The Right Coast quotes an example

of why you shouldn't trust the first economic reports the government puts out. The media always trumpet the first reports, they never trumpet the revisions.

The payroll survey counts jobs, not workers. But counting payroll jobs is a questionable way of measuring America's evolving work force, especially in light of declining job turnover. The payroll survey's biggest problem is that it systematically double counts workers when they change jobs. Since somewhere between 2 percent and 3 percent of the work force changes employers every month, payrolls tend to be noisy. The illusion of lost jobs in recent years occurred because job turnover declined after 2000, first with the recession, then even more sharply after 9/11. As a result, 1 million jobs have been artificially "lost" in the payroll survey since 2001.
...
This is not the first time the survey has been off. That's why the Labor Department warns against using the real-time payroll figures in the footnotes of its monthly release. In 1992, the media proclaimed a jobless recovery based on preliminary payroll data. Only later did benchmark revisions correct the data that the public sees today, which show the net creation of 900,000 jobs in the year prior to the 1992 elections. The next major payroll revision won't occur until January 2005.

E. G. Ross taught me to watch out for that sort of thing. This post doesn't make that exact point, but it's close, particularly the last section, Balancing Context for Positives. If I find the one I'm looking for, I'll update.

Off the point somewhat, here's something everybody should meditate on: "It is the man-made that must never be accepted uncritically: it must be judged, then accepted or rejected and changed when necessary."
—Ayn Rand