Comments Guidelines

All comments are pre-moderated. No spam, slurs, personal attacks, or foul language will be allowed.

Wednesday, March 28, 2012

The Best Data on Middle Class Decline (Updated)

The flurry of posts earlier this month on middle class decline (me, Lane Kenworthy, Matthew Yglesias, Kevin Drum) made me think some more about what the best way is to show what's happened since the peak of real wages in the early 1970s. While in my opinion there is no perfect measure, there are a lot of choices to be made, and I argue below why real wages for production and non-supervisory workers, with an adjustment for non-wage compensation, is the best single measure.

Choice 1: Household/family vs. individual

While we all live in households or families, over the past 40 years, there has been a decline in persons per household (see Kenworthy) and an increase in incomes per household as women's labor force participation has increased. The decline in persons per household means that a household needs less income than in the past to have a fixed per capita income. The increase in incomes per household has meant that households have had higher real income even as real individual income has fallen, as pointed out by commenter peggy_Boston in the comments thread of Drum's article. To my mind, this is partly causal; that is, because real wages have fallen, families have had to have more incomes in order to maintain their living standards. Indeed, falling real wages have forced families to run up high levels of debt, with non-mortgage debt reaching 1/3 of family income by 2005. Therefore, I think individual data is the right choice here.

Choice 2: Median income vs. production and non-supervisory workers' income

The median (middle value, with an equal number of observations above and below it)  has big advantages over the arithmetic mean in trying to show the typical situation in a distribution of values. It is especially useful for income distributions, where the presence of very high incomes means that the mean is much higher than the median. In fact, the literature on "decoupling" (see Kenworthy above) demonstrates just how much this is the case. But I think that "production and non-supervisory workers" captures our intuition about who is in the middle class even better than the median does. This series, in Table B-47 of the Economic Report of the President, is an average of the earnings of employed persons in private (non-government), non-agricultural jobs. It includes about 80% of the private workforce and 64% of the total non-agricultural workforce. Despite being an average, its exclusion of supervisory workers means that virtually all of the extremely high values that distort the mean of the entire workforce are eliminated. It is, essentially, the mean income of the bottom 80% of private workers. The biggest drawback to this dataset is that it does not include non-supervisory government workers, but I think that is outweighed by its broader coverage of the middle class than the pure median income (or middle quintile, as in Kenworthy's post).

For the counterargument, that changes in composition of production & non-supervisory workers can cause distortions that the median wage is not subject to, see Dean Baker (p. 9).

Choice 3: Weekly vs. hourly

Baker mentions hourly earnings rates in some cases. As I discussed in the comments section of my March 11 post, the decline in hours worked per week (from 36.9 hours in 1972 to 33.6 hours in 2011) suggests to me that we need weekly, not hourly, wages.

Choice 4: Which inflation data to believe?

Shortly after President Clinton's first election, I predicted to my students that, because his message of middle-class stagnation ("It's the economy, stupid") was dependent on how inflation was measured, that conservatives would soon attack the official Bureau of Labor Statistics inflation data. The issue is, if inflation is overstated, then the decline in real wages reported by the BLS could be overstated or even non-existent. Conversely, if BLS data understates inflation, then real wages have fallen even faster than shown in Table B-47.

Unfortunately, I did not publish this prediction, so you'll have to take my word for it that I predicted the attack on inflation data that culminated in the Boskin Commission in 1995. I always took this to be a political attack rather than a scientific one. My attitude has always been that trade theory (i.e., the Stolper-Samuelson Theorem; see Ronald Rogowski's great book Commerce and Coalitions for an explanation of this topic, which I intend to discuss in a later post) predicts that real wages in a relatively labor-scarce country like the United States will fall as trade expands, and the data shows that real wages indeed fell: so what reason do we have to question the data? In the end, though, the Commission concluded that inflation was being overstated by about 1.1 percentage points a year, and the BLS was mandated to adjust its methodology.

Barry Ritholtz takes an even more jaundiced view of the Boskin Commission than I do. Paul Krugman, on the other hand, is not convinced that inflation is now significantly underreported, citing the work of the Billion Prices Project. For the moment, I do not see reason enough to toss out the BLS data, despite the possibility that the Boskin Commission may have introduced distortions into it.

Choice 5: Wages vs. compensation

Martin Feldstein and other economists argue that it is not sufficient to look at wages alone, because the non-wage share of compensation has been growing over the past few decades. As I posted before, total employee compensation includes everyone from the CEO to the janitor, so it overlooks the fact that the top 1% have made almost all the gains from decades of economic growth. Nevertheless, it is clearly true that non-wage compensation has grown faster than wages, as we will see below. In fact, Yglesias suggests that the 2000s actually saw real compensation growth at the median, but it was all in the form of health insurance benefits. Of course, there is some debate over how much value actually comes from extra employer payments for health insurance, as Baker's paper (p. 10) details

A different way to factor in compensation that I had seen before on the Economic Policy Institute's website was explained to me in an email by Jared Bernstein and is documented in the footnote of his blog post here. It takes the ratio of total compensation to total wages, both of which are in National Income and Product Accounts Table 1.12 (you can set it to a wider range of years, as I did). Whereas he applies it to median wages, I apply it to Table B-47 and get the following results:

Year          Weekly Real Earnings     Comp/Wages     Weekly Compensation
                  (1982-84 dollars)                                     (1982-84 dollars)

1972          $341.83                         1.14                   $388.01
1975          $314.75                         1.16                   $366.63
1980          $290.86                         1.20                   $348.93
1985          $285.34                         1.22                   $347.10
1990          $271.12                         1.21                   $328.99
1995          $267.07                         1.22                   $326.23
2000          $284.79                         1.20                   $341.49
2005          $284.99                         1.24                   $352.87
2010          $297.67                         1.24                   $370.28
2011          $294.78                         1.24                   $365.77

Note: Last two columns rounded from spreadsheet calculations

Sources: Economic Report of the President 2012, Table B-47, National Income and Product Accounts, Table 1.12, and author's calculations

 By this measure, compensation in 2011 for most workers was still almost 6% below its 1972 peak. The advantage of this adjustment over Feldstein's procedure is it strips out the wage inequality of the compensation data, although there is still some overstatement based on inequalities in non-wage compensation. Still, I think this gives us our most accurate picture of what's happening to the bottom 80% of workers.

That is not to say that this is a perfect measure even with those caveats. It matters what is happening at the top, too. If high wage earners were seeing their income fall faster than middle-class workers, then inequality would be falling and we would probably object less to what would then look like the much-vaunted "shared sacrifice." But of course, as Kenworthy notes, the share of the top 1% more than doubled from 1979 to 2007, from 8% to 17%. With inequality rising as it is, we now seem to be in danger of a consequent sharp shift of political power to the 1%, as MIT economist Daron Acemoglu told Think Progress' Pat Garofolo.

I look forward to your comments, especially if I've gotten something wrong.

UPDATE: By way of comparison, here is Lane Kenworthy's chart.


In it, you can see that by either median family income or 3rd quintile household income, incomes started rising shortly after 1980, whereas in my table compensation-adjusted real wages continued to fall until 1995. You can also see the divergence in median family income and Q3 household income between 2000 and 2007, as noted by Yglesias. Whereas the increase is made up entirely of nonwage compensation in the Q3 household income series, in my table at the individual level we have an increase made up partly of wages and partly of nonwage compensation.


  1. A few questions, some of which are tangenitally related to your overarching point (which I don't necessarily disagree with, just think more information would be good):

    On Choice 1, could you elaborate on why families have chosen to borrow to maintain living standards? I don't think that's an obvious point. Could they instead have borrowed to keep up with the rich as income inequality rose (in which case no matter how great real income is, as long as the top pulls away from the bottom, families will keep borrowing)?

    Choice 4: How should we account for the proliferation of high-quality, low-cost products that surely boost utility without imposing very much larger living costs (Facebook, YouTube, blogging, Skype, working from home via secure online connections, Google search, huge improvements in smartphones and other gadgets without commensurate increases in costs, etc.?)

    In other words, if I claim that the quality of life today is far better for the median worker than it was in 1990, how would you square that with the claim that because inflation-adjusted wages haven't risen, quality of life by definition cannot be better?

    On Choice 5: how do you most accurately measure intangible compensation? Compensation in the form of moving out of the factory floor and into air conditioned offices, safer working conditions as the service sector has grown, increasing emphasis on creativity and human interaction instead of uniform physical labor, etc.? In other words, if the disutility of labor is fallnig, how do you capture that?

    Thanks in advance.

    1. Thanks very much for your comments. I'll try to answer them as best as I can.

      On choice 1, I think that if income falls, you need a second income to maintain the same consumption. But your question resonates with something Juliet Schor argues, that there is a psychological element to consumption as well. As you say, keeping up with the rich Joneses is more difficult if inequality is increasing.

      On choice 4, I actually wrote about that a long time ago ( You are right that the gadgets are cheap, but other costs, particular health care and college education, have skyrocketed in cost much faster than general inflation. I present the data there. You should also check out the Matt Yglesias post I link to in that one.

      On your last point, I'm not sure I know a good way to measure such intangibles. If you know of any research on it, I'd be interested to see it.

      Thanks again!