This piece exists because a reader, an economist and professor, wrote in after the last one and pointed out that it gave the margin on the first estimate and never the margins on the revisions. I looked into it... and the details turned out to be more interesting than the omission.
Remember: August payrolls came in at 162,000. The Bureau of Labor Statistics says the margin around a monthly figure like that is roughly plus or minus 119,000. I shared a little bit more about this on LinkedIn and in the last edition, How Much of a Single Monthly Print Can You Act On?
Now, if you wait one month, you'll see that typically the same figure carries a margin of plus or minus 98,000. Getting smaller!
That second number is published. It sits in a table on the BLS site that I have never seen anyone quote. It is worth knowing, because it turns "the jobs number gets revised" from a complaint into something you can plan around. But the more useful part is not the 21,000 jobs of margin you buy by waiting. It is the reason the margin shrinks at all.
Six of the last twelve monthly prints sit inside that first margin. Not wrong, not fabricated, just not distinguishable from no change at all on the strength of one month. That includes September, November and December of 2025, and May, June and July of this year. The number is real. The confidence you can put on it, alone, in the month it lands, is smaller than the way it gets reported implies.
Firms did not stop reporting
Here is the part I did not expect. The survey does not reach every business. BLS asks a sample, one it says covers roughly a quarter of all payroll jobs, and puts the same question to it every month. On the morning the number is published, it has heard back from about six in ten of that sample. That share has been falling for a decade: 75.2 percent in 2017, 60.4 percent in 2024, 63.0 percent so far in 2026.
That much has been written about. What has not is the other line.
By the time the same month reaches its third estimate, it has heard from more than nine in ten of that same sample. And that share has barely moved. Measured to the same year, 2017 to 2024, the release-day rate fell 14.8 points and the final rate fell 4.6, from 95.5 percent to 90.9. Last year it was 91.5.
So here's the important point: the businesses are answering, just later. The gap between what BLS has on release day and what it has two months on used to be about twenty points. It is now about thirty.
That reframes the whole complaint. The jobs data did not get less reliable. The first draft of it did.
