Lies, damned lies, and the Bureau of Labour Statistics

[J]

For some time now I've suspected the official US unemployment numbers were being manipulated, now I have some evidence that something is very fishy.

First, we have the official BLS unemployment report for July, which can be foundhere for another couple weeks.

Employment Situation Summary


Transmission of material in this release is embargoed USDL-09-0908
until 8:30 a.m. (EDT) Friday, August 7, 2009

Technical information:
Household data: (202) 691-6378 CPSinfo@bls.gov http://www.bls.gov/cps
Establishment data: (202) 691-6555 CESinfo@bls.gov http://www.bls.gov/ces

Media contact: (202) 691-5902 PressOffice@bls.gov


THE EMPLOYMENT SITUATION -- JULY 2009


Nonfarm payroll employment continued to decline in July (-247,000),
and the unemployment rate was little changed at 9.4 percent, the U.S.
Bureau of Labor Statistics reported today. The average monthly job
loss for May through July (-331,000) was about half the average
decline for November through April (-645,000). In July, job losses
continued in many of the major industry sectors.

247,000 jobs lost in July, remember that number.

Next we have the table of mass layoff events, also from the BLS. It can be found here

Code: Select all

Table 2. Mass layoff events and initial claimants for unemployment insurance, August 2005 to July 2009, not seasonally adjusted
Table 2.  Mass layoff events and initial claimants for unemployment insurance,  August 2005 to 
July 2009, not seasonally adjusted 

                                       Total            Private nonfarm        Manufacturing  
             Date                                                                             
                                           Initial               Initial               Initial
                                Events    claimants   Events    claimants    Events   claimants

  2009                                                                             

January ....................     3,806     388,813     3,633     375,293     1,461     172,757
February ...................     2,262     218,438     2,173     210,755       945     103,588
March ......................     2,191     228,387     2,107     221,397       940     114,747
April ......................     2,547     256,930     2,385     243,321       887     100,872
May ........................     2,738     289,628     2,572     274,047     1,005     123,683
June .......................     2,519     256,357     2,051     216,063       674      85,726
July .......................     3,054     336,654     2,659     296,589     1,133     154,208

Note the numbers for July. The number of jobs lost to mass layoffs alone is greater than the total number of jobs lost as claimed by the official BLS unemployment report. There's also nothing to suggest that hiring has somehow soaked up the difference, the only sector with a gain in employment was healthcare with 20,000 jobs added.

So yeah, change you can believe in! The BLS moves from fudging the unemployment numbers with the birth/death model adjustments and goes straight to pulling numbers out of thin air and claiming whatever the heck it wants. I wonder when 2+2 will equal 3?

[Lagmonster]

I can't claim to understand how these figures are generated on a casual review, but is it possible that the first bit details 'non-farm payroll employment' loss (ie actual job loss), whereas the second list, which seems to reference unemployment insurance claimants, simply tells us how many people stated in July that they qualified for unemployment (ie everyone making a first claim whether they lost their job in July or June or coming from some other situation)? If so, the discrepancy would actually be in showing measurements of two separate phenomena, not two differing measurements of the same phenomena. Correct me with better details if I'm misunderstanding the results.

[Surlethe]

Even if they are measuring the same thing, why should we expect perfect consistency? The two measures are on the same order of magnitude and only differ from each other by about 20%. Lagmonster's point aside, especially in social sciences the expectation of 100% accuracy and consistency between all measurements of X - and the claim that the stated measurements of X are fraudulent otherwise - is a black-and-white fallacy.

[Broomstick]

I would like to point out that, in addition to the healthcare industry, government is also hiring. Unfortunately, many of those jobs are temporary, but between the states and the Feds there might be enough new hires to affect the unemployment figures, at least in some areas.

Also, some folks who were unemployed may have started businesses or started working as independent contractors (officially, I am the latter). This, too, will have some effect.

[J]

Lagmonster The mass layoffs are based on the initial unemployment claims filed when a person loses his or her job. They go through the initial claims data and identify companies which have more than 50 initial claims filed against them in a certain time period, if such a dataset is found then those claims are entered into the mass layoffs basket for the month.

Even if they are measuring the same thing, why should we expect perfect consistency? The two measures are on the same order of magnitude and only differ from each other by about 20%.

I don't expect perfect consistency, but when they're simply counting jobless claims in a computer system you'd think they can get the numbers a lot closer than 20%. The only way the numbers could work is if a large portion of those who lost their jobs found new employment within the month, this can almost certainly be ruled out since the average length of unemployment is now 25 weeks, with the median at nearly 16 weeks (table A9). Fat chance that a sufficient number of mass layoff victims plus regular layoffs sufferers found jobs soon enough to push mass layoff figure down to the official 247k.

I would like to point out that, in addition to the healthcare industry, government is also hiring. Unfortunately, many of those jobs are temporary, but between the states and the Feds there might be enough new hires to affect the unemployment figures, at least in some areas.

From the BLS unemployment chart the government added ~7000 new jobs in July. Doesn't look promising, a less populous with a bunch of hires would notice it I suppose.

Also, some folks who were unemployed may have started businesses or started working as independent contractors (officially, I am the latter). This, too, will have some effect.

This is accounted for in the birth/death model adjustments (much as I distrust them) which for the month is +32,000.

---

I suppose it's possible for the numbers to add up if

• the margin for error is fairly large
• at least 70.000 mass layoff victims found new jobs within the month
• there were no additional job losses, if there were then that number has to be made up in the above

Can it happen? Maybe, but I'd rank the chances around say, winning the Powerball lottery, give or take an order of magnitude. When the BLS has a history of questionable numbers, I feel it's pretty reasonable to believe it's not just errors and incompetence. There's a consistent pattern to the errors, IMO, either their models & methods are hopelessly broken or they've taken lessons from Treasury and started messing with the numbers.

[Count Chocula]

The BLS numbers are as accurate as can be established for 300+ million people, based on its underlying assumptions regarding the various categories of unemployment. The real number to watch is U-6:

Table A-12. Alternative measures of labor underutilization
HOUSEHOLD DATA HOUSEHOLD DATA

Table A-12. Alternative measures of labor underutilization

(Percent)



Not seasonally adjusted Seasonally adjusted

Measure

July June July July Mar. Apr. May June July
2008 2009 2009 2008 2009 2009 2009 2009 2009

(Snip U-1 through U-5)

U-6 Total unemployed, plus all marginally attached
workers, plus total employed part time for
economic reasons, as a percent of the civilian
labor force plus all marginally attached workers.. 10.8 16.8 16.8 10.4 15.6 15.8 16.4 16.5 16.3

NOTE: Marginally attached workers are persons who currently are neither working nor looking for work but indicate that they want and
are available for a job and have looked for work sometime in the recent past. Discouraged workers, a subset of the marginally attached,
have given a job-market related reason for not looking currently for a job. Persons employed part time for economic reasons are those
who want and are available for full-time work but have had to settle for a part-time schedule. For more information, see "BLS
introduces new range of alternative unemployment measures," in the October 1995 issue of the Monthly Labor Review. Updated population
controls are introduced annually with the release of January data.

U-1 through U-3, sometimes U-4, are the only stats reported in the newspapers and television. You have to find other sources for U-6, which is a more accurate measure. U-6 in July '08 was 10.8%; now, it's 16.3%. In short, the government and media know the real numbers, but choose not to talk about them.

[Surlethe]

The only way the numbers could work is if a large portion of those who lost their jobs found new employment within the month, this can almost certainly be ruled out since the average length of unemployment is now 25 weeks, with the median at nearly 16 weeks (table A9).

Are you sure that's the only way? Perhaps because the CPS is a random sampling, which assumes some degree of homogeneity, it undercounts unemployment due to mass layoffs, which are probably localized. In any case, I'm still skeptical that this is evidence the BLS is intentionally publishing false information; if they were doing that, you'd more likely expect too-high agreement among its numbers, rather than loose agreement like we see.

[CmdrWilkens]

U-1 through U-3, sometimes U-4, are the only stats reported in the newspapers and television. You have to find other sources for U-6, which is a more accurate measure. U-6 in July '08 was 10.8%; now, it's 16.3%. In short, the government and media know the real numbers, but choose not to talk about them.

Just as a small additional note what we today call the U-6 is much closer to what was called "unemployed" during the middle decades of teh last century. Unemployment during the great depression, in other words, is more closely tracked against U-6 than against U-3 as the later doesn't include those who are marginally attached or otherwise under working/temporary working due to economic conditions.

[KrauserKrauser]

But remember kids, it's increased visibility and change you can believe in these days...

Oh wait.

I need to find a more pessimistic adviser to match my feelings of the future market. The numbers can only be fudged for so long until the shit hits the fan and this is just more of the same we've been seeing for years now. If anything Obama feels like he needs to fudge the numbers or the big correction that has been waiting around will fall on his shoulders and not the next guy.

Damned if you do, damned if you don't but his claims of increased honesty and visibility reveal him to be a complete hypocrite, but what can you expect from a politician, the unvarnished truth? Likely not.

[Surlethe]

Did you read the criticisms of J's argument and decide to ignore them in favor of a conspiracy theory with zero evidence? Or did you just skip over them entirely?

[J]

Chris Martenson's commentary on the BLS's benchmark revisions which were released last Friday. Over 800,000 phantom jobs were created by the birth/death model from the start of the recession to March of this year, and it's likely that another 800,000 more have been pulled out of thin air since then. Also, real job losses based on the household survey came in at just under a million for the month of September as opposed to the official headline number of 263k. It's nice of the BLS to admit their model is wrong and retroactively correct the numbers, but it would be even nicer if they fixed their models to better reflect reality. Maybe next year.