Every month (on average), for about a year, there has been a startling discrepancy between employment as measured by the household survey and jobs as reported by the establishment survey.

I believe the discrepancy is yet another Obamacare artifact.

Jobs vs. Employment Discussion

Before diving into the details, it is important to understand limits on data, and how the BLS measures jobs in the establishment survey vs. employment in the household survey.

Establishment Survey: If you work one hour that counts as a job. There is no difference between one hour and 50 hours.
Establishment Survey: If you work multiple jobs you are counted twice. The BLS does not weed out duplicate social security numbers.

Household Survey: If you work one hour or 80 you are employed.
Household Survey: If you work a total of 35 hours you are considered a full time employee. If you work 25 hours at one job and 10 hours at another, you are a fulltime employee.

Recall that the definition of fulltime under Obamacare is 30 hours, but fulltime to the BLS is 35 hours.

Next, consider what happens under Obamacare if someone working 34 hours is cut back to 25 hours, then picks up another parttime job.

Obamacare Effect

Prior to Obamacare
34 hours worked = 1 parttime job household survey
34 hours worked = 1 job establishment survey

Enter obamacare
Person cut back to 25 hours and takes a second job for 10 hours
Here is the new math

25 + 10 = 1 fulltime job on the household survey.
25 + 10 = 2 jobs on the establishment survey.

In my example, the household survey totals up all the hours and says, voilla! (35 hours = full time). So a few extra hours that people pick up working 2 part time jobs now throws someone into full time status – thus no surge in part-time employment, but there is a surge in jobs.

I am quite sure this is what is happening, but I cannot prove it.

Household Survey Normalized

The BLS has a chart (shown below) that normalizes the household survey to the establishment survey, but that just transfers establishment survey double-counting to the household survey!

I contacted the BLS and asked if they could please weed out duplicate social security numbers. They can’t because they do not capture social security numbers.

This is not a fault of the BLS. They wish they had more data but they don’t.

Does Any Available Data Lend Credence to My Theory?

Yes, and overwhelmingly so. An unusual discrepancy between the household and establishment surveys is the key to the puzzle.

Household survey: http://research.stlouisfed.org/fred2/series/CE16OV
Establishment Survey: http://research.stlouisfed.org/fred2/series/PAYEMS

Numbers are in thousands.

October Employment and Jobs vs. October in Prior Years

Year-Over-Year Gains or Losses vs. Prior Years

Fore the year ending October 2009, 2010, 2011, and 2012, the household survey and the establishment survey were very well aligned.

However, something happened between October 2012 and October 2013. In the last year, the household survey says employment rose by 20,000 a month while jobs rose by 194,000 per month!

Let’s drill down by month and take a look.

Month-over-Month Gains or losses vs. Prior Month

Because of the government shutdown, some will object (and rightfully so) about October. So let’s throw that month away.

12-Month Results Excluding October 2013

Even after eliminating the government shutdown effect, the difference between the two surveys is still huge.

From October 2012 through September 2013, the household survey suggests employment rose by an average of 111,000 per month. The establishment survey suggests 207,000 jobs per month on average.

Which is correct?

Actually because of what they measure, both might be. Thus my blog subtitle “And Way Less Job Growth than Anyone Thinks” is not technically accurate.

Practically speaking however, job growth has been nowhere near as good as it looks. People picking up a second parttime job following cutbacks in hours does not do a thing for the economy except perhaps waste gasoline.

However, in spite of strong evidence, this is still a theory. To prove it, we need to weed out duplicate social security numbers. The BLS can’t, but ADP can. I contacted them twice but to no avail.

I would like ADP to crunch the data and determine how many duplicate social security numbers show up vs. the same months in prior years. If I am wrong it won’t be the first time. But let’s have a look at the numbers and see what they say.

Mike “Mish” Shedlock