Where August’s payroll gains occurred—and how to read them

Food services and drinking places added 59,000 jobs in August 2026 while information lost 23,000. Those contrasting results sat inside a preliminary national increase of 162,000 nonfarm payroll jobs—jobs on employer payrolls outside agriculture—reported September 4 by the Bureau of Labor Statistics. The national gain exceeded the published monthly average of 31,000 over the prior 12 months.153

Local government education added 42,000 jobs, largely offsetting July’s decrease, with little net change since January 2025. That makes the rebound different from sustained expansion. BLS described manufacturing as trending upward, adding 16,000. Health care added 13,000, slower than its prior-12-month monthly average of 32,000. Construction’s larger positive estimate, 22,000, was described as little change. The supplied evidence does not establish why BLS used that description.1

The release also revised June’s gain from 20,000 to 31,000 jobs and July’s change from a loss of 23,000 to a gain of 21,000. These revisions incorporate additional employer reports and recalculated seasonal factors.1

The size and sign of a single monthly estimate do not alone establish a trend over time. Sampling uncertainty matters: surveying part of the population leaves room for chance differences from the full population. Comparing industries therefore requires more than ranking their monthly numbers. The payroll survey’s sample design prioritizes precision for the national total; that does not establish equal precision for each industry. This explains a limit on the comparison, without identifying the statistical reason for a particular industry’s description.52

For a job seeker, the first interpretive step is identifying what the employer category contains. The Current Employment Statistics survey, or CES, gathers payroll information from businesses and government agencies. It groups worksites by their primary business activity and by ownership: private or public, with public employers divided into federal, state and local government. Together, business activity and ownership identify the part of the economy being measured.2

An occupation describes the work a person does. An industry contains different occupations working for the same type of business, so its payroll change cannot identify opportunities in one particular occupation.4

The next distinction is the unit being counted. Payroll employment counts jobs; the separate Current Population Survey, or CPS, surveys households and counts employed people. BLS supplies a useful job-changing example: someone on two employers’ payrolls during the same reference period contributes two jobs to the payroll measure but one employed person to the household measure. The overlap is enough to produce different counts without requiring two people. Household employment also has broader coverage, including agricultural workers and the self-employed.32

The observation windows differ, too. The household survey uses a seven-day calendar week, usually the week containing the 12th. The payroll survey uses each employer’s pay period containing the 12th, which can run weekly, biweekly, monthly or otherwise. A monthly report therefore does not mean both surveys observe employment over an identical stretch of time. BLS says the effect of these timing differences is unknown and cannot be adjusted away in its comparison. These distinctions explain limits on comparison; they do not establish what caused August’s results.3

The unemployment rate expresses unemployed people as a percentage of the labor force: employed plus unemployed people. Classification requires no employment during the reference week, availability for work that week except for temporary illness, and an active job-search effort within the four weeks ending that week. People temporarily laid off and expecting recall are exempt from the search requirement. The rate describes people’s labor-force status; it does not track applications and offers, so it cannot give an applicant’s chance of receiving an offer.4

Unemployment duration presents another trap. It measures weeks of continuous searching among people still unemployed, including the reference week. For temporary layoffs awaiting recall, it counts full weeks on layoff. Median duration is the midpoint: half have shorter durations and half longer. These spells remain in progress: their eventual endings have not been observed. The median therefore cannot tell a new applicant how long a completed search will take.4

To help readers compare months, BLS uses seasonal adjustment to remove normal recurring variation, such as effects associated with holidays and weather, and reveal underlying trends. That method does not establish a seasonal cause for education’s movement.5

The revision timetable gives readers another way to place a monthly number. Initial payroll estimates are preliminary, revised the following month and again two months after first publication, then held until the scheduled annual benchmarking update. Receiving more information updates an estimate; it differs from the sampling uncertainty that complicates comparisons of monthly changes. Reaching the final monthly revision does not make a sample-based measure an exact count.5

Report an error · Corrections policy