US Payrolls Revised Down: A Statistical Error or a Weaker Labour Market?

Payrolls revised down by 79,000

For the 12 months from April 2025 to March 2026, nonfarm employment was revised down by 79,000 (-0.1%) from the initial estimate. Private employment was revised down by 178,000, while government employment was revised up by 99,000. The two adjustments offset each other to produce the total revision. Before the release, the median forecast among 20 economists had been an upward revision of 183,000, leaving a gap of about 260,000 between expectations and the result.

The 79,000 adjustment changes the historical data for that year, not the latest employment reading. At the same time, Fed Chair Warsh was making his Jackson Hole debut since taking office. The two events coincided.

Why are annual payroll figures revised down?

Monthly payroll estimates come from the BLS establishment survey. A sample of employers is used to estimate employment nationwide. The results arrive quickly, but sampling error is inherent in the process.

Each year, the BLS benchmarks the estimates against QCEW records based on unemployment-insurance tax filings. These cover almost all employers and provide a more accurate count, but are released with a lag. Put simply, monthly payrolls are estimates; the annual benchmark uses more complete records to correct those estimates.

Over the past decade, the average absolute annual revision has been 0.2% of total nonfarm employment, compared with 0.1% this time. Another common misunderstanding concerns unemployment: this revision affects only establishment-survey data. The unemployment rate comes from a separate household survey and is not affected.

Did the BLS calculate the data incorrectly?

Last year's preliminary benchmark revision lowered employment by 911,000, the largest downward revision on record. The BLS also notes that sampling errors become larger as data are broken into finer categories. This time, employment in the information sector was revised by +3.0%, while the revision to total nonfarm employment was only -0.1%.

Sampling error is part of the statistical system's design, rather than an accident. Revisions of around 0.2% occur regularly and should not be equated with a calculation error.

Does the revision show that US employment has weakened?

The revision can support the view that employment was weaker over the period concerned, but it cannot establish the current state of the labour market. After the adjustment, average monthly employment gains over that year were only about 11,000 on a not-seasonally-adjusted basis.

But the July payroll decline of 23,000, the combined downward revision of 103,000 to May and June, and the labour-force participation rate of 61.4% all concern developments after March. They fall outside the period covered by this benchmark revision.

The composition also matters: the total downward revision of 79,000 combines a 178,000 reduction in private employment with a 99,000 increase in government employment. The private-sector adjustment alone was larger.

There is no need to choose between the conclusions that the statistics were wrong and that employment has already deteriorated.

One point deserves attention. Last year's preliminary revision was 911,000, while this year's narrowed to 79,000. That means the gap between sample-based estimates and administrative tax records has narrowed; it does not mean the economy itself has improved. Benchmark revisions reset the starting point of the data, not the destination of the economy.

What should we watch next?

September 4: the August payroll report will provide a clearer picture of US employment after March.