September's Unemployment Number Still Suspicious

Jack Welch and many others think the September unemployment rate was strange. Mr. Welch said this.

"I doubt many of us know any businessperson who believes the economy is growing at breakneck speed, as it would have to be for unemployment to drop to 7.8% from 8.3% over the course of two months."

My original argument for suspicion had nothing to do with economic growth; it came from the Household Survey itself. I had discovered that the Household Survey never had as huge an increase from August to September as it did this year. Also, the Household Survey had never showed a bigger increase in September than the Establishment Survey showed.

(The Bureau of Labor Statistics conducts two surveys every month, one by contacting employers and the other by contacting households. They both measure how many people are working, but the results can and do differ. The Household Survey number is the one used to compute the unemployment rate.)

And when something happens that has never happened before, I think suspicion is called for. Statistics 101 backed me up.

Mr. Welch and others brought up a good point: the unemployment rate does not usually go down quickly unless the overall economy is growing rapidly. And Mr. Welch specifically cited the two-month drop of 0.5 percentage points.

How many times has that happened since the BLS started calculating unemployment? Thirty-six times, 33 of which were prior to Obama. So while it is a relatively rare occurrence (less than 5% of the time since 1948), it is not unheard of. But if we look at how Gross Domestic Product was growing when such unemployment rate drops occurred, the September number looks suspicious yet again.

I tabulated all two-month changes in the unemployment rate since 1948. There were 775 data points, from March 1948 to September 2012. The table below shows all the 36 times the unemployment rate drop was at least 0.5 percentage points.

Unemployment rate drops over two months of 0.5% or more

Here are my observations from that table:

• There were no such drops in the unemployment rate through all of President Clinton's "boom" years. In fact, there were no such drops from 1985 to 2009 - a full quarter-century, and a quarter century of mostly healthy economic growth.
• Since 1962 and prior to President Obama's presidency, there were 9 such occurrences. Eight of them happened during the Reagan recovery of 1983-84 when GDP was growing at rates of 7.1% to 8.5%. The only other occurrence was during President Ford's recovery in 1976, when real GDP was growing at a 9.4% annual rate.
• Of the 33 occurrences prior to Obama's presidency, only two occurred when real GDP growth was slower than 2.5% annually: 1949 and 1956.
• All three of the occurrences in Obama's presidency occurred with a GDP growth rate of 2.2% or less. (Although we do not know what the growth rate was in September, no one expects it to be too different from the 1.3% rate of April-June.)

In short, something that had not happened at all since 1956 happened three times in Obama's presidency. In more than a half-century preceding Obama, the only times the unemployment rate dropped 0.5% in two months were when real GDP was growing at annual rates of 7.1% to 10.5%.

Those rates were breakneck rates of growth. We haven't seen such rates since 1999. As Jack Welch said, no one "believes the economy is growing at breakneck speed." Not even Obama's economists would claim anything near 7%. It's been 2% or less for the last six months.

Prior to Obama, the chance that the unemployment rate would drop 0.5% or more over two months with a real GDP growth rate under 2.5% was two out of 33. The chance of that same thing happening three times in a row then, as it did under Obama, is 0.0223%. Or let's look at it this way: what had happened only twice in the 62 years from 1948 through 2009, happened three times in the last three years under Obama.

A straightforward statistical test, again, says that should not happen by random chance. That is, we reject the "statistical anomaly" hypothesis and accept the alternate hypothesis that those Obama data points were genuinely different from the rest somehow.

Statistics cannot tell us "how" those points were different, just that they don't pass the "statistical anomaly" smell test.

To review, looking simply at the historical Household Survey itself, we found that September 2012 was unprecedented. Comparing the Household Survey to the Establishment Survey, we found, again, an unprecedented difference. And now, looking directly at the unemployment rate and comparing it to GDP growth, we find more anomalies. That is three statistical strikes against the September number.

That might not be enough evidence to convict, but it is sure enough for "probable cause" to dig deeper. Anyone besides me care to do such digging? It seems kind of important if it means someone in Obama's administration is purposely cooking the books.

(Data source: St. Louis Fed/FRED. The unadjusted Household Survey is the Employment Level (LNU02000000) series. The unadjusted Establishment Survey is the PAYNSA series. The unemployment rate is the UNRATE series. Real GDP is the GDPC96 series.)

Randall Hoven can be followed on Twitter.