the bottom fifth, compared with 3.2 persons and 2.1 workers in the top fifth. Consumption per capita or per worker is much more equal than the consumption shares indicate, and consumption shares are much more equal than income shares (particularly if taxes and transfer payments are excluded).

What about inequality of disposable income? The initial essay by Burtless said, “if you look at some of the most comprehensive definitions of income, it turns out that inequality increased less, possibly much less, after 1989 than indicated by the Census Bureau’s headline number.” Burkhauser likewise concluded that, “Since 1989 household income inequality has risen very little.” Aside from the one-year spike in 1993, in fact, he finds “remarkably little increase in income inequality” for the bottom 99% of the population since 1983. Bernanke mentioned disposable income but started with 1979 and used the wrong data. Thoma has not questioned any income distribution statistics offered by Burkhauser or me. So where is the disagreement?

Any remaining controversy about total income or labor income (“wages”) appears to have been narrowed to small fractions of the population (1-10%) rather than the population as a whole. If gains among the top 1-10% had a significant impact on middle- class shares of disposable income, however, that would have shown up as a rising Gini coefficient. That did not happen, as I showed graphically.

There is still considerable misunderstanding about the main reason that Henderson and I questioned the CBO’s misuse of income tax data to estimate disposable income of the top 1% (Piketty and Saez do not estimate disposable income). I will deal with those misunderstandings in a separate note, and with Thoma’s concerns about wealth inequality.

This comment begins with the 90/10 ratio – the ratio of average income or wages of the top 10% to the income or wages of bottom 10%. Such calculations exclude 80% of the population and usually exclude transfer payments and taxes, so they do not actually relate to the issue as I described it – namely, inequality of disposable income “among the U.S. population as a whole.” Since 90/10 ratios have nonetheless been presented as evidence that inequality of pretax income or wages has supposedly increased, the data demand a closer look.

Krueger and Perri write that 90/10 ratios “have the desirable properties that are not affected by changes in top-coding procedures.” Using the CES survey, they conclude that “the 90/10 ratio increases from around 5 in 1989 to around 6 in 2003.” After examining internal data from the Current Population Survey, however, Burkhauser, Feng and Jenkins discovered that 90/10 ratios based on public use files are seriously affected by top-coding. That is likely true of CES data too. Top-coding of 2004 CES public use files applies to wage income above $150,000, for example, and any higher wage is replaced with a mean average of all wages above that critical value.

There are other problems with using a consumption survey to estimate income. The small sample (7500) is made smaller and less random by the fact that some respondents refuse or neglect to report their income. Income is included in the data if respondents provide values for even one source, but many people have additional unreported income from more than one source (such as savings or transfer payments).

Burtless uses a 95/50 ratio for wages alone – that is, changes between two years in the wages of the top 5% of workers and median workers, whether full-time or part-time. He writes, “In 1988 – Reynolds’ preferred base year – a wage and salary worker who earned the median hourly wage received $13.20 an hour (measured in constant 2005 dollars). By 2005 the median worker’s wage increased to $14.29 an hour, an increase of 8.3%. Over the same span of years, a worker earning the 95th-percentile wage experienced a 20.3% gain in pay.”

I have no “preferred base year.” On the contrary, I strongly object to such two-year comparisons because they do not reveal what happened when. One reason, as I have repeated many times, is the break in the data in 1993 when the Current Population Survey (CPS) began to include higher incomes and the income share going to the top 5% suddenly spiked as a result. That same problem is likely to affect the “hourly wage cutoffs” that Burtless cites from the Economic Policy Institute (EPI), because those estimates are based on the “authors’ analysis of CPS wage data.” This is the same data Janet Yellen relied on in the speech Thoma cited.

The EPI’s 95/50 ratio was unchanged at 2.6 from 1985 through 1993, when it suddenly jumped to 2.8 in two years because, as the EPI explained elsewhere, “a change in survey methodology in 1993 led to a sharp rise in measured inequality.” The 95/50 ratio has fluctuated between 2.8 and 2.9 ever since. The Burtless comparison of 1988 and 2005 does not reveal that the ratio was flat before 1994 and after 1995.

Unlike Krueger and Perri’s 90/10 ratio for income, the EPI’s 90/10 ratio for wages shows no sustained increase since 1986. The EPI 90/10 ratio has fluctuated narrowly between 4.3 and 4.4 since 1986, with the familiar CPS spike from 1992 to 1994 and touching 4.5 twice since then. The increase since 1994 was partly because, as the EPI notes, “changes to the survey in 1994 led to lower reported earnings for low-paid workers.” In any case, the ratio was 4.4 in 1987 and 4.4 in 2004, so I too could play the two year game and say that proves there has been no increase in inequality since 1987.

Click on Burtless’s link to the EPI table and see for yourself that all of the years from 1987 to 2005 show no significant and sustained trend in the 90/10 ratio for wages. If Krueger and Perri’s CES data show an increase in the 90/10 ratio of income (not just wages), such different results may be because consumption by the bottom decile is mainly financed by uncounted transfer payments rather than wages.

The EPI estimates do not describe average wages of the top 5-10%, but only the “cutoffs” or minimum thresholds defining where the 90th or 95th percentile begins. They estimate that in 2005, for example, you had to make at least $41.70 to be in the top 5% of wage and salary workers. That figure is, as Burtless noted, is up 20.3% from 1988. But that cutoff or threshold was not pulled up from above (the EPI excludes wages above $100) — it was pushed up from below.

Third Way, a progressive think tank, notes that “From 1979 to 2005, the percentage of prime-age households earning over $100,000 in current dollars grew 12.7 percentage points, while those earning between $30,000 and $75,000 shrank 13.3 percentage points.” That huge increase in the percentage of “rich” households explains why it takes higher earnings than it used to in order to still be included in the mean average of top 5% incomes. With too many people crowding the EPI threshold from below, the threshold moved up. The bar was raised by a general increase in the percentage of workers earning high incomes. And because mean income above the higher threshold no longer encompasses lower earnings of $36-40 an hour that used to be included, a mean average of earnings above the higher threshold was bound to increase. I call this “threshold illusion” in my book, and explain it as follows:

A rising mean income among the top 5, 10 or 20 percent has been routinely misinterpreted as indicating that income gains were confined to only that top group. In reality, rising incomes among those with incomes below the rising threshold have caused the definition of top income groups to exclude incomes that had formerly been among the top group. The mean average of income in top income groups can be pulled up by a few unusually high incomes at the top. But the average can also be pushed up from below by rising numbers of people moving up — leaving what used to be considered a “middle class” income and “joining the ranks of the rich.”

In most of the income distribution data we have been discussing, such as Gini coefficients or the related income shares by decile or quintile, the income of top groups is described by a mean average. We just add up all the income above some threshold and then divide by the number of households, taxpayers or consumer units. Mean averages mislead in this case too, just as they do for total income.

New York magazine’s 2004 survey of Manhattan incomes identified a famous hedge fund manager who earned $1.02 billion. For that same year, the Census Bureau defined the top 5 percent of households as everyone earning more than $157,185. Blending together all incomes from $157,195 to $1,020,000,000 and then dividing that total by the number of households (113,146,000) produces a hodge-podge “average” of $264,387. But such a mean “average” tells us nothing about typical incomes of those 5.7 million households.

Any mean average of income for the top 1-10% is greatly distorted by a small number of outliers, unlike mean averages for all other fractiles which are bounded by an income ceiling. The table below shows that mean and median incomes are virtually identical for the bottom four quintiles. For the top 10%, however, mean income is 64-66% larger than median income. The last column, the source of Figure 7 in my January 11 presentation, shows that real median income of the top 10% increased by 20.7% from 1989 to 2004 – before taxes. Yet real median income of the bottom two quintiles also increased by 20-21%, although in-kind transfers are excluded.

It is important to note that for the top 10% alone, the increase in real income from 1989 to 2004 was smaller for mean income (18.5%) than for median income (20.7%). If the growth of top 1 percent incomes had been nearly as great as suggested by Piketty and Saez or the CBO, then mean income of the top 10% would have grown much more rapidly than mean income, rather than the other way around.

Median income for all households was $43,200 in 2004 according to the SCF, and mean income was $70,700. Everyone would rightly argue that it would be extremely misleading for me to say the higher mean income describes an “average” household. I argue that is equally misleading to say that a mean average describes an “average household” within the top 10 percent. Why am I wrong?

Although I have followed academic convention by presenting mean income figures for the top 5 percent’s income share, and Gini coefficients based on mean income after taxes and transfers, I find such uses of mean averages inherently biased toward exaggerating the typical level of income among top income groups.

Even using conventional measures of mean income, however, no evidence has yet been presented to show any significant and sustained increase in inequality of disposable income among the U.S. population as a whole. That one-time 1993 spike in Census income keeps popping up, and the 1986 spike in capital gains realizations, but that is about all we have seen so far, aside from comments about my temperment or assumed policy agenda. Has anyone contemplated the possibilty that I might simply be right?

I am pleased to see that Alan Reynolds is finally taking a closer look at some of the evidence that works against his claim that inequality has been stagnant in recent decades, though he predictably dismisses it. I will not convince him the evidence is valid, and he most certainly has not convinced me that it isn’t, so I encourage anyone who is still puzzled about the evidence that profits have been mismeasured and that it matters for assessing changes in inequality in recent years to look at the research and draw their own conclusions. I have no doubt that a fair reading of the evidence will lead to the conclusion that inequality may in fact be worse than we thought which runs opposite of Reynolds’ claims. The essence is fairly simple, if we’re mismeasuring real investment, then we are also mismeasuring profits. Given the concentration of corporate ownership at higher incomes, and the extent of the mismeasurement, this correction matters.

One additional note on Reynolds’ responses since my last post, then I’d like to move on to other issues, particularly those raised in the contributions others have made to this debate. Given Reynolds wholly unsubstantiated and uncalled for attack on the ethics of Piketty and Saez in a commentary on the opinion pages of the Wall Street Journal where he accuses them of fabricating results in academic journals to support an agenda, and given other things he has said at other times, it made me chuckle to see him say in his latest response that “all we have seen so far” are “comments about my temperment or assumed policy agenda” as though no evidence rebutting his stance has been presented, just personal attacks on his character or charges that he’s pursuing an ideological agenda. Reynolds says there’s not a strand of evidence that he’s wrong (“no evidence has yet been presented to show any significant and sustained increase in inequality”). Not a strand he’ll acknowledge anyway, but as has been pointed out in previous posts here, and has been documented elsewhere in many different ways, there’s overwhelming evidence against his claims.

I want to follow up on the post from Dirk Krueger and Fabrizio Perri (“Inequality in What?”) because I think they bring something important to the discussion, the academic underpinnings of how we approach the measurement and assessment of inequality changes. In their introduction they say:

Inequality is a fascinating subject, one that provokes discussion and makes it hard to settle the apparently simple question of whether income inequality in the US has increased since 1988. …. Our main point … is to argue that to focus only on the evolution of current income inequality is insufficient if one is interested in the evolution of the distribution of living standards in the U.S.

They then go on to explain, correctly, that the academic literature does not support looking at current income to measure inequality, a broader lifetime measure of consumption and leisure opportunities must be considered, i.e. some concept of the present value of expected lifetime utility is needed (the data on current consumption Reynolds uses in some of his arguments is one proxy for lifetime resources under permanent income stories of consumption, but it is an imperfect proxy with acknowledged problems making any results from these data difficult to interpret reliably). Their conclusion is worth repeating:

One conclusion we would … like the readers to take home is … that understanding the welfare effects of changes in measured inequality, and possibly the appropriate policy measures to deal with it, is a complex task that involves more than reporting the distribution of current resources. Ideally one should understand and measure the distribution of lifetime resources. In order to understand how lifetime resources translate into observable indicators, and what these indicators are, it is crucial to have a thorough understanding of how and to what extent households can transfer resources through time and across states of the world using financial markets. Our own previous work has highlighted the importance of using consumption as an indicator, but recent exciting work is being done by leading researchers in the economics community stressing the role of inequality and dispersion in other variables, too, such as labor effort or wealth, and assessing their impact on incentives, the allocation of resources, and the distribution of welfare.

This is the point I’d like to follow up on because it gets at the essence of Reynolds’ point, measurement issues. What does the academic literature tell us about measuring inequality and has the debate as presented here conformed to those standards?

There is a considerable body of work on measuring inequality, so I will only scratch the surface and point to a couple of surveys on the topic and highlight some of the key results that relate to the discussion here. Perhaps as others respond they can add links to additional resources people can use to learn more about what the academic literature says about measuring changes in inequality.

One such resource is Some New Methods for Measuring and Describing Economic Inequality, by R. L. Basmann, K. J. Hayes, and D. J. Slottje which was reviewed by John A. Bishop in the Journal of Economic Literature in June, 1995. Since much of the evidence presented by Reynolds and by Richard Burkhauser in this debate has been in the form of Gini coefficients, let me quote one passage from the review (e.g. Burkhauser’s argument that I am wrong about the persuasiveness of the overall evidence relies on Gini coefficients):

First, the authors stress the shortcomings of cardinal evaluations of inequality that rely upon a single index such as the Gini coefficient. They argue that there is no one best inequality index –each contains an implicit set of distributional weights and the precise weights underlying many familiar indices, if clearly understood, would probably not be widely accepted by policy makers.

Another summary of techniques for measuring inequality can be found in Frank Cowell’s Measuring Income Inequality, 1995. Two more collections of the work on measuring inequality are given in The Handbook of Income Inequality Measurement, by Jacques Silber, 2000, which is reviewed by Charles Beach in The Journal of Economic Literature, and in The Handbook of Income Distribution, edited by A.B. Atkinson and F. Bourguignon, 2000. Let me also be sure to recommend my colleague Peter Lambert’s The Distribution and Redistribution of Income, 2002.

These surveys and texts discuss topics such as using stochastic dominance techniques along with Lorenz curves to assess inequality, equivalence scaling, parametric and non-parametric approaches to measurement, welfare comparisons, allowing for different family sizes and compositions, horizontal versus vertical inequality measurement, intertemporal measurement of inequality, and all sorts of other important theoretical and measurement issues you won’t generally find in popular discussions of the topic and that have not, for the most part, been a part of the evidence and discussion Reynolds has presented.

There has been quite a bit more work since these overviews were published, but they constitute a good introduction to many of the important theoretical and empirical issues in the debate over inequality. But as an example of more recent work, another good source of information on this topic is the Journal of Economic Inequality which, coincidentally, has a paper in its latest issue about an issue Reynolds is worried about: robust estimation in the presence of contamination of data in the tails of distributions (note that Reynolds is critical of

Krugman’s efforts to look at data contamination at the top of the income distribution that follow roughly along the lines suggested in approach (2) of this paper):

Robust stochastic dominance: A semi-parametric approach, by Frank A. Cowell & Maria-Pia Victoria-Feser, Journal of Economic Inequality, April 2007:

Abstract Lorenz curves and second-order dominance criteria, the fundamental tools for stochastic dominance, are known to be sensitive to data contamination in the tails of the distribution. We propose two ways of dealing with the problem: (1) Estimate Lorenz curves using parametric models and (2) combine empirical estimation with a parametric (robust) estimation of the upper tail of the distribution using the Pareto model. Approach (2) is preferred because of its flexibility. Using simulations we show the dramatic effect of a few contaminated data on the Lorenz ranking and the performance of the robust semi-parametric approach (2). …

And that’s just one paper in the most recent issue on a single journal, there is a considerable volume of work on these issues. The important thing to realize from all of this is that no single measure of inequality is perfect. Thus, looking at a variety of measurements using a variety of data sets and state of the art techniques, and being fully aware of and acknowledging the shortcomings of the process at every step along the way so that results can be interpreted properly is important in establishing how inequality has changed through time, and in presenting a balanced overview of the results. When researchers go through these exercises carefully and weigh the evidence objectively they conclude, with few exceptions, that inequality has been rising in recent years.

It appears from Mark Thoma’s last posting that he is inching his way into the consensus of Reynolds, Burtless, and Burkhauser, which says that it is hard to find much of an increase in household size adjusted income among the bottom 98 or 99 percent of the United States population since the 1980s using standard Gini measures and consistently top coded data from both the public use and internal restricted access Current Population Survey. And, though he doesn’t say so, it also appears he is inching his way toward the view that the gains from economic growth were more equally distributed over the last major business cycle (1989-2000) than the previous one (1979-1989) within this population.

Thoma has posted some very valuable references to the more technical economics literature, which uses alternative measures of income inequality and struggles to extract the most information about the entire distribution from imperfect data by making certain assumptions about the shape of the entire distribution given limited data at the top. He especially notes the problem that outliers at the top of the distribution cause in such efforts. Hence it appears he now recognizes that you sometimes need to be a weatherman (or at least need to carefully listen to several of them) to know which way the wind is blowing.

But instead of then agreeing that the literature on how exactly to capture this very high end of the distribution (the top 1 or 2 percent) is still developing — as is the literature on how exactly to use such measures to tell us about what has happened to income in this very high income population over the last thirty years — he concludes with a most amazing non-sequitur: “When researchers go through these exercises carefully and weigh the evidence objectively they conclude, with few exceptions, that inequality has been rising in recent years.”

A careful reading of the most recent article Thoma provides us makes no such claim:
“Robust stochastic dominance: A semi-parametric approach,” by Frank A. Cowell & Maria-Pia Victoria-Feser, Journal of Economic Inequality, April 2007. Rather, the article is a cautionary tale of the difficulties of making such judgments, and the sensitivity of such finding to outliers at the top of the distribution. While the paper uses British data, similar problems are likely to be found using the CPS and, I suspect, the other data sets we have discussed over the course of our conversation. That is why much of the income inequality literature using the CPS has used “trimming” or consistent top coding to avoid the problem of outliers and instead talks about the bottom 98-99 percent of the income distribution. Pinning down what has happened to the top 1 or 2 percent of the income distribution is the hard work that remains to be done before we can state definitively what has been happening there and how it impacts overall income distribution.

The Wall Street Journal article I wrote with David Henderson, critiquing the way the Congressional Budget Office (CBO) allocates corporate profits, must have been unclear in some respects because the main points have been misunderstood by Thoma and perhaps also Burtless.

In the National Income accounts, a distinction is made between the capital income of persons (included in personal income) and the retained earnings of corporations (included in national income). This distinction is also present in the Piketty-Saez study, and all other income distribution studies except those of the CBO. Yet Thoma depicts the distinction between corporate and personal income as a mere semantic quibble. He notes that I “didn’t like the use of the word capital income and wanted ‘corporate profits’ to be used instead, even though the issue is the income of the top 1% of Americans. That’s fine, call it what you want; it doesn’t change the point.” On the contrary, the table below shows that simply substituting Federal Reserve wealth statistics for the CBO’s invalid proxy reduces the CBO’s estimates of the top 1 percent’s pretax income from 16.3% in 2004 to 11.3%, and turns an apparent increase in that share into a decline.

The CBO is the only agency that attempts to distribute corporate profits among households. They do that because they are trying to estimate who bears what share of the corporate income tax. But that, in turn, requires adopting a theory about the incidence of the corporate tax.

The 1962 theory chosen by the CBO back in the seventies was based on a closed economy, and assumed that none of the corporate tax could be shifted to workers or consumers. The corporate tax was assumed to be borne by domestic owners of capital in general (because of arbitrage) – not just owners of corporate stock, and not just the owners of taxable investments. This old theory now has plenty of critics, including CBO economist William Randolph, who estimates that labor bears about 74% of the corporate tax.

What is critical to understanding the Reynolds-Henderson analysis, however, is to realize that the CBO theory about the incidence of corporate taxes does not suggest that the corporate tax is borne in proportion to income from capital (much less the small fraction of such income that is realized and taxable). Using capital income as a proxy for wealth distribution was simply a statistical shortcut. It is grossly erroneous for the CBO to use taxable capital income as a roundabout proxy for the top 1 percent’s share of wealth, for reasons the Reynolds-Henderson article explains. Besides, we have much better direct estimates of wealth distribution.

Thomas says, “there has been substantial growth in asset holdings for [the top 1%], and therefore the misallocation does not appear to be very large if it is there at all.” Using the same wealth data that Thoma cited we showed the misallocation is enormous. We wrote that, “Kennickell … concluded that the top 1 percent’s share of wealth declined slightly from 34.6% in 1995 to 33.4% in 2004. Yet the CBO says that share rose from 43.2% to 59.4% in those same years.”

Thoma replies that “the 39% to 59% increase he cites says nothing by itself. That is simply the amount of corporate profits allocated to the top income group.” What is the word “simply” doing here? Could anyone possibly imagine that “simply” adding 59% of corporate profits to the top 1 percent “says nothing by itself”? Compared with the Kennickell estimates (which are higher than most), the CBO’s technique inflates the top 1 percent’s income by 25% in 2004.

Thoma goes on to say, “that figure [39% or 59%] says nothing at all about whether the growth in capital income reflects the actual change in asset holdings for this group.” That is an argument with the CBO, not with me. Asset holdings are precisely what these percentages are assumed to reflect, according to the CBO’s incidence theory. The figures 39% and 59% reflect percentages of corporate profits allocated to the top 1 percent in the past and present. That is not the “capital income” of the top 1 percent, as Thoma believes. On the contrary, it is added on top of the interest, dividends, capital gains that taxpayers report. As the graph in Reynolds and Henderson shows, that is why the CBO’s estimates of average incomes of the top 1% are so much larger than those of Piketty and Saez (which do not include corporate profits).

The Table below show what the top 1 percent’s share of combined personal and corporate income – before taxes – looks like after we substitute Kennickel’s direct estimate of the top 1 percent’s share of wealth for the CBO’s invalid proxy. By correcting this one error, the top 1 percent share of pretax income no longer appears to rise from 12.5% in 1989 to 16.3% in 2004 but instead falls from 11.8% in 1989 to 11.3% in 2004.

Before CBO statisticians can estimate what share of the corporate tax is borne by the top 1%, they must first estimate the top 1 percent’s share of all assets — not just corporate assets, and not just taxable assets. To do that, the CBO relies on shares of “capital income” income received from taxable investments as a proxy for the distribution of assets in general, whether those assets are taxable or not. In their own words,“CBO assumes that corporate income taxes are borne by owners of capital in proportion to their income [reported on tax returns] from interest, dividends, capital gains and rents.”

The CBO has surely been missing a rising amount of unseen income of middle-class taxpayers, but (contrary to Thoma) that is not our main point. Our main point is that this causes them to hugely exaggerate the level and trend of average incomes among the top 1%. That, in turn, is due to the untenable way by which they allocate corporate profits — not capital income — by income group.

Thoma says, “The main thing to notice is how Reynolds uses misleading statistics in an attempt to support his case. Saying, as he does, that ‘The CBO added 39% of corporate profits to top 1% incomes in 1989 and 59% in 2004, thus fabricating a wholly artificial increase in the top 1 percent’s share’ implies that all of the increase represents a misallocation when that’s not the case at all. It is the change from 39% to 59% relative to the change in asset holding across groups that matters.” That comment is nearly indecipherable, but let’s try. The CBO infers that the top 1 percent now holds 59% of assets which, by definition, must mean “relative to … asset holding across groups.” That’s what 59% means. In reality, the 59% figure is derived from the top 1 percent’s observed share of taxable capital income, since the CBO has no estimates of asset holdings (much less nontaxable asset holdings) and chooses not to use the Fed’s. Unlike Thoma, the CBO certainly does not confuse corporate profits with personal income from dividends, capital gains, interest and rent. The CBO uses personal income from dividends, capital gains and rent as a proxy for asset ownership in general. The point of the Reynolds-Henderson article is to show that this procedure does not result in a credible approximation of the top 1 percent’s share of wealth, and to explain why.

Thoma’s latest post says no single measure is perfect, which is why I have presented a wide variety of measures of inequality of disposable income, consumption, wage and wealth and also discussed some of the data problems (such as the data breaks of 1986 and 1993). I also discuss some problems with Gini coefficients in pages 14-20 of my book, which displays real income data by quintile (from Census and CBO) rather than relying on summary measures.

Thoma claims increased inequality of something (income?) since 1988 “has been documented elsewhere in many different ways,” but I have explained why the outside sources he previously mentioned (notably, Ben Bernanke) were mistaken. Thoma even claims “there’s overwhelming evidence,” but never explained what it is or where it is. He thinks the government has underestimated the profit share of national income. But I noted that a low profit share has normally been associated with nasty recessions like 1982, not with prosperous periods like the 1960s. The share going to the top 1% has also fallen in every recession since 1920, according to Piketty and Saez, but that certainly does not demonstrate that recessions must be good for workers because they “reduce inequality.”