Friday, August 16, 2013

The truth about the decline in real median household income


- by New Deal democrat

This is the final installment in a series I began a couple of months ago, after Pulitzer Prize winning journalist David Cay Johnston was taken in by a Doomer argument based on unit labor costs falling over 5% on a real basis in the first quarter, and It was picked up on Prof. DeLong's blog. The number is true, but all it did was take back a little more than half of the +8% rise in unit labor costs from the 4th quarter of last year. In other words, it reflected the shifting of income forward from 2013 to 2012 due to the so-called "fiscal cliff" and its unknown (at the time) tax increases. As of this writing, real personal income as reported monthly is higher than at any time except for last December. Second quarter unit labor costs will be reported on August 29. (Update: They were actually reported this morning, up 1.4% before inflation, or up about 0.9% after inflation).

In any event, Johnston's article went on to opine that the unit labor cost drop was part of a bigger dismaying trend: a decline in real household income that dates back to the turn of the Millennium. He cited, among other work, that of Professors Saez and Piketty of Berkeley, who have documented a 10% decline in the incomes of the "bottom 90%" and further how the top 5%, 1% 0.1% and 0.01% have pulled away from all other households.

And it is true. Real family and household income fell about 10% between 2008 and 2011. Most likely it ihas recovered some since then, but is still far below its pre-recession level. As we will see, it is mainly a result of the decline in the employment to population ratio, and not about a decline in wages themselves.

Provoked by the exchange, I decided to take a detailed look at both mean and median wages and household income. As to wages, the results showed that in real terms both average and median wages actually rose during the great recession, almost entirely due to the decline in gas prices from $4.25 in July 2008 to $1.40 in December 2008. Since that time, gas prices rose back as high as $3.95 before retreating to oscillate around about $3.60 as they have for the last year. This rise worked its way through the general price level, causing both real mean and median wages to fall again. Average (mean) wages are currently about where they were in 2009, having peaked in 2010. Median wages peaked in 2009, and have fallen back down to 2007 levels. Both progressions are shown in the graph below:



Thus, if we measure from 2010 and 2009, respectively, both mean and median wages fell by about 3% and bottomed in 2011 or 2012 before rebounding slightly this year, as measured by average hourly wages, the employment cost index, or the median weekly wage.

So how do we square this with the reported 10% decline in household incomes? The natural thought after hearing those numbers is to think something like, "OMG, wages have really been cut, and far worse than we were led to believe!" That would be wrong, not that certain Doomers intentionally or ignorantly don't understand.

The explanation is more complicated. Income includes more than wages. For example, it includes interest on bank accounts (you remember when that used to happen, right?) and from investments like bond mutual funds, held by many more affluent households. It also includes dividends on stocks, including those held in 401(k)'s.

Not only that, but when we talk about "families" and "households", we are talking about something other than "wage-earners." As longevity has increased and Baby Boomers age, the proportion of households of one or more retired persons has been booming as well. There is a second type of non-wage earning household, and that is one wage-earner has lost his or her job. Both of these kinds of households are included in calculations of real median family or household income, and as it turns out, that makes all the difference.

Now let's turn to the evidence. Real family income and real household income is typically reported from three sources: (1) the Census Bureau; (2) Sentier Research; and (3) Professors Piketty's and Saez' studies of income inequality.

The Census Bureau issues an annual report on average and median household income. Households are defined to include any with a member over 16 years of age, and so includes retirees. The last such report was issued in September of 2012, with data through 2011. In its 2012 report, the Census Bureau found thaat the median household in the US still has not exceeded its 1999 high of $54,932. Afer falling in the early 2000's, median household income rose so far as $54,489 in 2007, before falling to $50,054 in 2011. This is close to a 10% loss.

Sentier Research has taken the data one step further. Using questions from the Household Survey, they publish a monthly update on real household income. Doug Short always includes the report at his site, with a very helpful graph, below, showing that real median household income as calculated by Sentier peaked at $56,550 in 2008, and fell all the way to $50,722 in 2011 before rising slightly since. As of June, Sentier reported that median household icome was $52,098:



Profs. Saez and Piketty report on median family income as opposed to household income. For their purposes, a household may include more than one family, especially if there are extended family members or in-laws in the house. Their methodology is different: they examine income tax returns, and report on the median return for the 0 - 90th percentile (i.e. the 45th percentile) as well as various breakdowns of the top 10% of families. They also update data on an annual basis. They last issued a preliminary report for 2011 (pdf), which will be updated shortly. Their results are very similar to those published by the Census Bureau and by Sentier. They find that the median income of the bottom 90% has fallen by almost 14% from $35,173 in 2007 to $30,437 in 2011. Here's the graph of median family income for the bottom 90% from their most recent report:



I have no reason to quarrel with any of the methods used by the Census Bureau, by Sentier, or by Saez and Piketty, nor do I have any reason to doubt the accuracy of any of the their results. (NOTE: That Piketty and Saez do not limit their data to working households and wages is set forth in their above study in Note 1 to figure A1b as to the definition of income, where it appears that not all non-wage income is excluded.  Similar although not identical definitions of income, and the definition of family unit, are used for Table A0 and Table 1, as well as under the tab "Explanations."  In any event, it appears that interest income is included for all of these tables. It should also be noted that Piketty and Saez do not include transfer payments like Social Security in their definition of income).

But, as I have indicated above, it would be mistaken at least, or misleading or worse, to interpret these reports as telling us anything about wages (this is the mistake that the blog Credit Writedowns made last weekend). Most importantly, all of the data sets include households, or families, where the adults are retired. Whether a family or household is of working age or retirement age makes a huge difference to their incomes. Let me show you two graphs showing just how big a difference that is.

First, here's a graph of all households vs. working age households only using the 2011 Census Bureau data:



Now here's a second graph, this one comparing households in the 35-44 age group with retirement age (65+) households:



Notice that not only are working age households considerably above the median for all households, but retirement age households are far beneath it. Even though over time the financial position of retirement age households has improved (largely because of longevity causing a large increase in elderly workers, most pronounced at the age of 75+!), retirement age households evan as late as 2011 were earning only a little more than 1/2 of the median working age household. In short, choosing to retire almost always requires accepting a very large decline in income, as retirees must live on accumualted savings and investments and interest from them, as well as Social Security, and if they are very fortunate a pension. With 10,000 Boomers hitting age 65 each and every day, the proportion of households consisting of retirees is increasing dramatically - and that is pulling down median housheold income.

But that is only part of the story. The other part is that about 9 million jobs were lost in the recession, and only a fraction of those have been recovered on a population-adjusted basis:



As a result of both people dropping out of the labor force due to giving up looking for work, and the tidal wae of Boomer retirements, the employment to population ratio has plummeted, and has barely recovered from its 2011 low:



The above explanation of why household incomes have declined so much makes sense on an intellectual basis. But can we test it empirically using available data? Not perfectly, but we can come pretty close.

If median wages remained the same, and if dropping out of the labor force for any reason resulted in a household falling into or remaining in the bottom 50% of all households, then the change in median family or household income would simply be a function of the employment to population ratio. If, for example, 10% of all households dropped out of the labor force, and they all were thereafter in the bottom half of the income distribution, then if they were randomly distributed beforehand throughout the income distribution, median income would be fall from the former 50th to former 45th percentile.

We can closely reproduce this result in the data, by starting with the median wage as measured by the Employment Cost Index, deflating it by the CPI, and then multiplying the result by the employment to population ratio. This allows us to factor in both what is happening with real median wages and also the percentage of people who have dropped out of or entered the labor force. A 10% fall in the employment to population ratio ought to put us at about the former 45th percentile as the new median, if roughly half of the fall comes from households previously above the median, and half from households already below the median. As graphing luck would have it, we know from the 2011 Census Survey that there was a $12,000 difference bewtween the 40th and 50th percentile in median households. Just over half of that puts us at a little over a $6000 decline in moving from the 50th to 45th percentile. So a 10% decline in the employment to population ratio should give us about a $6000 decline in real median household income. Here's what we actually get:



While this isn't a perfect fit, beginning to drop off about a year before the Sentier data does, (but in line with the peak of Saez and Piketty's data), it is a very similar decline of nearly 10%, and a slight rebound thereafter. I have scaled the result by 1750 to match the peak $ value on Sentier's data for ease of reference. Our replacement graph peaks at about 56,500, declines to just over 51,000, and then rebounds to 52,000 as of the end of June 2013 - very close to the values in the graph of Sentier's results.

So our empirical reconstruction, while not perfect, largely duplicates the results from the Census Bureau, Sentier, and Saez and Piketty. We should get the 2012 information from the Census Bureau in about a month. Based on the above information, I expect that report to show a sideways move in median real household income in 2012, and possibly a very slight increase.

In conclusion, the next time you hear about a decline in real median family or household income, remember that the majority of that decline is not because there has been a widespread decline in real wages. Rather, there was about a 10% drop in median family and household income betwen the onset of the great recession and 2011. According to Sentier, there has been a slight rebound thereafter. While the decline in real median wages explains about 3% of that 10% decline, the remainder is most likely almost entirely explained by a decline in the percentage of working adult households, both from Boomer retirements and from discouraged workers giving up the job search. Median wages themselves are at about the same level they were at just before the recession. They have risen and fallen within a range since then largely in response to changes in the price of gas, but overall they have stagnated.