Don't Trust a Single Number in This Jobs Report (or Any Jobs Report)

It is a sad statement about either American politics or American sanity that when today's jobs report showed the unemployment falling below 8% for the first time in Obama's presidency, the reaction from some administration critics was: The numbers are cooked!

The conspiracy claims are beyond silly, as my colleague David Graham explained today. But the employment-truthers are right about one thing. No number in this jobs report is to be trusted.

They're all wrong, probably.

The reason why isn't a cover-up that reaches into the highest levels of government. There's a more prosaic reason you should be wary about the precise numbers in each jobs report -- and especially in a first estimate of the previous month. Each report represents a statistical estimate that will be repeatedly revised. Just about every number in today's jobs report will probably change in the next few months.

The proof that you shouldn't trust any initial number from the Bureau of Labor Statistics is right here ... in today's report from the Bureau of Labor Statistics.

The change in total nonfarm payroll employment for July was revised from +141,000 to +181,000, and the change for August was revised from +96,000 to +142,000.

The BLS is telling us, basically, that it messed up. The first guesses for job creation in July and August were really wrong. In fact, 86,000 people got jobs that weren't counted at first.

This is natural. When the BLS releases initial job creation figures, its "confidence interval" for each monthly change is plus- or minus-100,000. We're only creating 150,000 jobs a month. The confidence interval is two-thirds our monthly job creation!

That's not all. Every year, the BLS announces an even bigger whoopsies. In its annual "benchmark revision" released two weeks ago, the agency reported that the economy had created a whopping 450,000 jobs more than previous estimates in the last year. That figure could be revised again before next year, but for now, BLS is admitting that its estimates have been off by an average of 37,000 jobs per month. That's a big deal.

These revisions on top of revisions on top of revisions tell us two things.

First, counting jobs in a labor force of 150 million is hard stuff. The BLS uses two surveys -- one that polls employers and counts jobs, and another than polls households and calculates the unemployment rate. Sometimes the surveys sing in unison. Sometimes merely in close harmony. But other times, they scream inconsistencies at each other. Still, after a few corrections, they usually offer a fairly clear picture of the economy.

Second, exactly because counting jobs is hard, corrections can swing wildly and change our impression of the economy's health. So don't put too much faith or attach too much significance to one set of numbers. Reach back a few months. Seek the broader trends.

And remember: Just last year, we first thought we created a grand total of zero jobs in August. Zero. Economists and policymakers experienced a collective nervous breakdown. But BLS now says we created 85,000 jobs that month. What explains such madness? Not a conspiracy. Not a cover-up. Just stats and revisions.