What the Jobs Report Counts, and Why the Number Changes After It's Announced
The most watched number of the month is a first count from a survey of employers. It is revised twice, then again a year later, and the revisions have a direction.
October 28, 2026
In one sentence
"Jobs added" is the change in the number of people on nonfarm payrolls from one month to the next, estimated from a survey of about 120,000 businesses and government agencies and published on the first Friday of the following month.
It counts jobs, not people: someone with two jobs is counted twice. It leaves out farm workers and the self-employed, which is what "nonfarm payroll" means. A second survey, of households, produces the unemployment rate and its own count of people working; the two agree over a year and disagree in most individual months.
New articles and charts arrive in one short email on Monday mornings. Get the weekly digest
The analogy
The first count on election night. By eleven o'clock you know roughly who won, from the precincts that reported fast. Over the next days the mail ballots and the slow counties come in and the margin moves, sometimes a lot, rarely the winner. The first jobs number is the fast precincts: the employers who returned the survey in the first two weeks, about two-thirds of them. The rest arrive over the next two months and the number is revised each time. Then once a year the whole count is checked against the tax records every employer files, and the previous year is recounted at once. That is the benchmark revision, and it can move a year's total by several hundred thousand.
How it reaches your kitchen table
Markets: within the hour. Stocks, bonds and mortgage rates move on this number at 8:30 Eastern on jobs Friday, in the direction of the surprise rather than the level. A weak print raises the odds of a Fed cut, which lowers bond yields, which lowers mortgage rates the same morning.
The Fed: at its next meeting. This is the headline reading on the "maximum employment" half of the Fed's mandate. Several weak months argue for cuts; several strong ones with inflation high argue the other way.
Your raise: with a lag. Steady job growth means employers competing for workers. Whether they are competing with money is in the same release: average hourly earnings, which real vs. nominal explains how to read.
Your own job: barely. A national figure of plus 29,000 says nothing about whether your employer is hiring. The industry table in the same release does, and it has told a consistent story for three years: health care adding, manufacturing and federal government shedding, most of the rest flat.
What moves it
Hiring minus separations, across about 160 million jobs. A reading of plus 29,000 means the two nearly matched; it is not 29,000 decisions. Then the noise: the survey's margin of error on the monthly change is roughly 130,000, so a swing from plus 130,000 to plus 30,000 can be nothing at all. Then the size of the workforce. When population growth slows, as it did when immigration fell in 2025–26, the number of jobs needed each month to hold unemployment steady falls with it, from around 200,000 during the 2022–24 immigration surge to well under 100,000 now by most estimates. A plus 50,000 that would have read as weakness in 2019 can be enough today.
How to read this month's number
Noise first. Anything within about 100,000 of last month is the same number. Use the three-month average on the chart; October 2026's 165,000 sits in an average of about 137,000.
Then the revisions. The number you remember from jobs Friday is not the number in the record. Since 2023 the gap between each month's first print and today's figure has averaged about 65,000, nearly all of it downward; in 2024 and 2025 the first prints ran about 90,000 a month too high. That pattern is not new; first prints ran high through 2008 as well. Surveys count late responders, and businesses that are closing don't return surveys.
Then the other survey. The household survey's employment count (a companion on the chart) is noisier, counts the self-employed, and resets every January when new population estimates come in. When the two surveys diverge for months, economists usually trust payrolls month to month and watch whether the household count catches up.
Then wages and hours. Average hourly earnings and the length of the workweek are in the same release and often matter more than the headline: a weak count with rising hours is employers stretching staff before hiring; a strong count with falling hours is the reverse.
Where the analogy breaks
On election night the ballots exist; they are merely uncounted. The jobs number is an estimate from a sample, and part of it, the birth-death adjustment, is a model guessing how many businesses opened and closed that nobody has surveyed yet. When the economy turns, the model lags, which is why revisions tend to run downward while it is slowing. And elections end. The jobs count is revised for a year and then the benchmark resets the level, so there is no final result, only the latest one.
What people argue about
Whether the numbers are "cooked." In August 2025 the President dismissed the head of the Bureau of Labor Statistics hours after a weak report that revised the previous two months down by a combined 258,000, calling the figures rigged. Critics of the bureau point to the run of downward revisions and to a model that fills in businesses nobody has surveyed. Defenders, including former commissioners appointed by both parties, say the pattern is what a slowing economy looks like in a survey that counts late responders; that every method is published before the data; that the commissioner never touches the estimates; and that the household survey, run separately, told the same story. The chart's first-print line shows the same downward pattern through 2008, under a different administration and a different commissioner. What it cannot show is intent.
What a healthy number is. One side: plus 50,000 a month is stall speed and the warning before a recession. The other: with the workforce barely growing, plus 50,000 holds unemployment flat and is the new normal; judge the labor market by the unemployment rate and prime-age employment, not the count. Both agree the breakeven number has fallen. They disagree by how much, and the estimates range widely enough to hold both readings.
Two jobs, one person. Payrolls count jobs; the household survey counts people. When payrolls grow and household employment doesn't, one reading is that more people hold two jobs; another is that the household survey undercounts recent immigrants. The multiple-jobholder rate, published in the same report, has stayed in a narrow band for decades, which favors the second reading without settling it.
See also
- The unemployment rate: the other survey, and who it leaves out.
- Jobless claims: the weekly number that moves before this one.
- Real vs. nominal: reading the wage line in the same release.
- The almanac's jobs added with every overlay, and the first-print view.
Sources
- BLS, Employment Situation technical note, the two surveys, response rates and margins of error.
- BLS, CES benchmark revisions and the birth-death model.
- FRED: Nonfarm payrolls (PAYEMS), household employment (CE16OV), multiple jobholders (LNS12026620); ALFRED vintages of PAYEMS, the first prints.
- Axios, "Trump fires labor statistics chief after weak jobs report" (Aug. 1, 2025).