The White House, Treasury Secretary Snow, and Labor Secretary Elaine Chao have been pointing out that the American economy has created 4.4 million new jobs created since May of 2003.

Inquiring minds want to know: How legitimate is that number? Note that this is a very different (but related) question to the one Floyd Norris looked at in the Sunday NYT: How does this post-recession Jobs recovery compare to prior ones?

Both questions should be familiar to readers of this blog.

Let’s start with my question: How legitimate is that 4.4M number?

The answer is, it depends upon how you look at it: Its either 1) Very Legitimate; 2) Legit, but Misleading; 3)About a Third Fabricated Projected;  4) Not nearly as legitimate as it appears. Here’s how they breakdown:

1) The Very Legitimate argument is simply math: In May 2003, there were 129,827,000 people employed, as of November 2005, there were 134,289,000. That reflects 4,462,000 new jobs. So the 4.4 million number is very accurate — at least mathematically.

2) The Legit, but Misleading take is based upon the time period selected: Note that May 2003 happens to be the low point for total employment. The only way to get at that 4.4M number is by advantageously selecting time periods on a trough to peak basis.

If this were a mutual fund advertisement, such a selective time line would be disallowed by the SEC as misleading:>

 > >> > > > >>>  Total Employed (in 1,000s)

Source: BLS

What if we took a less randomly selected time period? If measured from the start of the President’s first term, the job creation numbers are much worse: about 1,835,000.

As we pointed out earlier, we can make the numbers better or worse depending upon when you mark the beginning of your time period.   

3) The Fabricated Projected Answer is based upon a simple adjustment: The 4.4 million new jobs number includes a newer adjustment called Birth/Death adjustment. Its been gradually phased in since 2001, becoming fully implemented by 2003. This is an estimate (not a measurement) of the jobs created by newly born firms and lost by newly dead firms. Out of the 4,462,000 new jobs measured from the March 2003 low, ~1,639,000 were this statistical fiction "adjustment." That works out to be 36.7%. That’s a hefty number by any measure.   [NOTE: But see Tim's comments below]

4) The prior three issues have been quantitative. Our last measure — "Not nearly as legitimate as it appears — is more qualitative: 

As I’ve noted previously, the jobs recovery compares rather poorly with prior post WWII recessions and their aftermaths. We’ve seen that the jobs created are unusually dependent upon the real estate complex, they pay less and have weaker benefits than the lost jobs they replaced. And, there have been an unusually large amount of government jobs created. So the overall quality of this recovery is pretty mediocre.

These four measures show that this has been a fairly mediocre jobs recovery. The reason for this is quite simple in my opinion: Its because this is a post-bubble economy — the 2,000 crash was a 100 year flood — and looking at this recovery as if its just another post war cycle misses the bigger issues.


Which brings us to the next query, Norris’ question:  How does this post-recession Jobs recovery compare to prior ones?

The bottom line is reflected in the table accompanying Norris’ Column:

     Historical Comparisons, Job Creation, Presidential Terms
Table courtesy of NYT


Here’s an excerpt:

"In terms of economic growth, this recovery ranks eighth for the four-year
period after the official end of a recession, as measured by the National Bureau
of Economic Research. It is just better than the period after the previous
recession, under the first President Bush, ended in 1991, and considerably
better than the period after the first recession in the Eisenhower
administration. Then, a new recession began three years and three months after
the first one ended.

The recovery that began in 1991 became the longest period of uninterrupted
economic growth in American history, an indication that there is nothing wrong
with a slow start. But the current recovery so far is far from impressive.

Consider jobs, the focus of the Treasury chart. A unique aspect is that the
job count continued to fall for 18 months after the 2001 recession ended.
number of jobs in November was up 3.4 percent from the job low 30 months

That measurement, which is the way the Bush administration chose to look at
the data, ranks eighth among the 10 postwar recessions, a fraction ahead of the
recovery after the 1990-91 recession, and better than the period after the
recession that ended in July 1980, when another recession followed a year later.

Were job growth instead to be measured from the end of the recession, this
recovery is the slowest ever, with the job count up 2.6 percent in four years.
The previous low was a 4 percent gain in the four years after the 1953-54

This is yet another example of contextualizing data. Too often, the headline (4.4 million new jobs!) is misleading when compared to the actual data beneath . . .


The Jobs Recovery Looks Good, Until a Longer Look Is Taken
NYTimes, December 10, 2005

Category: Data Analysis, Economy

Please use the comments to demonstrate your own ignorance, unfamiliarity with empirical data and lack of respect for scientific knowledge. Be sure to create straw men and argue against things I have neither said nor implied. If you could repeat previously discredited memes or steer the conversation into irrelevant, off topic discussions, it would be appreciated. Lastly, kindly forgo all civility in your discourse . . . you are, after all, anonymous.

10 Responses to “How Strong is this Jobs Recovery?”

  1. muckdog says:

    The recession was pretty mild, though. And at 5% unemployment, some might argue that we’re at “full employment.” So is this economic upswing weak, or are we just in an upturn from a not-so-dramatic 2001 recession?

  2. JWC says:

    I watched Ms. Chao on CNBC last Friday and couldn’t stand the spin. Even with their soft ball questions she looked like nothing more than a shill for the administration. I continue to be amazed at the pushing of the 5% unemployment number while ignoring the percent of the population that have dropped out of the labor force.

  3. Kash says:

    Barry: You make an excellent, and under-understood, point about choosing starting points when looking at data.

    I put up some more long-term context for the current recovery on Angry Bear today. Context is key…


  4. Tim says:

    The birth-death estimates for the previous year are replaced with actual state employment insurance data every January, and the entire previous year’s data gets revised. Look for that next month – sometimes it’s a big number, sometimes not. So, it’s not really fair to say that it’s a fabrication – just estimates until the yearly revision.
    I continue to believe that all these jobs are real, there are just too many in categories like Food Service and Construction and too few in Profesional Services.

  5. Hey Tim,

    I spent an hour on the phone with BLS today — no one there could give me that info — I specifically asked about benchmarking, data correctiona nd substitution — they told me its a pure projection –

    How did you derive this info ?

  6. calmo says:

    Too funny. It WAS darts all along. There never was a fancy model. I believe that info ( not the darts) can be obtained from the web site.
    And what is all that wrong about a fabrication as long as it works which the annual benchmark revisions confirm, no? [Is that benchmark revision just another fabrication?]
    Even if we skip the low paying service , continuing temporary, marginal , self-deprecating nature of these jobs, even if the UE rate drops to 2%, can we just ignore the cost of this wonderful job recovery, the ~$3B/day?
    Are these the normal costs that we expect from job creation –the ones that will pay big dividends down the road? All that R&D investment in those new jobs (that only appear to be rewriting mortgage agreements) are in fact gateways to a new standard of living?
    So there is that trick of picking your starting date for a suitable conclusion and then there is that trick of ignoring the general flood conditions by noting that the rain has eased somewhat.

  7. Tim says:

    Here are a couple links and a couple excerpts. In re-reading my comment, I’m not sure it is 100% accurate, but this is what it was based on:

    Accounting for Business Births and Deaths in CES: Bias vs. Net Birth/Death Modeling

    The Longitudinal Data Base (LDB) is the universe from which BLS draws the CES sample and is derived from state unemployment insurance (UI) records submitted by employers who are subject to state unemployment insurance laws. The LDB contains data on approximately 7.5 million records for the year 2001, representing nearly all nonfarm business establishments of the U.S. economy. These reports provide information on the number of people employed and the wages paid to employees each quarter3. This database is also the main source for the annual CES benchmarks.

    U.S.: Birth/Death Confusion Redux

    Contrary to current conspiracy theories, the monthly birth/death adjustments are not some sort of ad hoc or subjective adjustment to the numbers based on the whims of the BLS staff. The monthly adjustments are based on a model that incorporates incoming actual data on business births and business deaths as it becomes available. The procedure roughly consists of the following: First, create an employment estimate series after each benchmark month by assuming that businesses that die are replaced by businesses growing at the same rate as the surviving businesses. In other words, if there were five companies in the sample in May, but one disappeared in June, replace it with an imputed company that had payrolls equal to the dead business in May times the growth of the four surviving businesses from May to June. This imputed level of payrolls extrapolated forward provides a first estimate of payroll growth from net birth/death. Second, compare this imputed series with actual data on the universe of employment from state unemployment records to determine the errors each month in the basic imputation method. Third, create a times series of these monthly errors over the prior five years. Fourth, use this time series of past errors to forecast birth/death adjustments in the current period. Fifth, update the error series and re-estimate each quarter when more detailed unemployment insurance records become available from the Quarterly Census of Employment and Wages (QCEW).

  8. D.Wallener says:

    According to the table posted in the article, the unemployment rate this far after a recession is one of the very lowest on record. To spin this simple fact to the negative by focusing on “growth” is not intellectually honest unless you also emphasize the extremely low level of unemployment the last recession created.

    As another commentator pointed out, the birth/death forecasts are a temporary issue, as the numbers are replaced with actual data over time. This is even shown in the tables the article linked to.

  9. howard says:

    here’s the number i think we should remember: bush promised that the 2003 tax cut would create 5.5M jobs between july 1, 2003 and december 31, 2004.

    so here we are, 1 year later, and we’re still 1M jobs short.

    i do realize that “bush” and “promise” are, in some sense, oxymoronic.

  10. tee says:

    Wow Best site I liked look my – sexteen