Reading the US jobs report, from payrolls to the unemployment rate

Two surveys in one release

The establishment survey, the Current Employment Statistics (CES) programme, asks employers how many people are on their payrolls. BLS’s technical note says it covers about 119,000 businesses and government agencies, representing about 622,000 worksites. Its reference period is the pay period that includes the 12th of the month. It produces the payroll count, average hourly earnings and average weekly hours.

The household survey, the Current Population Survey (CPS), interviews about 60,000 eligible households about whether the people in them are working, looking for work or neither. It produces the unemployment rate, the participation rate and the employment-population ratio.

The two measure different things. The payroll count is a count of jobs, so a person on two payrolls appears twice; the household survey counts people once, and includes agricultural workers, the unincorporated self-employed, unpaid family workers and private household workers, whom the payroll survey leaves out. They often disagree from month to month. In September 2026 household employment (BLS series LNS12000000) rose from 162,746,000 to 163,152,000, a gain of 406,000, while payrolls rose 29,000 (our arithmetic on BLS levels). The household sample is much smaller, so its monthly changes are noisier.

Worked example: the payroll change and its average

BLS publishes payrolls as a level, in thousands, seasonally adjusted (series CES0000000001). The headline number is the change in that level. Using the values in the BLS database on 20 October 2026:

September: 159,044 − 159,015 = +29 thousand. August: 159,015 − 158,882 = +133 thousand. July: 158,882 − 158,892 = −10 thousand.

Because single months are noisy, many readers also look at an average. The three-month average change is (29 + 133 − 10) ÷ 3 = 50.7 thousand (our calculation). The release itself compares September with the prior 12 months, which it puts at an average gain of 45,000. That figure can be rebuilt from the levels: August 2026 minus August 2025 is 159,015 − 158,472 = 543 thousand, and 543 ÷ 12 = 45.25 thousand.

BLS marks the two most recent months, August and September, as preliminary. They will be revised in the next two releases.

Worked example: the unemployment rate and its rounding

The unemployment rate is the number of unemployed people divided by the labour force, published to one decimal place. In September 2026 the household survey counted 7,109,000 unemployed (series LNS13000000) in a labour force of 170,262,000 (LNS11000000): 7,109 ÷ 170,262 = 4.175%, published as 4.2%. In August the figures were 7,031,000 and 169,777,000: 4.141%, published as 4.1% (our arithmetic).

So the published rate rose 0.1 percentage point, but the unrounded rate rose only about 0.03 point. A one-tick move in the headline rate can come from a much smaller change underneath it. BLS’s technical note gives the 90% confidence interval for the monthly change in the unemployment rate as about ±0.3 percentage point, larger than either figure.

The participation rate (LNS11300000), the share of the adult civilian non-institutional population in the labour force, was 61.8% in September against 61.6% in August. The release describes both it and the unemployment rate as little changed.

Worked example: wage growth from average hourly earnings

Average hourly earnings for all employees on private nonfarm payrolls (series CES0500000003) were $37.81 in September 2026 and $37.76 in August. The monthly change is $0.05, and 0.05 ÷ 37.76 = 0.13%, which BLS reports as 0.1%. A year earlier, in September 2025, the figure was $36.70, so the 12-month change is 37.81 ÷ 36.70 − 1 = 3.02%, reported as 3.0% (our arithmetic on BLS levels).

Earnings are revised too. The August figure was first published as $37.75 on 4 September 2026; in the database it is now $37.76. That one-cent revision is part of why the September monthly gain shows as 5 cents rather than 6.

Average hourly earnings is an average across whoever is on payrolls that month. If lower-paid jobs are cut, the average can rise without any individual being paid more, so the series reflects the mix of jobs as well as pay rates.

Sampling error, revisions and benchmarks

Every headline figure is an estimate from a sample. BLS’s technical note says the 90% confidence interval for the monthly change in total nonfarm employment is on the order of ±122,000. For September that gives a range of 29 − 122 = −93 thousand to 29 + 122 = +151 thousand (our arithmetic), which includes zero. That is the statistical sense in which BLS calls the change “little changed”.

Monthly revisions come as more employer reports arrive and seasonal factors are recalculated. July 2026 shows how far they can travel: first published as −23,000 on 7 August, revised to +21,000 on 4 September, then to −10,000 on 2 October. August went from +162,000 to +133,000 in its first revision.

Once a year the payroll survey is benchmarked to universe counts of payroll employment from unemployment insurance administrative records, and the gap between the survey and those counts for March is the benchmark revision. In the release of 11 February 2026, BLS revised the seasonally adjusted March 2025 payroll level down by 898,000; the technical note puts the average absolute benchmark revision over the prior ten years at 0.2%. The household survey has its own annual step: new population controls, which in 2026 BLS said it would introduce with the February data rather than January.

Seasonal adjustment and gaps in the data

Hiring and lay-offs follow a yearly calendar: retail staffing before the holidays, school staff in summer, construction in warmer months. The technical note says this seasonal variation can be very large, and gives the example of education payroll employment, which falls by about 20% at the end of the spring term. BLS removes those regular patterns with seasonal adjustment, and the headline figures are all seasonally adjusted. The not-seasonally-adjusted series are published alongside them.

The household series in the BLS database have no October 2025 values. BLS marks that month “Data unavailable due to the 2025 lapse in appropriations”, so a 12-month comparison of the unemployment rate for October 2026 will have no October 2025 starting point from the household survey.

When it comes out, and how to check the numbers

BLS’s release schedule lists the next three Employment Situation reports for 6 November 2026 (October data), 4 December 2026 (November) and 8 January 2027 (December), each at 8:30 a.m. ET. Each release also prints the date of the next one at the foot of its summary. The same dates appear in the release schedule in our site header, and the Treasury yields that traders compare before and after each release are on /rates with their dates.

To redo the arithmetic above, open the BLS public data API or the BLS data viewer and request the series IDs: CES0000000001 for payrolls, LNS14000000 for the unemployment rate, LNS13000000 and LNS11000000 for its two parts, CES0500000003 for average hourly earnings and LNS11300000 for participation. FRED carries the same figures as PAYEMS and UNRATE. The current release text, at bls.gov/news.release/empsit.nr0.htm, states the revisions to the prior two months; the archive of past releases shows what each month looked like when it was first published.