Tuesday, May 31, 2005

Re-critiquing Krugman, with Friends

The last post generated some interesting responses in the comments, as well as on the blogs of Robert Waldmann, Brad DeLong, and PGL at Angry Bear.

Here are some points worth synthesizing:

1) Robert states:

The number cited in the op-ed criticized by Samwick is payroll employment, unshaped, unsliced, and selected because it is very very important.

My critique of Krugman has nothing to do with any reference to payroll employment. Krugman is discussing the labor force participation rate, the unemployment rate, and the employment-population ratio. All of these are drawn from the household survey and not the payroll survey, because they deal with decompositions of the whole population into those in or out of the labor force and of the labor force into those employed or unemployed.

2) Brad states:

I will concede that Paul Krugman should have written "primarily" rather than "only."

That concession is appreciated. We can differ on how important we think the distinction is.

3) However, that concession may not be enough. What generated the several posts of mine last fall was my attempt to figure out the extent to which the decline in the unemployment rate is due to the fact that "some of those without jobs stopped actively looking for work, and therefore dropped out of the unemployment statistics," as Krugman asserted.

In my original post, I considered the three usual augmented measures of the unemployment rate that the BLS tabulates in Table A-12 each month that specifically track those who are not in the labor force because they are not actively looking for work. Reviewing those numbers for the period between 6/2003 and 9/2004 (the period in question in Krugman's op-ed), we get:

  • Unemployment Rate: Fell from 6.3 to 5.4 percent
  • UR, Incl. discouraged workers: Fell from 6.6 to 5.7 percent
  • UR, Incl. discouraged and other marginally attached workers: Fell from 7.2 to 6.4 percent
  • UR, Incl. marginally attached workers and those employed part-time for economic reasons: Fell from 10.3 to 9.4 percent
Subsequent exchanges with readers and with some helpful folks at the BLS revealed that the BLS also collects information monthly on those who are not in the labor force but want a job, even if they have not searched recently enough and/or are not currently available to work. So we could add another bullet point (from my last post in the series in October) to include these people:

  • UR, Incl. any who wants a job: Fell from 9.2 to 8.4 percent
Even measures of the unemployment rate that are specifically designed to pick up people who are out of the labor force because they stopped actively looking for work declined over the period in question. All of them fell by 0.8 - 0.9 percentage point, compared to the decline of 0.9 percentage point in the official unemployment rate.

4) I don't even get "primarily" from this. I would like to have Robert, Brad, PGL, Krugman, or anyone else show me an unemployment rate that fell over this period even primarily because "some of those without jobs stopped actively looking for work, and therefore dropped out of the unemployment statistics." This is what a fact-checker at the New York Times would have required of Krugman (even leaving aside his inappropriate use of the word "only" in the original statement) and what the rest of us, presumably including Okrent, would have hoped for as the standard followed at the Times. In my view, Krugman's assertion does not even rise to his own standard of using the standard data in the standard way.

5) I posited, in the absence of such a measure, that the unemployment rate could be falling because people were voluntarily leaving the labor force and being replaced by those who were formerly unemployed. This story fits all of the facts cited above. In the absence of a better story to fit those facts, I'll stick with it.

Brad makes a reasonable point that over the time period in question (and many months on either side), compensation has fallen relative to productivity. Tracking his calculations as best I can from this table, and using the average of 2003:Q2 and 2003:Q3 to approximate Krugman's starting point of 6/2003 and the average of 2004:Q3 and 2004:Q4 to approximate Krugman's ending point of 9/2004, we see that:
  • Real compensation per hour in the nonfarm business sector increased over this period of time, by about 2%.
  • Output per hour in the nonfarm business sector increased over this period of time, by about 4%.
Brad's point is that compensation trailing productivity does not support the story I have used to explain the facts above, since my story makes labor appear to be more scarce and scarcity would push up compensation at least as fast as productivity allowed.

Bearing in mind that this does nothing to resuscitate Krugman's specific point, I agree that it is surprising that 2% real growth in compensation per hour was enough to accommodate any additional scarcity created by people voluntarily leaving employment. Leaving aside the scaling by productivity for a moment, it suggests a fairly flat demand curve for labor.

The weak growth in real compensation per hour may also suggest that a very large share of the productivity increase that is driving the trend in compensation relative to productivity came in the form of improvements in the production process that substitute directly for additional labor. There may be a growing divergence between the average and marginal productivity of a worker. Admittedly speculative, but this is my initial reaction to Brad's statement:
I cannot come up with a story according to which the Samwick channel is the dominant one that is consistent with the growing gap between wages and productivity.

6) PGL states:

While Andrew Samwick has joined the cottage industry of those who parse every word from Paul Krugman, ...

I don't consider myself to be in this cottage industry. I gave one example because an example was called for. I do not believe that one example would constitute the "disturbing habit" mentioned by Okrent, but I also don't presume that this example is the only one. I consider myself to be much happier when I am not asked to read Paul Krugman's op-eds, let alone have to parse every word. There are better places to go for similar points of view.

Other blogs commenting on this post


PGL said...

Three points: (1) I did over at Angrybear graph the labor force participation rate v. the employment to population rate, but I've admitted folks may have stopped looking for work for a variety of reasons; (2) "only" has other meanings besides 100%; and (3) chill-out a bit as the point of my post is that fact checking can be - and perhaps should be - applied across the spectrum. While I don't wish to associate with the likes of Donald Luskin, your post is likely to be cited by this crowd. It ain't my choice of folks to run with - and I'd caution anyone else to stay far, far away from their insanity.

Andrew said...

Actually, people age 55-64 and 65+ are showing increases in their labor force participation rates (LFPR) and employment-population ratios (EPR).

Using April data (to get an annual interval because the data are not seasonally adjusted), take a look at the following for those 55-64:

1995 57.3 55.1
1996 57.3 55.5
1997 59.0 57.3
1998 59.5 58.1
1999 59.6 58.0
2000 59.2 58.0
2001 60.6 58.9
2002 61.6 59.2
2003 62.5 59.9
2004 62.3 60.0
2005 63.3 61.1

(These are series LNU01300095 and LNU02300095 from www.bls.gov.)

PGL said...

gast - we all need to take a step back. After all, Andrews is letting me comment even if I happen to take an occassional different point of view. And Andrew just posted some interesting information on the 55-64 crowd that Bill Polley and I were musing about a little while ago. Heat rather than light is being generated here - but let's note this is how Don Luskin likes it (as evidenced by his latest dishonest rant over at NRO - which maybe misrepresenting the Fortune interview).

Patrick Sullivan said...

Prof. Samwick would be doing gast a favor by removing his ignorant and crude comment. But, I doubt he will ever stoop to DeLong's tactic of removing comments that refute his arguments effectively. Which is something Okrent clearly got right about Krugman:

'Krugman would likely be more willing to contribute to the Frist for President campaign than to acknowledge the possibility of error.'

Examples of which abound. Such as when Bush's former partners in the Texas Rangers humiliated Krugman in a letter to the editor of the Times, correcting this July 16, 2002 statement:

'"[Bush] who put up 1.8 percent of the Rangers syndicate's original capital, was entitled to about $2.3 million from that sale. But his partners voluntarily gave up some of their share, and Mr. Bush received 12 percent of the proceeds — $14.9 million. So a group of businessmen, presumably with some interest in government decisions, gave a sitting governor a $12 million gift. Shouldn't that have raised a few eyebrows?"

For those who don't know, Bush was the MANAGING GENERAL PARTNER (i.e. he had unlimited liability, not limited liability) which is the reason he had a share greater than his investment. A commonplace arrangement in LPs.

How did Krugman deal with it? Not in a column, but in a post on his website, "Power and the Rangers":

'A few people have asked me about that letter from Bush's former business associates, regarding the nature of his deal with the Texas Rangers syndicate. They assert something I didn't know: that he was granted a 12 percent share of the profits despite having put up only 2 percent of the money back in 1989, when the deal was initialized, rather than in 1998, when the franchise was sold. Assuming this is true - it would be nice to see the contract - does this make everything clean and above-board?

'Actually, if anything it makes things worse. '

And then goes on to commit a number of errors in fact and logic explaining why.

Not to mention that we all remember the classy way he handled his apology to Thomas White for libeling him as a securities fraud felon.

Andrew said...

gast's first comment has been removed. I do not allow profanity (in this case) or personal attacks (in the case of the few other comments that I have deleted) in the comments section.

On the substantive issue that gast raised, I only allow people who have registered with Blogger to comment because that is one of the few ways, albeit imperfect, to deal with comment spam on Blogger.

Subject to the constraints of having no profanity or personal attacks, and subject to the requirement that commenters register, all comments are currently allowed and commenting is encouraged.

On gast's other comment, I quoted Mankiw's statements about Paul Krugman because I think it summarizes my frustration with reading Krugman's columns.

On PGL's comments, I understand that there are others who do seem to parse every word Krugman writes and make the same errors that they accuse him of in making their critiques. I think that your lumping me in with that cottage industry is unfair.

I am also surprised, given how much good writing has appeared at Angry Bear on this issue, that you won't simply acknowledge my point that Krugman, at the very least, overstated his case. What is your definition of "only" that allows his statement to be true? How important can Krugman's posited effect be if the declines in the augmented measures of unemployment all fall by similar amounts to the official rate?

You have rightly taken Kudlow to task for his cheerleading (and I have never used Kudlow or Luskin as a source). Why won't you do the same with Krugman for his excess in the other direction?

Robert said...

Thanks for the link. Reading your reply to my post I wonder what I had in mind. Of course, as you note, Krugman mentioned the unemployment rate and the labor force participation rate.

I think what I meant was that Krugman's point was based on the employment series. Given stagnating employment, unemployment could only decline due to decreased labor force participation. If the household survey were invalid, Bush's point (about unemployment) would be invalid, but Krugman's counter point would still be valid. Still my sentence was silly.

I still think that you are disagreeing with Krugman's analysis not his statistics. The numbers are from two surveys, but it is perfectly reasonable to analyse them together. Everyone does almost all of the time except, of course, for Fresh water guys who ignore the unemployment series entirely.

On the question of substance I too highly recommend Brad's blog.

I have a thought too. One thing about the household vs payroll survey debate is that the weakest number seems to be population which is based on, among other things, a guess about illegal immigration (which probably declined after 9/11). I got this from Angry Bear I think.

This means that the decline in the employment to population ratio is over estimated. The same reason that the household survey gives bad estimates for employment (a ratio multiplied by a guess) would mean it gives good estimates of the employment to population ratio (the payroll survey gives estimates which are a count divided by a guess).

The anomaly in labor force participation is there in the household survey, but I think that the sense that it is huge comes from comparing the two surveys.

Of course you have noticed how reliable I am. Thanks again for the link

Kirk said...

I'm not certain which side it supports, but I'll note an interesting series when you break down the participation rates by age. BLS, cps charts. Series id LNS12300012Q is 16-19 year old employment to population ratio, 1995 to present (quarterly, seasonally adjusted). Throughout the reported 1990s, the numbers are in the 43.0-45.0 range with most in the mid-to-high 44s. In 2000 continuing this the respective quarterly numbers are 45.2, 46.0, 44.6, 45.0. In 2001 a decline begins ending the year at 40.9. This decline plateaus briefly in mid 2002 around 39.7, then plunges to the mid-low 36s. In other words, the ratio drops almost 10 percent from 2000 to 2003. In contrast, the other age groups for whom I've been able to assemble the numbers (LNS12300036Q - ages 20-24; LNS12300089Q - ages 25-34; LNS12300091Q - ages 35-44; LNS12300093Q - ages 45-54) show a decline of approximately 2-3% over the same interval.

I've no explanation. I'll suggest, however, that this may be a relevant item. I'll even go so far as to offer the opinion that the decline we've seen isn't experienced workers but is instead new workers - people only recenly eligible to enter the job market. Supporting this opinion is the point that these entry workers wouldn't class as unemployed. They've not lost a job and so are seeking another, they're seeking their first.

Of course there could be other reasons. I just think that variance is something worth noting.

Kirk said...

Fresh Air said:

I don't think new entries into the work force would vary that much year-over-year.

I say:

Sorry, it's not an opinion. It's the data from the BLS - I cited the appropriate series for that reason.

I also fail to understand how most of your points relate to the original point of this blog's thread - that the unemployment rate has fallen while the employed fraction of the population has remained steady. The fuzzy point appears to be in the unemployed looking for work. I noted a specific instance of a population which would not be considered unemployed and so might not be caught in the 'looking for work' surveys AND it is working at a significantly lower rate. This might explain the peculiarity, or at least contribute to it.

Kirk said...


If it's flawed data, it's been pretty longterm.

Using Q-IV data, here is the (seasonally adjusted) 16-19 year old non-institutionalized population and civilian labor force level for 1995-2004 (numbers in thousands) from the BLS:
Year Population Labor Force
1995 14712 7769
1996 15136 7913
1997 15473 7947
1998 15809 8319
1999 16117 8374
2000 15853 8198
2001 15994 7806
2002 15943 7449
2003 16162 7003
2004 16725 7175

In 1999, the 16-19 age group formed approximately 6% of the civilian labor force level. By 2004 this is just under 5%.

The BLS listed the 16-19 age group unemployment rate (worked and looking for work) as 13.7% in 1999-IV, and 17.1% in 2004-IV. In raw numbers, in 1999-IV 1147K of the total 5716K, or 20% of the unemployed (prior job, looking), were from the 16-19 age group. In 2004 1226K of 8004K, or only 15% of the unemployed came from this age group.

The numbers tell the same story. This age group has had a very large decline in work force contribution over the preceding five years. It's a large enough group to have an impact on the rest of the whole.

I'm willing to acknowledge there may be other causes. But these numbers aren't insignificant. And dismissing them as such because they don't support your guesses of what's going on is frustrating to me. Dismissing? "Oh, sounds like bad numbers." "Can't possibly be a significant proportion."

You've stated a belief that the data set has problems. Cool - justify it. Cite verifiable sources. Or set up a test that can validate or invalidate your beliefs.

Because right now I'm citing facts and normally accepted authorities and you're responding with opinions. And for some reason that doesn't give me confidence that your position has validity.

Kirk said...

FA, no, I'm being a pedantic, egalitarian jerk. You set a standard for Dr. Krugman of clearly supported (un-nuanced) opinions and conclusions, and I insist that you play by the same standards. You challenged my cite and my numbers' relevance, and I backed them up. But when I returned the challenge, you insulted me instead of supporting your position.

Again, your position may have validity. But I have no way of knowing this beyond your say-so. And as you are unwilling to provide supporting cites or tests and your profile has no detail that indicates you may know this professionally or avocationally - your say-so ain't enough.

alphie said...

Are there any independent checks on the numbers the BLS publishes?

Even the biggest Bush supporters have to admit he has trouble with accuracy...

spencer said...

One of the problems you are having is that all of the various unemployment rates you cite are impacted by both the number of unemployeed and the labor force.
Just as with the unemployment rate, given the number of unemployeed a drop or lower growth of the labor force will cause the
unemployment rate to be lower. But exactly the same thing happens for the marginally unemployeed.
A drop in the participation rate will impact the rate for the marginally unemployeed the exact same way it impacts the fully unemployeed. Because of this, citing the marginally unemployment rate does not add anything to your argument because the numbers are skewed -- a poor choice of words -- in the smae way the overall unemployment rate is. If the overall unemployment rate drop was due to a drop in the participation rate, it will have the same impact on the marginally unemployeed rate.

Andrew said...


I'm not sure that I follow your argument, but I'll do my best. Here are some equations to help:

Labor Force = E + U
Population = E + U + N


E = Employed
U = Unemployed
N = Not in labor force

The Unemployment Rate is:

UR = U/(E+U)

Normally, upon hearing that UR fell, we would think that some people shifted from U to E. However, we also know that the labor force participation rate:


has been falling. This means that N has been rising. Suppose we divide up N into two pieces:

N1 = Those who want to work
N2 = Those who don't want to work

Three of the alternative measures of the unemployment are variants of the form:

UR' = (U + N1)/(E + U + N1)

The critique Krugman is making of the official unemployment rate is that it would fall if some people from U shifted into N1. These alternative measures are invariant to such shifts. So if both UR and UR' are falling, then it must be through some channel other than U to N1. I was arguing for (E+U) to N2, which seems like the only channel left. Whether that is a good argument depends on how much we believe the distinctions the household survey draws between those in N1 and N2.

Kirk said...

Cyberike - that works until you look at the data. From 1992 through 2000 the employment among 16-19 year olds increased. Specifically, workforce as a proportion of total population increased AND the unemployment rate decreased. Yet videogames were available throughout that period.

I'll point out that to some extent this is Krugman's conundrum writ large. I'm still not certain this age group is relevant to the conundrum, but the peculiar disparity is there in full measure.