Technology

Spinning Employment Data

nvestors need an inoculation against the “spinning virus,” especially when it comes to employment data. The sources and reports are so complicated that it is easy to find something to prove your point – whatever that might be!

Get The REITs eBook in PDF

Get our PDF study on REITs and our other investor studies! Save it to your desktop, read it on your tablet, or email to your colleagues.

Source Timing Errors Revisions Accuracy/Value Timeliness
Business Dynamics About eight months old Small. An actual count, not a sample. Small, but stretched out over many months. Very accurate. Based on state employment premiums. Ancient history.
Initial Claims One week old. Small. Not a sample. Modest, and only delayed by a week. High accuracy, but only shows job losses, not gains. Excellent. Only one week delayed.
ADP Straddles the week including the 12th of the month. An actual count, not a sample of client companies, but only 25% coverage. Tweaks to extend the results. One revision, usually modest. Good accuracy, especially when viewed versus business dynamics rather than the BLS. Excellent.
Payroll Survey Based on the week including the 12th of the month. Counts jobs, not people. Confidence interval +/- 100K just for sampling error, which persists after initial revisions. Two revisions in the next months and eventual “benchmark” revisions. These may be significant. Good accuracy when measured against total jobs, but plenty of error on monthly changes. Pretty good. Sacrifices some accuracy for a speedier report.
Household Survey Also uses survey from mid-month Counts people, not jobs. One person might hold several jobs. Sampling error for total jobs is +/- 450K No effective follow-up or revisions. Interpretation requires a multi-month series. Reasonably accurate for labor force participation and unemployment rate. Less so for total employment. Sacrifices some accuracy to provide frequent reports.

 

The table can help you avoid some of the most significant errors. Many people confuse, for example, the time of the data release with the time of events. Here are two examples you will see:

  1. The payroll and household survey results cannot be an outcome of the tax cut legislation. The surveys were done before the legislation was passed.
  2. The initial claims – up to 250K last week – was a report from after the payroll and household survey periods. Whatever the implications of the increase in claims, we will not see it until next month.

You will also see claims of a “big miss” or “big gain” from changes that are well within the sampling error, without even considering the non-sampling error. A deviation of 50 – 70K, either way, means little.

Pundits know this, give lip service, and then proceed as if there is a high level of accuracy.

Many will opine about the seasonal adjustments. Especially those who no experience or training on this subject. It is often easy to make something look wrong based upon a single month.

And finally, the important context is that the jobs report is based upon a net change. Despite the hype, place this report in the context of other economic data. The actual number of jobs created (and lost) in a month is over 7 million. We are only analyzing the froth at the top of the waves.