Floyd Norris: A Skeptic of the Job Creation Stats

I must have problems with authority. One of my forms of recreation is looking out for statistics, particularly government releases, that don’t seem to comport with reality.

Fortunately, I have plenty of good company in this endeavor, including (but far from limited to) Barry Ritholtz, Michael Shedlock, Dean Baker. A frequent target is the nonfarm payrolls report (see here and here for past examples), particularly its dubious “birth/death model” which attempts to allow for jobs created by new firms (or conversely, lost due to businesses closures).

Oddly, the usual suspects refrained from attacking the validity of Friday’s report. Perhaps they felt the fact that it has been revised significantly and frequently is starting to speak for itself (last month’s 4,000 loss, for example, has somehow morphed into an 89,000 gain).

This month, the criticism comes from an unexpected quarter: the restrained and evenhanded Floyd Norris of the New York Times, in a blog post, “Assuming Jobs“. While he maintains his levelheaded tone, he makes no bones about not liking the data:

The employment report is being hailed as good news, proof that the economy is stronger than it appeared after last month’s bad report.

Well, maybe.

As reported, the last six months have seen an average job gain of 112,000 — the lowest such figure for any six-month period in three and a half years. But that is probably overstated.

The most interesting part of the report to me is that the government now thinks it overstated job growth in the year through March 2007 by 297,000 jobs. It will later restate those months.

The employment numbers come from a survey of employers, but the numbers put out by the government are not what the survey found. Instead, the statisticians add a fudge factor, to account for jobs created by new companies that the government does not know about, less those lost when employers went out of business and did not respond to the survey.

In the 12 months through March, that fudge factor added 1,073,000 jobs to the total reported. The government now thinks that about a quarter of those jobs did not exist.

Chances are that most of the overstatement came in the later months of that period, as the economy began to weaken, and that it is continuing now.

Since March, the fudge factor has added 839,000 jobs to the total. And the total increase it has reported, before seasonal adjustment, is 1,709,000.

What all this means is that a significant part of the employment gains this year came from the fudge factor, or the birth-death model, as the government prefers to call it.

Of those 839,000 jobs added, 150,000 were in the construction industry. Anybody want to bet that the number of jobs created by new construction companies, less those lost from failed builders, is that high? Anybody want to bet it is positive?

Comparing pre-seasonal adjustment figures to figures after adjustment is risky, and the fudge numbers are not available on a seasonally adjusted basis. But it is worth noting that, after seasonal adjustment, the government thinks that over the last six months the economy has added 671,000 jobs — fewer than the fudge factor has added before the adjustment.

The bottom line is this: Job growth this year is mediocre if you accept the figures as reported. But we have good reason to think they are overstated, and will be revised lower at some point.

Today’s figures were better than many expected, but they do not indicate a strong economy, and they should not provide a reason for the Federal Reserve to conclude all is well with the economy.

Update, 3:45 AM: Whoops. Dean Baker is indeed on to this item, and due to the difference in his style and readership, is more pointed than Norris:

However, it is worth noting another item in the report. The September report included preliminary benchmark revisions to the establishment survey based on state unemployment insurance records. These records, which provide a virtual census of payroll employment, show that the establishment survey overestimated job growth by 297,000 in the 12 months from March of 2006 to March of 2007, an average overestimate of approximately 25,000 per month.

The obvious culprit in this overestimate is the imputation for job growth in nearly created firms that could not be included in the survey. It would seem that the Bureau of Labor Statistics (BLS) overestimated job growth in new firms last year.

This fact is relevant to the September jobs data because BLS has actually imputed slightly more jobs into the establishment survey in the last three months than it did over the same period last year. If the imputation led to an average overestimate of job growth of 25,000 last year, it is reasonable to believe that the overestimate may be at least as large this year, since the economy seems weaker by most measures.

Reported job growth has averaged 97,000 over the last three months, so if BLS is overstating job growth by 25,000 a month due to a faulty imputation for jobs in new firms, the error would account for a substantial portion of reported job growth. We’ll know the answer to this one in September of 2008 when next year’s benchmark revisions are first released, but it is worth keeping an eye on this issue.

Print Friendly, PDF & Email