This article is the second in a three part series. Previous essay: Part One
In my last essay, I argued that despite the consensus that the middle class and poor have seen income declines since their pre-recession peak in 2007, a more complete measure of income and a proper accounting for inflation show that by 2011 incomes had probably fully recovered. This conclusion, while very different from the conventional wisdom, does not exactly make for cheerful news. If not for the safety net—including, importantly, noncash benefits like food stamps and the effects of tax policy changes such as the payroll tax cut—incomes would have fallen by an historic magnitude. However, the evidence does suggest that federal safety nets—permanent and temporary—did their job and that the Great Recession inflicted less damage on families than we might have believed.
This essay will further qualify conventional accounts of living standards by looking at trends since 2000. For many observers, notably New York Times columnist Paul Krugman, the 2000s were a “lost decade.” In these accounts, many economic indicators worsened, though a closer look often reveals that aggregate trends look worse than they are due to plateauing female labor force participation, rising immigration, rising school enrollment, and reduced work as the boomers approached and entered retirement.
In the case of income trends, measurement issues make interpreting the period very difficult. The first essay in this series highlighted the importance of correctly measuring changes in the cost of living and of focusing on disposable income in its broadest sense. The current essay will highlight data quality considerations.
The first issue that immediately arises is to define what is meant by “the 2000s,” or “the aughts”. Economists are careful to think not in terms of arbitrary ten-year windows when assessing income trends but in terms of business cycles. To determine what to make of long-term trends, one has to compare similar points in the business cycle. Conventionally, business cycle peaks are used to assess change over time, in part because such peaks have coincided closely with decade demarcations over the years, coming in 1969, 1979, 1989, and 2000. The last peak came in 2007, however, before the Great Recession. Comparing 2000 incomes to those in 2010 would lead to unduly negative conclusions about the trajectory of living standards, because it would compare a boom year to a bust year.
With that in mind, what happened to incomes between 2000 and 2007? According to the official Census Bureau measure of money income, which adds cash government transfers to privately-earned income and the earnings and retirement benefits of government workers, the median household’s income was 2 percent higher 2007 than in 2000. (All of the figures I cite, unless otherwise indicated, use the recently revised “PCE deflator” from the Bureau of Economic Analysis.)
When I added noncash government benefits to that measure, such as food stamps, Medicare, and Medicaid, and the value of employer-provided health insurance, the increase was not much larger. Depending on whether one looks at the median household or the median person, on how government health benefits are valued, and on whether or not incomes are adjusted for the fact that households of different sizes have different needs, median pre-tax income rose 3 to 5 percent over the period.
You may recall there were some tax cuts during the Bush years, and lower taxes raise disposable income no less than an equivalent increase in government benefits. However, I find that median post-tax and -transfer income rose by just 3 to 6 percent. This is only marginally more than the 1 to 4 percent increase I find in pre-tax and -transfer median income over the period. In short, contrary to the post-2007 period, taxes and transfers do not appear to be an important part of the story in explaining 2000 to 2007 income growth, which was tepid by any of the estimates from the Census Bureau data.
However, there is reason to believe that the Census Bureau figures understate income growth over time. Under-reporting of income in studies like the Current Population Survey (from which my estimates are drawn) is substantial when compared against federal administrative records. Economists Bruce Meyer, Wallace Mok, and James Sullivan find that less than 60 percent of food stamp benefits were reported in the CPS in 2004 and just half of TANF benefits. One Bureau of Economic Analysis researcher found that in 2001, business owners reported only half of their business income in the CPS. Dividends were reported at a 59 percent rate, retirement and disability income (excluding workers’ compensation and Social Security) at a 70 percent rate, and interest income at a 73 percent rate.
Buttressing these findings, survey reports of consumption exceed income among poorer households. That could reflect spending that comes from taking on debt or drawing down assets, but Meyer and Sullivan found that income under-reporting is substantial even for households with little of either.
If income under-reporting has increased over time, or if poorly-reported types of income have become a bigger share of what households bring in, survey-based income trends will understate improvement in living standards over time. The evidence on trends in under-reporting is, unfortunately, thin, but reporting of means-tested government transfers has clearly worsened. Reporting of other types of income, such as self-employment earnings, is sensitive to changes in marginal tax rates.
There are two alternatives to survey-based income trends—using administrative data on income or using survey data on consumption.The Congressional Budget Office has created the most complete measure of income based on administrative data. It assigns tax filers in IRS data to matches created synthetically in the CPS. It then retains the income from the IRS data, gives the tax filers the cash and noncash benefits from the CPS, subtracts out taxes (some of them estimated), and groups the tax filers back into the original CPS households.
The CBO estimates suggest stronger income growth during the aughts than CPS data does. Market income rose by 6 to 7 percent, pre-tax income by 10 to 11 percent, and post-tax income by 14 percent. These estimates suggest annual income growth in the aughts that was stronger than in the 1980s and even the 1990s. In contrast to the CPS estimates, the CBO data suggest that perhaps half of the increase in disposable income during the aughts was due to rising government benefits and falling taxes, with the two roughly equally important. Taxes and transfers were just as important to rising incomes in the 1980s, but much less so for 1990s income growth.
To the extent that matching CPS records to tax-return records replaces under-reported income amounts with better-reported ones, it may improve on the survey-based estimates. However, because the matching technique is based on complicated modeling that attempts to appropriately pair different tax filers in two datasets, the CBO estimates arguably suffer from the necessary assumptions involved.
Consumption data from the Consumer Expenditure Survey are more straightforward, though far from perfect. Like income, expenditures are under-reported in surveys, and increasingly so. Unlike income under-reporting, however, consumption under-reporting appears to be concentrated at the top of the income ladder, which suggests that consumption reports may be more accurate than reported income as an indicator of poor and middle class living standards. In fact, Meyer and Sullivan have shown that most of the largest spending categories are fairly well-reported.
Using Meyer and Sullivan’s results, but replacing their cost-of-living index with the PCE, I estimate that median consumption rose by roughly 18 percent from 2000 to 2007. Like the CBO data, Meyer and Sullivan’s figures suggest that the 2000 to 2007 expansion featured stronger income growth than the previous two (especially the 1980s expansion).
Since median income was likely flat from 2007 to 2012, when paired with this analysis a reasonable guess is that disposable income today is at least 10 percent higher than it was in 2000. If this seems implausibly high, note that the fact that under-reporting of consumption is increasing means that median consumption is understated by more in 2007 than in 2000. For that matter, many experts argue that conventional inflation indices bias improvement in living standards downward.
What about the poor? In the CPS, pre-tax and -transfer income at the twentieth percentile—the household poorer than 80 percent of all households—fell by 1 to 8 percent from 2000 to 2007, and the official definition of money income (which includes cash transfers) fell by 1 to 3 percent. Accounting for non-cash benefits raises the income of the poor by 5 percent at most. Factoring in changes in taxes indicates either a small drop in income or a small gain.
On the other hand, CBO estimates suggest an increase in pre-tax and -transfer income of 16 to 17 percent for the bottom fifth of households, an increase in pre-tax-post-transfer income of 8 to 14 percent, and an increase in post-tax and -transfer income of 10 to 16 percent. Consumption at the 10th percentile—the household poorer than 90 percent of households—rose by roughly 16 percent according to the Meyer and Sullivan data. If the CBO data are to be believed, these gains had little to do with transfers or taxes, and on an annual basis, they were as strong as in the 1990s (and stronger than in the 1980s).
The CBO disposable income estimates and the consumption estimates from Meyer and Sullivan give quite a different conclusion than the one drawn by analysts who simply use the data closest at hand uncritically. For instance, Larry Mishel, president of the Economic Policy Institute recently claimed in an essay titled, “Already More Than a Lost Decade,” that, “In the full business cycle from 2000 to 2007, poverty actually increased and, for the first business cycle on record, incomes for those at the middle did not rise.” In another piece, he argued that median income for nonelderly households fell by 11.6 percent between 2000 and 2012. CBO data, on the other hand, indicate that the disposable household income of the median nonelderly family rose by about 15 percent.
I do not want to suggest that the impression left by the CBO and Meyer/Sullivan figures is definitely closer to reality than the conventional figures. But they ought to leave economic pessimists less certain that the 2000s were a lost decade. Interpreting trends in living standards requires more tolerance of ambiguity the further back we look. The next essay in the series—looking back to 1979—will make that point dramatically clear.