The numbers are broken.
The September job numbers are out. The Washington Post reports: 
The U.S. economy created far fewer jobs in September than
anticipated, but revised job figures for previous months showed a
surprisingly substantial uptick over what was previously reported.
national economy created 51,000 jobs in September — analysts had been
predicting 120,000 — and the unemployment rate dipped slightly from
4.7 percent to 4.6 percent, according to figures released today by the
At the very end of the article, in the very last paragraph, in the very last sentence, comes the truly newsworthy part of the article. Here’s the last paragraph:
The good news was that job gains for both July and August turned out to
bigger than previously estimated, taking some of the bite out of
September’s figures. The report also showed that job growth during the
12 months ending in March may have been 45 percent higher than
previously reported. The Labor Department said payrolls for the 12
months that ended in March 2006 will be revised upwards by a whopping
810,000 jobs, the biggest revision since 1991.
I like that word "whopping." The Post reporter, Daniela Deane, realizes that an average monthly error over 12 months of 67,000 jobs is an incredible error.
These revisions are because the Bureau of Labor Statistics establishment survey is a sample of existing businesses that has trouble capturing some sources of new job growth—small businesses and the self-employed. The BLS household survey which used to move roughly in parallel as a measure of job creation has been diverging wildly from the establishment survey. The pessimists quotes the establishment survey. The optimists quote the household survey. Revisions of this size suggest the optimists had a point.
(HT: Brian Wesbury)