New jobs is the most difficult of numbers for the BLS to measure. They are trying to count a change of 50,000 to 150,000 each month out of 165,000,000. That's a change of 0.03% to 0.09%. These are very tiny numbers they are trying to suss out.It is interesting to see this larger revision in the context of the CPI accuracy discussion.
The question that comes to my mind: should we expect more or larger revisions to similar reports from the BLS or other U.S. economic institutions?
Note that the US Census takes many months and uses about 500,000 census takers to get their numbers. The BLS is doing it every month with about 2000. And the census response is legally mandated. Business responses to BLS surveys are totally voluntary. And when times are tough or business conditions uncertain, businesses are more reluctant to respond to voluntary surveys. History shows that you get larger revisions when economic conditions are in transition because response rates are lower.
They don't have the personnel to survey every business so they have to do with sampling and statistical estimates. There's a lot of noise in the measurements. There are bout 5 million people who lose and gain jobs every month which alone is an order of magnitude bigger than the new jobs number they are trying to measure. Another big estimate is the millions of businesses in the "birth-death" model. This is the churn of millions of new and defunct businesses that can't be sampled.
So they have to get the best numbers than can by sampling from about 120,000 businesses of the 35 million businesses in the US. Compare this to an election poll in which 4% or 5% is considered are good margin of error. The BLS is trying to measure 100 times that accuracy.
For all of these reasons, the monthly reports will always have a large margin of error. In fact, they state clearly in each monthly release that this is just a first estimate with a confidence interval of about 136,000 plus or minus. But of course most folks don't read the footnotes and don't realize it is a first estimate and then get all upset about revisions they should have expected.
As more data comes in over the following months, revisions are made to increase the accuracy of the original numbers. One of the biggest assists is actual payroll tax numbers from the IRS. But businesses only file these taxes to the IRS once a quarter and it takes some months after that to get to the BLS. And many businesses that did not respond to their original voluntary survey, finally get around to in following months.
So in the end, up to a year later, with all the revisions, the numbers are very accurate, but one should not be surprised that revisions occur. And when economic conditions change is when bigger revisions are expected.
Compared to the employment surveys, the CPI measurements are infinitely simpler.
Statistics: Posted by Franz — Fri Sep 12, 2025 12:04 pm — Replies 364 — Views 40389