They measure different things, and both are surveys. The establishment survey counts jobs on company payrolls. The household survey counts people. One worker holding two jobs is two in the first and one in the second, so the two can move in opposite directions in the same month with neither being wrong.
The number worth carrying: how big a move has to be before it means anything. A payroll change has to reach about 122,000 to stand out from sampling noise. A household employment change has to reach about 650,000. Most monthly headlines are inside that band, which is why a single month is close to uninformative and a run of months is not.
January breaks the household series. The survey resets its population estimates every January. In January 2026 that reset alone took 0.4 of a point off the participation rate in one month. A December-to-January comparison mixes a real change with a definitional one, so it needs care or a footnote.
Payrolls get checked against tax records once a year, and the revisions are large. The preliminary estimate of that benchmark arrives around late August and the final one the following February. August 2024 and August 2025 were each revised down by about a million jobs. So a level you quoted last year may not be the level today, and the revision is not noise: tax records are closer to a census than the monthly sample is.
There are two different layoff numbers and they answer different questions. JOLTS layoffs count people let go during a month, a flow. Permanent job losers count people laid off and still out of work, a stock. Those can diverge, and in 2026 they did: the flow stayed flat while the stock rose, because the same number of people were losing jobs and taking longer to find the next one. Median time out of work went from 9.6 weeks to 11.4. Read only the flow and you conclude nothing is happening.
And one month does not exist. October 2025 has no household survey, because it was not carried out during the government shutdown. Any code that fills a gap by interpolating will invent a value there, and any year-on-year comparison through that month is comparing against nothing.
The series are on https://www.tigzig.com/tremor (US macro, labour). The substantive question this sits under: is US unemployment worse than the headline. Full write-up: https://www.tigzig.com/post/us-jobs-calm-before-storm-sep2026.
Building something like this? How I work covers the rates, the availability and what I take on.