The jobs report released earlier this month reported a surprising amount of jobs growth, even though it continued to show that we aren't producing enough to dig ourselves out of 9.6 percent unemployment. Last week, I noticed that Henry Blodget, the disgraced former investment analyst who has reinvented himself as a financial journalist, had a new post up on his business news website: "Wow -- Check Out How Blatantly Our Government Misled Us With The October Jobs Numbers!"
Blodget links to another story that ran on his website arguing that October's results were overstated and that the Bureau of Labor Statistics is fudging the numbers by changing the way it adjusts the employment report for seasonal changes, implying some sort of unspoken conspiracy. He concludes:
In other words, it wasn't that there were a surprising number of jobs created in October. It was that the government changed its "seasonal adjustment" assumption in a way that made it look as though there were a surprising number of jobs created in October. .... If our government is going to publish a number like this that represents such a major "surprise," we would expect it to at least be upfront about the reasons for the surprise. And in this case those reasons had NOTHING to do with the jobs market, and EVERYTHING to do with the seasonal adjustment assumption.
That's a pretty bold conclusion, considering that Blodget and the people he quotes didn't bother to call anyone at the BLS. I did, and it turns out Blodget doesn't have much to back his claim. When the BLS collects data from businesses -- the "establishment survey" -- to track the number of jobs, it wants to capture meaningful economic trends. That means the number most people focus on is "seasonally adjusted," so that we factor out the change in jobs related to holidays, the weather, and other temporary stimulus -- we just want to see the overall trend in the labor market. Blodget and his sources charge that the BLS is manipulating the seasonal adjustment process to create "phantom jobs":
[In October 2010], the seasonal bar which the payroll data must jump was (inexplicably and dramatically) lowered from prior Octobers... Thus, in October 2009, the BLS set the bar at 870,000 jobs, similar to the 840,000 it anticipated in October 2008. This year, by contrast, it lowered the bar to 768,000. Mumbo, jumbo, payrolls presented "an upside surprise" of 100,000.
I spoke with Chris Manning, an economist who is the National Benchmark branch chief and a 15-year veteran of the BLS. Manning explained that seasonal adjustment process hadn't changed in October. As ever, it was run by X-12-Arima, software that averages the last 10 years of October jobs data to get a sense of how many jobs are typically created. It also weighs other factors, including the usual change from September to October, and the length of the month (more on this later) to create the number that Blodget's sources complained about -- the "bar" above which we count employment gains as seasonally adjusted.
This "bar" is called normal seasonal movement. We can track it by subtracting the seasonally adjusted one-month change in employment from its unadjusted equivalent in each October. When you graph the results, it looks like this over the last decade:
As you can see, the October numbers from 2008 and 2009 that Blodget's sources found so damning are actually outliers themselves. Manning suggests one explanation for that: In 2008 and 2009, the measuring period for October -- from the week of the 12th of September to the week of the 12th of October, known as "reference weeks" -- was five weeks long, not four, so X-12-Arima would have priced-in a larger-than-expected change in employment than in a four-week period.
Manning also reminds us that regardless of how the seasonal adjustment is done, it doesn't add or subtract jobs over the course of an entire year, making concerns about phantom jobs seem a bit silly. And to Blodget's point about the data having "NOTHING" to do with the jobs market, it is worth noting that this past October's unadjusted data -- before the BLS runs it through X-12-Arima -- shows the largest growth in jobs during the month (919,000 jobs) than any October in the past 10 years with the exception of 2004 (982,000). Maybe it's something about election years?
Blodget hasn't revisited his post on unemployment numbers. It's too bad he felt the need to spread an unconfirmed conspiracy theory, but perhaps he felt the real explanation -- involving boring old statistics and a five-week month -- wouldn't attract readers seeking entertainment. But if his readers wanted to understand whether economic data gathered by the government is trustworthy, they'd be a bit lost.
-- Tim Fernholz
(This post has been updated for clarity.)