Why September's Unemployment Number is Suspicious

Randall Hoven
The Bureau of Labor Statistics said the unemployment rate dropped from 8.1% in August to 7.8% in September, after having been over 8% for the previous 43 months. In fact, this latest figure puts the unemployment rate exactly back to where it was in the month of President Obama's inauguration.

On its face, that is not necessarily suspicious. First, the rate had been trending downward since reaching its peak in October 2009. And a drop of 0.3 percentage points is not unprecedented. In fact, since 1948 it has dropped at least that much 63 times, including as recently as January 2011.

The change in total non-farm jobs in the Establishment Survey was also in line with expectations. The "expert" consensus was an increase of about 111,000. My own prediction was 99,000. The BLS figure was 114,000 -- not all the different from expectations.

The seasonal adjustment to the employment level used in the unemployment rate calculation was also not out of line. The adjustment to the employment level was downward, and by a bigger amount for any September from 1985 through 2009. So this does not look like a case of magical seasonal adjustment.

But here is what I do think is suspicious: the raw employment level of the Household Survey. That number gets seasonally adjusted and then goes into the unemployment rate calculation. If that raw number is off, the unemployment rate will be off. And that number is certainly off if we judge by history.

  • The August-to-September change, +775,000, was the largest upward Aug-Sep change in the history of the Household Survey.
  • That was the largest difference between the Household Survey and the Establishment Survey (Aug-Sep change) in the history of the Household Survey.
  • That was the only time ever that the Household Survey showed a greater increase in September than the Establishment Survey. 

 

We have been assured by "budget wonks" that "It's a statistical anomaly, not a conspiracy." If so, that was quite a statistical anomaly: the biggest in history, meaning the biggest in 65 years (for the Aug-Sep change).

I'll do some math for you. Based on the previous 64 years (and assuming Aug-Sep changes are independent of each other year to year), the chances of seeing an increase that big in the Household Survey in September was 0.14%. That is well outside most reasonable statistical "confidence intervals." In short, most statistical tests would say this was not a "statistical anomaly."

Of course, nothing can be proved. Rare and improbable events can happen. It's possible that an event that is expected to occur only once about every 700 years happened to occur for the first time in 65 years just one month before a presidential election.

Unlikely things can happen. Like Mahmoud Ahmadinejad and Hugo Chavez getting re-elected and Saddam Hussein getting 100% of the vote. So maybe there's no reason to be suspicious.

Randall Hoven can be followed on Twitter.

The Bureau of Labor Statistics said the unemployment rate dropped from 8.1% in August to 7.8% in September, after having been over 8% for the previous 43 months. In fact, this latest figure puts the unemployment rate exactly back to where it was in the month of President Obama's inauguration.

On its face, that is not necessarily suspicious. First, the rate had been trending downward since reaching its peak in October 2009. And a drop of 0.3 percentage points is not unprecedented. In fact, since 1948 it has dropped at least that much 63 times, including as recently as January 2011.

The change in total non-farm jobs in the Establishment Survey was also in line with expectations. The "expert" consensus was an increase of about 111,000. My own prediction was 99,000. The BLS figure was 114,000 -- not all the different from expectations.

The seasonal adjustment to the employment level used in the unemployment rate calculation was also not out of line. The adjustment to the employment level was downward, and by a bigger amount for any September from 1985 through 2009. So this does not look like a case of magical seasonal adjustment.

But here is what I do think is suspicious: the raw employment level of the Household Survey. That number gets seasonally adjusted and then goes into the unemployment rate calculation. If that raw number is off, the unemployment rate will be off. And that number is certainly off if we judge by history.

  • The August-to-September change, +775,000, was the largest upward Aug-Sep change in the history of the Household Survey.
  • That was the largest difference between the Household Survey and the Establishment Survey (Aug-Sep change) in the history of the Household Survey.
  • That was the only time ever that the Household Survey showed a greater increase in September than the Establishment Survey. 

 

We have been assured by "budget wonks" that "It's a statistical anomaly, not a conspiracy." If so, that was quite a statistical anomaly: the biggest in history, meaning the biggest in 65 years (for the Aug-Sep change).

I'll do some math for you. Based on the previous 64 years (and assuming Aug-Sep changes are independent of each other year to year), the chances of seeing an increase that big in the Household Survey in September was 0.14%. That is well outside most reasonable statistical "confidence intervals." In short, most statistical tests would say this was not a "statistical anomaly."

Of course, nothing can be proved. Rare and improbable events can happen. It's possible that an event that is expected to occur only once about every 700 years happened to occur for the first time in 65 years just one month before a presidential election.

Unlikely things can happen. Like Mahmoud Ahmadinejad and Hugo Chavez getting re-elected and Saddam Hussein getting 100% of the vote. So maybe there's no reason to be suspicious.

Randall Hoven can be followed on Twitter.