Every first Friday of the month, the US Bureau of Labor Statistics publishes its jobs report. Within seconds, the dollar moves. Bond yields jump or fall. Stock markets lurch. Billions of dollars change hands — all based on a single number that is routinely, sometimes dramatically, revised weeks or months later. This page explains how employment data is collected in both the US and UK, why revisions happen, how large they can be, and what that means for the reliability of markets that move on first-print numbers.
How US Jobs Data Is Collected — The Three-Publication System
The US Bureau of Labor Statistics (BLS) publishes jobs data three times for every single month. Understanding this system is essential to understanding why revisions happen and why they can be so large.
The Birth-Death Model — The Biggest Source of Error
The BLS cannot survey companies that do not yet exist or that have just closed. New businesses do not appear in the survey frame until they register with state employment agencies — a process that takes months. Closed businesses stop responding but are not immediately removed. The BLS attempts to estimate the net employment impact of these business births and deaths using a mathematical model.
This model is based on historical patterns of how many jobs new businesses typically create and how many jobs closing businesses typically destroy, by industry and by season. It works reasonably well in normal economic conditions. It works very poorly at turning points — when the economy is changing direction — precisely when accurate data matters most.
In 2021 and 2022, the birth-death model systematically underestimated the post-pandemic hiring surge. The model expected normal levels of job creation from new businesses; the actual surge far exceeded this. This meant the monthly figures consistently understated job growth — which was only revealed in the annual benchmark revision.
In 2023 and 2024, the pattern reversed. The model continued to add estimated jobs from new businesses at a rate consistent with a strong economy — but the actual economy was slowing. The result was that monthly figures consistently overstated job growth, which the 2025 benchmark revision then removed: 911,000 jobs that had been reported simply were not there.
Major US Jobs Revisions — A Record of Error
The pattern of large revisions is not new. Here are some of the most significant examples, all of which moved markets dramatically on first print and were then revised substantially.
| Period | Original Estimate | After Revision | Change | Direction of Error |
|---|---|---|---|---|
| Mar 2025 (annual benchmark) | Previously reported | −911,000 jobs | −911k | Overstated for full year |
| Mar 2024 (mid-year preliminary) | Previously reported | −818,000 jobs | −818k | Overstated — revised Aug 2024 |
| Apr+May 2026 combined | 179k + 172k | 148k + 129k | −74k total | Both months overstated |
| Jul 2024 | +114,000 | +89,000 | −25k | Overstated; shocked markets |
| 2021 full year | Understated | Revised up | +large | Understated — missed surge |
| 2019 annual benchmark | Previously reported | −505,000 | −505k | Overstated for full year |
Illustrative chart showing the pattern of US monthly revisions 2023-2026. Negative values mean the initial estimate overstated jobs created. Source: BLS, CEIC.
Why Markets Move Billions on Numbers That Are Later Proved Wrong
This seems like an obvious problem. If the jobs number is frequently and substantially revised, why do markets treat the first print as definitive? The answer reveals something important about how financial markets actually work.
Reaction is relative, not absolute. Markets do not move because they believe the exact jobs number is correct. They move because the number is better or worse than what economists expected. The consensus forecast — typically compiled from 60-80 economists' predictions — becomes the benchmark. A figure above the consensus is "strong"; below is "weak." The actual level is secondary. This means that even a number later revised away completely still moves markets if it differs from the forecast.
Positioning drives the reaction. Professional investors position themselves ahead of major data releases based on their own forecasts. When the number comes in, those who positioned correctly profit; those who positioned wrongly rush to adjust. This creates rapid, large price movements in the seconds after publication — entirely regardless of whether the underlying number will survive revision.
The revision is not tradeable. By the time the revised figure arrives — four to five weeks later, buried in the next month's release — the market has already moved on to the next month's first estimate. There is rarely a clean trade to reverse the original move based on the revision.
The Fed responds to first prints too. The Federal Reserve explicitly monitors the jobs data. If the first print shows weakness, it affects market expectations about whether the Fed will cut rates. Those expectations move bond yields and stock prices. Whether or not the figure is later revised, the Fed's reaction function has already been updated.
The UK Has Its Own Data Crisis — The ONS Labour Force Survey
Britain's equivalent of the US nonfarm payrolls is the ONS Labour Force Survey (LFS) — a quarterly survey of approximately 29,000 households across the UK. In recent years it has suffered a reliability crisis that is arguably more severe than the US revision problem, because at certain points the data was so poor it had to be withdrawn entirely.
🇬🇧 UK LFS — The Crisis
- Response rates fell sharply during Covid as face-to-face interviewing stopped
- By July-September 2023, responses hit their lowest point ever recorded
- Detailed LFS estimates suspended entirely October 2023 to January 2024
- Accredited statistics status withdrawn — data rebadged "official statistics in development"
- Sample was 55% smaller than pre-pandemic levels at the low point
- Only 21% response rate in Q1 2025 — meaning 79% of surveyed households did not respond
- ONS itself recommended using multiple data sources rather than trusting the LFS alone
🇺🇸 US BLS — The Issues
- Survey covers 119,000 businesses — large but still a sample
- Birth-death model adds estimated jobs that may not exist
- Annual benchmark revisions can remove hundreds of thousands of reported jobs
- Monthly revisions go unnoticed by markets despite being significant
- Seasonal adjustment models can produce implausible sector results
- Two separate surveys (establishment + household) often diverge significantly
- Secretary of Labor publicly criticised data integrity in 2025
The ONS has been rebuilding the LFS since late 2023. The sample was increased by 55% from January 2024. Response rates have been recovering. But the five-wave structure of the survey means changes take up to 15-18 months to fully feed through — so data collected in early 2024 may still reflect the transition period rather than the true labour market.
UK vs US — Different Problems, Same Core Issue
The US and UK employment data problems are different in nature but share a common root: counting jobs is genuinely hard, and statistical methods that work well in normal conditions break down when the economy or survey participation changes rapidly.
| US Nonfarm Payrolls | UK Labour Force Survey | |
|---|---|---|
| Method | Establishment survey of 119,000 businesses | Household survey of ~29,000 homes |
| Frequency | Monthly | Quarterly |
| Biggest weakness | Birth-death model for new businesses | Falling response rates |
| Revision system | 3 publications per month + annual benchmark | Periodic reweighting, methodology changes |
| Largest recent revision | −911,000 jobs (annual benchmark 2025) | Data suspended entirely 2023-24 |
| Market impact | Moves stocks, bonds, dollar within seconds | Moves sterling and gilts; slower market impact |
| Current status | Published but contested | Published with quality warnings |
| Alternative sources | Household survey, ADP private payrolls | PAYE/RTI data, Workforce Jobs series |
The UK has an advantage the US lacks: PAYE Real Time Information (RTI) data from HMRC. Every time a UK employer pays wages, they report it to HMRC in real time. This administrative data covers virtually all employees and is much more comprehensive than any survey. The ONS uses it alongside the LFS to cross-check employment trends — and has increasingly relied on it during the period of LFS weakness. This kind of administrative data cross-check is something the US is still catching up on.
What This Means For Ordinary People
You might wonder why any of this matters if you are not a professional investor. The answer is that employment data errors have real consequences that flow through to everyone.
Interest rate decisions. The Bank of England and the Federal Reserve both use employment data to calibrate interest rate decisions. If the data shows a strong labour market when the actual labour market is weakening, rates stay higher for longer than necessary — which means higher mortgage costs, more expensive business loans, and slower economic growth. Every UK homeowner with a tracker or fixed-rate mortgage coming up for renewal is affected by rate decisions made partly on flawed employment data.
Government spending decisions. UK government spending on welfare, housing benefit and tax credits is partly calibrated against employment levels. If official employment data overstates how many people are in work, public spending may be insufficient to support those who have actually lost jobs.
Market volatility. When markets move violently on a jobs figure that is later revised, this creates genuine economic harm. Pension funds that rebalanced portfolios, businesses that cancelled investment plans, and households that delayed major purchases all respond to the news environment shaped by the first-print number.
US Jobs: First Print vs Final Revised Figure (2023-2026)
Illustrative comparison based on BLS published data. Blue bars show initial first-print estimates. Red markers show final revised figures where available. The systematic gap shows the first print was repeatedly higher than reality from mid-2023 onwards. Source: BLS Employment Situation releases.