Skip to content
Sign inGet started free

The record · 5 September 2026

Washington shed 339,000 jobs

One country, one monthly survey, and two numbers that point in opposite directions. The month looks ordinary and the year does not, because a single employer is doing something no other sector in this table is doing. Everything below is checkable against the live market.

· 8 min read

815,000jobs added by the private sector over the twelve months to August 2026, against 603,000 added by the whole economy, because government subtracted 212,000
11.25%of federal employment gone since October 2024, while state government fell 0.31% and local government rose 0.31% over the past year
339,000federal jobs fewer than at the October 2024 peak of 3,013,000, a fall of 11.25% in twenty-two months

An ordinary month on top of an unusual year

The August report, published this week, was the kind that moves nothing. Employers added 162,000 jobs, better than July’s 21,000 and June’s 31,000. The unemployment rate stayed at 4.1%, exactly where it was in July, and below the 4.3% of a year earlier. On the month, nothing to explain.

The twelve-month view is where it stops being ordinary. American payrolls grew by 603,000 between August 2025 and August 2026. The private sector grew by 815,000 in the same period. Both are true at once, and the arithmetic that reconciles them is the subject of this piece: government payrolls fell by 212,000.

Put plainly, the private sector added more jobs than the economy did. That is not a paradox or a revision artefact. It is what happens when one large employer moves in the opposite direction from everyone else, and it is worth noticing because the number people quote, 603,000, is the one that hides it.

Provenance: Source
American payroll employment by sector, August 2026, seasonally adjusted, with the change over the twelve months since August 2025. Sorted by percentage change, the steepest fall first. These are PAYROLL POSITIONS counted at employers, not people: someone holding two jobs is counted twice, and the unemployment rate quoted in the text comes from a different survey of households. The federal row is the only one in this table that falls by more than one per cent.
SectorJobs, August 2026Change over the yearPercent
Federal government2,674,000-242,000-8.30%
Information2,745,000-115,000-4.02%
State government5,470,000-17,000-0.31%
Manufacturing12,638,000+23,000+0.18%
Local government15,179,000+47,000+0.31%
Professional and business services22,527,000+152,000+0.68%
Leisure and hospitality17,001,000+131,000+0.78%
Construction8,359,000+120,000+1.46%
Health care and social assistance18,521,000+378,000+2.09%

It is federal, not government

"Government" is three employers in American statistics, and they did three different things this year. Federal employment fell by 242,000, or 8.3%. State government fell by 17,000, which is 0.31%. Local government, by far the largest of the three at 15,179,000 jobs, ROSE by 47,000, also 0.31%.

So the 212,000 that government subtracted is not a public sector in retreat. It is one layer of it. Local government, which is schools, police, sanitation and county administration, added jobs. State government was flat to three decimal places of its own size. The entire movement, and more than the entire movement, is federal.

The longer view sharpens it. Federal payroll employment peaked at 3,013,000 in October 2024. In August 2026 it stood at 2,674,000. That is 339,000 fewer positions, or 11.25% of the peak, in twenty-two months. More than one federal job in nine.

Nothing else in the table is doing this

The exhibit sorts every major sector by how much it moved over the year, and the shape is worth reading before the federal row. Health care and social assistance added 378,000, the largest gain of any sector and 2.09% of itself. Construction added 120,000, 1.46%. Leisure and hospitality, professional and business services and manufacturing all added, in that order of intensity.

Two sectors fell. Information, which is publishing, broadcasting, telecoms and data services, lost 115,000, or 4.02% - a real decline, and the second steepest here. And federal, at 8.3%, fell more than twice as fast as information and more than twenty-five times as fast as state government.

That is the whole finding. In a year when almost every part of the American labour market grew, one employer removed more positions than every shrinking sector combined, and the headline number absorbed it without showing it.

Provenance: SourceWhat a payroll count counts
  1. The establishment survey asks EMPLOYERS how many people were on the payroll. It counts POSITIONS: a person with two jobs appears twice, and a self-employed person appears not at all. Every sector figure in the table comes from it.
  2. The household survey asks PEOPLE whether they worked or looked for work. It produces the unemployment rate and the participation rate quoted in the text, and it counts each person once.
  3. A position disappearing from a payroll says nothing about what happened to the person who held it. Retirement, resignation, dismissal, a transfer to a contractor and a post left vacant after someone left all reduce the count by one, and this survey does not distinguish them.

So "shed 339,000 jobs" is a statement about payroll positions between two dates, and it is the strongest statement this source supports. It is not a count of people who lost work, it does not say those positions were cut rather than left unfilled, and it names no reason. The unemployment rate held at 4.1% across the same month, from the other survey, and the two are not in contradiction: they measure different things.

Two different surveys produce the two numbers this piece quotes, and they answer different questions. Reading one as a version of the other is the most common way a jobs report is misread, so both are set out here.

What this piece will not tell you

It names no cause. Not one. A payroll series records that a position was on a payroll in one month and not in the next; it does not record why, and no amount of reading the series will produce a reason. Retirements not backfilled, resignations, dismissals, work moved to contractors and posts abolished all look identical here, and they are not the same thing.

It also says nothing about the people. The establishment survey counts positions at employers, not individuals, which is why the piece says "jobs" and not "workers" everywhere. Someone whose federal position ended may be counted the next month in professional services, or in health care, or not at all. This data cannot follow anyone.

And it is not a forecast. Twenty-two months of decline in one sector is a fact about twenty-two months. The August print itself, 162,000, was the strongest month since March; the federal series fell by 5,000 in that same month. Both of those are single months, and a month is a photograph.

What the data establishes

  • Over the twelve months to August 2026 American payrolls grew by 603,000 while private payrolls grew by 815,000, because government payrolls fell by 212,000.
  • The whole of that government fall is federal: federal employment fell 242,000 (8.3%), state government fell 17,000 (0.31%), and local government rose 47,000 (0.31%).
  • Federal payroll employment peaked at 3,013,000 in October 2024 and stood at 2,674,000 in August 2026: 339,000 fewer, or 11.25%, which is more than one job in nine.
  • In the same twelve months health care and social assistance added 378,000 jobs (2.09%) and information lost 115,000 (4.02%); federal fell more than twice as fast, in percentage terms, as any other sector in the table.
  • The August month itself added 162,000 jobs, against 21,000 in July, with unemployment unchanged at 4.1% and participation at 61.6%.

What it does not

  • Any cause, for any figure here. A payroll series records that a position existed in one month and not the next, and never why. Dismissals, retirements not backfilled, resignations and work moved to contractors are indistinguishable in this data.
  • Anything about people. The establishment survey counts positions at employers; one person holding two jobs is counted twice and a self-employed person is not counted at all. Nothing here follows any individual from one month to the next.
  • That the 339,000 positions were cut rather than left unfilled. The series records the level, not the mechanism, and the two produce the same fall.
  • That the private sector absorbed anyone. The piece prints both movements because they happened in the same economy, not because one caused or accommodated the other; this data cannot connect them.
  • Any trend or forecast. August is one month, the strongest since March, and a month is a photograph. Nothing here says what September will print.

Method

  1. One source, one report. Every figure is a US Bureau of Labor Statistics series read through FRED, which serves them as machine-readable CSV; bls.gov refuses automated reads. Reference month August 2026, the latest published, read on 5 September 2026. Series: PAYEMS, USPRIV, USGOVT, CES9091000001 (federal), CES9092000001 (state), CES9093000001 (local), USINFO, CES6562000101, USCONS, USLAH, USPBS, MANEMP, UNRATE, CIVPART. All seasonally adjusted.
  2. TWO SURVEYS, KEPT APART. Every sector figure comes from the establishment survey, which counts PAYROLL POSITIONS reported by employers. The unemployment rate and the participation rate come from the household survey, which counts PEOPLE. The piece never derives one from the other and never calls a position a person.
  3. NO CAUSE IS NAMED, and the restraint is the point on a subject this politically charged. The piece names no administration, no policy and no programme. A payroll series cannot distinguish a dismissal from a retirement not backfilled, and pretending otherwise would be an editorial act the data does not support.
  4. THE PEAK IS MEASURED, NOT ASSUMED. October 2024 is the maximum of the federal series over the period examined, found by searching it rather than by picking a memorable date. The 339,000 and the 11.25% are computed from that maximum and the August 2026 value.
  5. "MORE THAN ONE JOB IN NINE" IS AN UNDERSTATEMENT, DELIBERATELY. The fall is 11.25% of the peak, and one in nine is 11.11%. The piece says "more than one in nine" rather than "one in nine" because the second would be false by a small margin, and a small margin in the wrong direction is still wrong.
  6. Percentages are computed from the unrounded levels and rounded once for display. The health care figure is 378,500 jobs on a base of 18,142,200 and is printed as 378,000 and 2.09%.
  7. THE EDITION THIS REPLACES IS REDIRECTED, NOT DELETED. Its sixteen URLs had already been submitted to IndexNow when it was withdrawn, so they now redirect here rather than returning 404.

BabZituna scores every job against your profile on six real dimensions, and publishes what it can and cannot measure. See the live market.

The markets in this piece, live

Earlier editions