Employment Report Methodology

Today’s market focus was the monthly employment situation report.  The numbers get a lot of publicity, including the popular media.  Most active market participants know that the numbers are adjusted in various ways and that they also get revised.  Even the best-informed observers miss some key points.  If you read this through to the end, I’ll bet you learn something you did not know.

The featured number, the growth in non-farm jobs, uses the difference between the current and prior months.  Let’s see what that means.  (I’m going to put aside, at least for this particular discussion, the fact that there are seasonal adjustments and a "birth/death" model to account for the changing business population.  It is possible that there is significant non-sampling error.)

The reported results are based upon a survey.  About 400,000 workplaces are supposed to respond with information during the week of the month containing the 12th.  Most people figure that a survey that large must be pretty accurate, and it is.  The 90% confidence interval is about +/-0.1%.  This means that the Bureau of Labor Statistics is estimating the size of the workforce every month and doing it very accurately.

The problem is that the workforce is over 140 million, so a 0.1% error amounts to 100,000 jobs!

So today’s report estimated 135 million jobs.  This means that come September we can be 90% certain that the true number of non-farm jobs is 135 million +/- 100,000.

Why September?  The report was issued today.

Not all 400,000 workplaces report on time.  But they don’t.  Just as some percentage of people will be late filing their income tax returns, some businesses do not report employment data on time.  This is the source of the revisions in the payroll data.  (Sorry to disappoint all of the paranoid hedge fund managers who think that the President brings in a team to massage the numbers, but that is not how it happens.)

The first report, like today’s report for June employment, is based upon returns from about 65% of the workplaces in the sample.  That may seem like a good number, and it does provide meaningful data.  The problem occurs if there is some systemic bias in the businesses that are not filing on time — something business condition that relates to hiring or firing employees.  It does not take much of a bias to alter the result by a tenth of one percent — and that represents 135K jobs.

The second report, which we will not see until August, is based upon returns from about 80% of the workplaces.  The final revision, reported in September when everyone will see it as "old news" accounts for more than 90% of the reporting workplaces, which the government figures is good enough.

So what we did today was compare the first estimate for June with the second estimate for May and calculate the job growth.  We will not know the "official" job growth until September, when we have the final numbers for both periods.  Even then, the result for May and June are both +/- 100,000 jobs.

This is good to keep in mind when you see a parade of experts on CNBC trying to explain why there is a 50K difference from "expectations."

Oh — and here’s something they won’t tell you on TV.  Even those who realize that the job growth is an estimate forget it when they discuss the "internals" of the report.  (This is usually done in a deep and serious voice).

Well where does the BLS get the internals?  These are also survey results, of course!  While they do not report a sampling error for the hourly wage in the document, it is probably +/- five cents.  And to get the error down to that level we need to wait until September.

This is part of a consistent theme on "A Dash."  The immediate market reaction to data is often far too extereme, based upon the value of the new information.