My correspondent who had tried defending Krugman’s math (see DeLong Follies #7) turned out to be Wharton Research Data Services Director Michael Boldin. Following is a long e-mail I sent economist Boldin in response to recent correspondence from him plus a brief phone conversation.
Dear Michael Boldin:
Sorry it’s taken me a few days to get back to you after our phone conversation last week about Paul Krugman’s math. I’m still puzzled by the reason you gave for not wanting to continue the conversation (you claimed such “math” cannot be resolved over the phone—hardly true in the era of pencils, paper, and pocket calculators), but was glad to learn that you were the person writing me.
When I last spoke with you (about the Index of Leading Indicators), you impressed me as an extremely honest, even candid person, not at all defensive about admitting to error.
In this case, you made the totally understandable error of taking Paul Krugman at his word. You wrote me that “Krugman derived the 138 million full employment level from a Bush administration report,” adding that if I “have a problem with this number,” I should “complain to them.” But as I’ll soon explain, I can’t “complain to them” because what “Krugman derived” was a figment of his own imagination. The “Bush administration report” said no such thing.
I’ll also highlight two related ironies. First, your defense of Krugman’s math directly contradicts Krugman’s own defense of what he wrote. And second, your defense of this math directly contradicts the math that UC Berkeley Professor Brad DeLong—who is apparently privy to what Krugman had in mind—said Krugman used.
Now, if DeLong or Krugman would only apply this math to the numbers actually cited by Krugman in that column—which you are at least willing to do—they would see the results are nonsense.
Two Years?—or Four?
When Krugman first published that column in May 2004, the April ‘04 increase in payroll employment from the Establishment Survey had been reported as 288,000. No doubt you would have shared the general view at the time that this large an increase was unlikely to be sustained.
But suppose I had asked you to assume for the moment that it would be sustained. And then suppose I had also asked, If payroll gains do continue to average as much as 288,000 a month, how long do you think it will take for the unemployment rate to return to its 2000 low of 4.0%—a) about two years, b) about three years or c) about four years?
I choose a 4.0% unemployment rate because–as you know—“full employment” is generally defined in terms of a desired rate of unemployment. And when it comes to that desired rate, 4.0% is about as low as virtually anyone would go. (In fact, when the unemployment rate first hit 4.0% in early-2000, Krugman pronounced it inflationary and called for it to “rise a bit—from 4 percent to say 4.5 if we’re lucky, to 5 if we aren’t.”—New York Times, 5/7/2000.)
Now, don’t you think that you would have answered “a,” “about two years”? Or that, even if you answered “b,” “about three years,” you would have regarded “c” (“four years”) as too way-out to be taken seriously?
No matter how you did the math you would have noticed at the time that the unemployment rate had already fallen to 5.6% by April ‘04; and that, whatever plausible assumption you cared to make about faster growth in the labor force, 24 months of payroll gains averaging 288,000 a month would probably have been enough to bring the unemployment rate down to about 4.0%. And in 36 months…? Well, forget about it.
Then suppose I told you that Krugman thought it would take 48 months. You probably would have asked, What growth does he assume in the labor force? Based on his having written, “Employment…must rise by about 140,000 a month just to keep up with a growing population,” I would have told you, “about 140,000 a month”—which already assumed faster labor force growth.
Brad DeLong makes it clear that this was indeed the math Krugman had in mind.
In his own response to me, posted on his blog, DeLong writes:
“You see, the Bush administration in its 2004 Economic Report of the President did the same calculation as Krugman, using the same census and establishment survey sources.”
He then quotes the following passage from the report approvingly:
“Because the labor market is constantly expanding, employment must be growing moderately just to keep the unemployment rate steady. For example, if the labor force is growing at the same rate as the population (about 1 percent per year), employment would have to rise 110,000 a month just to keep the unemployment rate stable, and larger job gains would necessary (and are expected) to induce a downward trend in the unemployment rate.”
DeLong then adds that the “only difference” between Krugman and the “Bush Administration” is that Krugman assumed 140,000 a month instead of 110,000 “just to keep the unemployment rate stable.”
Based on these numbers, then, we would say that with monthly payroll gains averaging 288,000, the unemployment rate would hardly remain stable, but keep falling. And if the unemployment rate were 5.6% in April 2004, then it would likely take two years for it to reach 4.0%. To suggest that it might take closer to three years would strain credulity would strain credulity–and four years would sound ridiculous.
We also know, with the benefit of hindsight, that the unemployment rate did continue to fall to about 4.7% over the next two years, even though monthly payroll gains probably averaged less than 200,000. Had monthly gains averaged closer to 300,000 over this 24-month period, a 4.0% unemployment rate would have seemed likely.
Statistical Loopholes
You suggest various “outs” for Krugman so that his math might work better.
You suggest that a larger share of the growing population might participate in the labor force, thereby resulting in a faster-growing labor force. “The decision of the adult population on whether to be part of the labor force or not is driven by many factors,” you write, “including the number of jobs available.”
Quite right. But since DeLong already attributed to Krugman an assumption for labor force growth of 140,000—a number cited by Krugman himself–that loophole is not available.
You also point out that the monthly estimate of unemployment is only an approximation. “It is not too hard to find BLS official statements that explain that any month’s unemployment number could be off by a million or more,” you write, “due to sampling error.”
Although “a million or more” is a bit of an exaggeration, you are basically right. That’s why I argue in my own book (see Chapter 13) that the unemployment rate is best tracked as a 6-month moving average. But if you look at the monthly estimates in 2004, you’ll see that if the unemployed did not number 8.2 million—which is the figure I used—then 8.4 million was about as high as you could go. In other words, the range of monthly estimates is not wide enough to materially affect the result.
Then you suggest that perhaps Krugman wasn’t talking about the official unemployment rate at all, but about one of the broader definitions (what the Bureau of Labor Statistics calls “Alternative Measures of Labor Underutilization”), of which the official unemployment rate is only one.
Notice, in fact, that I discuss these other measures at some length in my own book (see Chapter 5), and fault Krugman for ignoring them. But in this case, even DeLong does not suggest that Krugman had one of these broader measures in mind.
And even if Krugman were writing for the professional, much less for the lay New York Times reader, don’t you think that whenever he does have an unconventional measure in mind, he first ought to say so? As an economic journalist, I could not imagine that readers should ever have to give me such latitude; nor do I think you would ever ask it for yourself. So why hold Krugman to even lower standards?
Krugman’s Misrepresentation
But readers do have a right to expect him to quote his sources accurately—which returns us to the error you made in thinking that he had.
For a surface clue that he was misrepresenting, notice that in May 2004, he referred to what “[Bush’s] own economists consider full employment” based on the 2002 Economic Report of the President. But that was literal nonsense, since the economists who wrote that 2002 Report were no longer working for Bush: Glenn Hubbard and his staff had been replaced by Gregory Mankiw and his staff, as Krugman himself had pointed out in at least one previous column.
Now read pages 52 to 53 of the 2002 Economic Report, and then try to make sense of Krugman’s statement that “The job forecast in the 2002 Economic Report of the President assumed that by 2004 the economy would have fully recovered from the 2001 recession”(my emphasis).
On page 53, you find the Report predicted an average unemployment rate of 5.2% by 2004. Even if a 5.2% unemployment rate is Krugman’s idea of “full recovery” (at the start of the ’01 recession, the unemployment rate ran 4.3%; through calendar year 2001, 4.8%), no such foolish claim was either stated or implied in that Report—again, as you can readily determine by simply reading pages 52 to 53.
Similarly, you wrote me that “Krugman derived the 138 million full employment level from a Bush administration report.” But you will be hard put to find out how Krugman managed to “derive” such a figure, since “full employment” is never directly mentioned on pp. 52-53 or even vaguely alluded to.
All you will find is a table (p. 53) in which garden-variety numbers are forecast through the year 2012. They include “Nonfarm payroll employment”–projected to average “138.3” million in 2004, and the “Unemployment rate,” projected to average 5.2% in that same year.
Again, you understandably assumed that Krugman could not so blatantly misrepresent a pretty straightforward document. (For an even more egregious example, where Krugman literally reversed the meaning of an article that appeared in the New York Times, see pp. 10-14 of my book.)
You thought I implied that you “do not understand the concept of full employment.” Sorry for the misunderstanding. I have no doubt that you do. You know as well as I that there is no such thing as a “level” of “full employment” of “138” (or however many) “million,” since the concept refers to the rate at which labor is being utilized. You apparently assumed—mistakenly, as you now can see–that Krugman’s math was some kind of reductio ad absurdum of this nonsensical “138 million full employment level” he had supposedly “derived” from the Bush economists.
Krugman’s Defense
Finally, you might reread the Chapter 2 of my book in which I go over this math. Notice that I did not bother to point out that Krugman had misrepresented the 2002 Report. Perhaps I should have. But since his reference to what “[Bush’s] own economists consider full employment” is so abrupt, I felt I might as well make the charitable assumption that Krugman was using the term in a coherent way: as some desired rate of unemployment.
Besides, as I explain in the book, when Daniel Okrent originally faulted Krugman for mixing “apples and oranges” in that column, Krugman defended himself by asserting (p. 19) that “All the numbers in that 5/24/04 column come from the same (establishment) survey.” So I even made the charitable assumption that we “ignore the reference to ‘full employment’ and treat that 140,000 as a thruway figure” (p. 25) to see if you could get four years by assuming, with Krugman, that all his numbers came from the Establishment Survey. As you know, you can’t.
To sum up, then: The math doesn’t work if you make Krugman’s assumption that “all the numbers” come from the Establishment Survey. It doesn’t work if you make DeLong’s assumption that Krugman was trying to determine how the increase in payroll employment would affect the unemployment rate. It doesn’t work if you associate full employment with a 4.0% unemployment rate. It doesn’t work if you require of Krugman that—unless he explains otherwise–he use terms and concepts that square with official definitions.
I don’t even think the math would work if what Krugman had “derived” from the Bush economists had been something other than his own fantasy. But that point is moot, isn’t it?
Yours respectfully,
Gene Epstein