Based on employers’ quarterly payroll filings. Why wait so long for revisions? Do it quarterly!
By Wolf Richter for WOLF STREET.
Every month, the Bureau of Labor Statistics asks 10s of thousands of employment locations about their payrolls as part of its Current Employment Statistics (CES). It then uses the data, plus a model for estimating the employment effects of firms having shut down (“deaths”) and firms having been created (“births”) during the period, to estimate the nonfarm payrolls for that month, whose month-to-month changes become market fodder. These preliminary data are then revised several times in the following months as more data become available, sometimes with stunning results.
Separately, all employers file payroll tax reports with their state agencies that become part of the unemployment insurance tax records. This is administrative data, not survey data. These detailed payroll data then become the primary input for the BLS’s Quarterly Census of Employment and Wages (QCEW).
But given the delay in these filings and the quarterly nature of the QCEW, these data are useless for monthly reporting and would not satisfy our hunger for monthly data – which is why the survey-based estimates from the CES are released monthly.
The BLS then adjusts (“benchmarks”) the monthly survey estimates of the CES to the actual payroll data in the quarterly QCEW for the 12 months through March.
Today, the BLS released the preliminary benchmark payroll revisions for the total 12-month period through March 2025. Then in February 2026, it will release the final benchmark revisions spread across the months.
The BLS today downwardly revised by 911,000 jobs, or by 0.6%, the number of nonfarm jobs in the period through March 2025.
Private sector jobs were revised down by 880,000 jobs; government jobs were revised down by 31,000.
Based on the monthly survey data, the BLS estimated job growth at 2.35 million jobs over the 12-month period through March, or an average of 196,000 per month.
Today’s revision lowered the pace of job growth to 1.44 million jobs over the 12-month period through March, or an average of 120,000 per month.
Where did this massive overestimation by the monthly survey data come from? The BLS said that “preliminary research” indicates two primary contributors:
- Response error: Businesses reported less employment to the QCEW than they reported to the CES survey.
- Nonresponse error: Businesses who were selected for the CES survey but did not respond reported less employment to the QCEW than those businesses who did respond to the CES survey.
And it said that “estimates of other errors, such as the forecast error from the net birth-death model [companies being created and shutting down], are not available at this time.”
This chart shows the preliminary benchmark revision of nonfarm payroll job growth, released yearly at this time of the year, for each March:
2 of the major categories were revised up:
| Transportation & warehousing | 6,600 | 0.1% |
| Utilities | 3,700 | 0.6% |
12 of the major categories were revised down:
| Leisure and hospitality | -176,000 | -1.1% |
| Professional and business services | -158,000 | -0.7% |
| Retail trade | -126,200 | -0.8% |
| Wholesale trade | -110,300 | -1.8% |
| Manufacturing | -95,000 | -0.8% |
| Information | -67,000 | -2.3% |
| Other services | -51,000 | -0.9% |
| Financial activities | -39,000 | -0.4% |
| Private education and health services | -35,000 | -0.1% |
| Government | -31,000 | -0.1% |
| Construction | -29,000 | -0.4% |
| Mining and logging | -4,000 | -0.7% |
The establishment survey data in the monthly jobs report will not be updated with these preliminary revisions. Instead, it will be updated with the final revisions that include monthly details, to be released in February 2026. We will cover that as we have in prior years with a chart with two lines, one color for the new revised data and another color for the old data.
Final revisions have differed significantly from preliminary estimates in the past. These preliminary revisions today don’t include a number of categories, such as the unincorporated self-employed, but they’re included in the final revision. So the final revision is likely less bad. The final revision released in February 2025 was substantially less bad than the preliminary benchmark revision of -818,000 at this time last year.
Why wait so long for benchmarking?
One of the immediate things to do for the new BLS commissioner would be to shift benchmarking to a quarterly basis. This would produce four preliminary benchmark revisions a year, with final revisions when available. And at least four times a year, we’d have more accurate nonfarm payroll data.
Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how:
E J Antoni, the new head of BLS, will straighten this mess out.
I am pretty sure Antoni will be better than former Commissioner McEnfarter. I would have preferred Wolf Richter to Antoni. A problem is there apparently is no reason for businesses to comply with the survey, nor is there any reason to think the numbers they submit are close to reality. Furthermore, instead of putting out a monthly preliminary print, just wait a couple of months when the data might be more reliable. Nobody is going to believe the preliminary monthly data any more.
As someone who receives those surveys for every business I have worked at, I can assure you there is no penalty for failing to comply. There are threats but never any teeth behind it – like most of our laws these days.
Those sit as the lowest priority item in my stack. Sometimes, after going to lunch and noticing 20% of the cars in the parking lot have expired tabs, I then return to my desk and toss them.
This small government libertarian would be angry if there were teeth from those surveys. Big government is growing not shrinking, intruding in our private areas. Excited to read Wolf interpretations of today’s PPI data.
thurd2,
You make an interesting point: “A problem is there apparently is no reason for businesses to comply with the survey, . . ”
While I can understand that business owners and employers have to follow IRS and state reporting for tax purposes. That being said, why the hell would someone running their business want to take the time and energy to answer to the BLS when sent a questionnaire?
Just asking . . .
So you have to distinguish between
1. surveys that people/companies may or may not reply to
2. payroll tax reports and UI reports that companies HAVE to file; and if they don’t file them, they can get into deep trouble, and owners can end up in jail.
In terms of responding to government surveys: I’m one of the guilty ones. I, as CEO of the Wolf Street Media Mogul Empire, a massive C corporation, ignored the last two surveys I got, despite increasingly urgent emails to comply (I don’t remember what they were for). I’m just tired of them. Why do I have to spend my time on them? Send them to Cookdoggie and let them deal with it.
The big errors in the nonfarm payroll data were from surveys not being responded to. I listed both of those types of errors in the article.
If you have business in the U.S with employees on payroll, there are quarterly (at least) tax forms due to the Feds and State for tax withholding, and Department of Labor forms due to the Feds and State for unemployment insurance. All of these forms specifically identify the number of employees on payroll for the reporting period, and their compensation (to the penny)
So, when the survey forms arrive, its a redundant exercise, when accurate data has already been provided to various Fed and State organizations
As a small business owner (max 2 employees at a time) I have also had to fill out these surveys and they are overly cumbersome with many non-sensical items for a small business. Even worse, when you fill out the next one, there is no way to pre-fill from the previous filing to then apply just differences, but you have to waste your time to go through each individual non-sensical question again.
So, like Wolf, I also gave up on responding to the surveys, and now haven’t received a survey for the last two years. I hope the authorities have just given up on me, rather than cooking plans in their drawers for my arrest due to non-compliance.
As far as the mandatory data filings goes: Apart from the ongoing IRS filings there is also various data collected during hiring: W-4s that also go to the IRS, I-9s that have to be filed with the DHS, and state unemployment insurance registration. One would think that out of all these mandatory filings, something meaningful and current could be derived, that should be better than the surveys.
Unfortunately that would require the government to get their act together.
Antoni is a hack with no significant administrative experience, and no particular insight into economics.
“[Antoni’s] track record is so alarming that many conservative economists” have denounced the nomination, including Kyle Pomerleau from the American Enterprise Institute, AEI colleague Stan Veuger, and Dave Herbert of the American Institute for Economic Research.”
Steady on lad ! Next you’ll be saying someone else should be Head of Health!
When an anti-vaccine guy goes shopping for a dog, does he ask if it’s had its shots? If not, it sure as hell ain’t going into American Kennel Club events.
The American Kennel Club has higher standards than the CDC under Cheryl Hines’s husband.
Here’s a list of conservatives who think he is an inept fool:
http://www.axios.com/2025/08/12/trump-bls-ej-antoni-economists
When these appointments are done on the basis of political opinion rather than ability (DEI anyone?) there is very little hope of their being a success. What a fool anyone is who believes otherwise just because Chairman Trump appointed them.
The Plum Book lists about 7000 Federal positions that are subject to some level of political selection/confirmation (because they clearly/directly deal with policy decisions).
Any reform movement (and god knows this $35+ trillion debt nation needs one) faces the challenge of finding 7000+ outsiders-to-the-Establishment to take over those positions from opposing political operatives/Establishment lifers if thoroughgoing reform is going to have a chance.
Your defense is that it’s hard to find competent people who want to do this stuff so it’s ok to pick incompetent people? That says a lot!
Wolf as someone who moved on from CS for your housing data series due to better/more timely alternatives being available, I am sure you have some thoughts on how better data could be collected or distributed moving beyond the BLS altogether?
I am concerned that the BLS is fundamentally not fit for purpose having being passed by the times and failing to adapt, e.g., INTC etc. for comparison.
Provide companies with some bureaucractic/IRS related reward for timely *and* accurate data filling at each month end. Release on first Monday of next month. I am clearly missing some important detail as never having been a business owner, but why is this so hard?
In terms of legit jobs (not under-the-table jobs), government payroll tax data, including unemployment insurance data, reported by companies to the government, are very accurate. They cover every employee in the US. They also cover the payroll-tax-paying self-employed. Only the government has this universe of payroll data.
The problem is timeliness. Most of this would be based on quarterly filings by companies with filing deadlines sometime after the quarter closes, so it doesn’t produce the reports we like on the first Friday after the month closes.
To come up with timely data, surveys are used — and they’re all more or less inaccurate and subject to big revisions, if they’re honest. ADP data, which extrapolates from its own client base to the overall country via a model, is also way off and subject to heavy revisions. Internet data is THE worst.
The IRS quarterly tax receipts would be a good indicator of the economy and employment.
They’re among the most misleading indicators of the economy and employment. You can just look at them. They’re all over place. They include estimated taxes and capital gains taxes and now tariffs. I posted a chart the other day:
https://wolfstreet.com/2025/08/28/us-governments-fiscal-mess-interest-payments-on-the-treasury-debt-interest-rates-tax-receipts-and-debt-to-gdp-ratio-q2-2025/
I concur. This is not 1950, it’s 2025. There has got to be a better way to do this.
Nearly everyone on a payroll is in a computer system that sends data (and money) to the IRS every paycheck: tax withholding. So the Federal Government already has an enormous stream of employment data.
That data contains far more accurate info on job gains and losses than what the BLS has. Given the importance of accurate econometrics, it wouldn’t be too hard to add a little more detail where needed.
Companies should only need to send one payroll report to the government every pay period, not tons of separate surveys.
There’s also room to replace the Household Survey, which tracks employment from the other side…
“Nearly everyone on a payroll is in a computer system that sends data (and money) to the IRS every paycheck: tax withholding.”
The reporting with the necessary detail is done quarterly one month after the quarter ends (Form 940), and annually (Form 941), as I pointed out. Ask me how I know this, LOL
Not only is it only done quarterly, it is fairly simple for companies to get extensions. This means the reporting might be 6 months late or more.
Yes, which is another reason by the final revisions are released in February the next year.
Gotta love armchair experts.
“I, who have no expertise in this particular area and have never bothered to learn anything about it, am absolutely sure that there must be a better way of doing things”
“Like what?”
*some random idea that is either fundamentally flawed, or actually describes how things are already done*
You realize that this is a logical fallacy on your part, right?
Indeed I am guilty of being too general. Of course some people have good ideas (Wolf’s in this post is an example of an idea that is both new and reasonable). I find that those who have new and reasonable ideas are both more knowledgeable about the subject area and more humble about the difficulties of doing it well.
30Y yield heading to 4.4% may even touch 4.3% if inflation is favorable. However, I do see a quick rebound to 4.5-4.55 range due to high deficits
It looks like it’s around 4.72% today. Where are you seeing heading to 4.4%?
Over 6 months
Hi Wolf,
At risk of incurring your wrath, I suggest that the calculation
‘ by 911,000 or by 0.6% ‘
is incorrect.
All good wishes.!
From 157.52 million nonfarm jobs to 156.61 million nonfarm jobs
Wolf,
New unemployment weekly claims are ranging in 230K range for a long time. This job market is going to get very confusing (if not already happened) from data standpoint and its impact on economy and I believe primary reason is impact of ICE on curbing illegal immigration. When Hyundai loses 700 people in a day (which is similar to letting go 700 people) but now they have to hire 700 legal immigrants to keep their factory running. Do you think we should just keep aside job creation and unemployment report on sideline for a while until we get immigration under control. it could be another 3 years of current administration and we may be still half way into our ICE operation.
why would hunday “hire” “legal immigrants”? They employ people in Korea, give them a visa for the country to build a factory in a country they are temporary send to. They might need to change the visa type but they are not going to hire someone in the US to send them to train them in Korea for 6 months, then have them build and dissasemble the machinery they then reassemble in the US thats not how any of that works.
I like the idea of more timely revisions. Seems like an easy step that can be implemented almost immediately.
This (to me) begs the question:
Why not just wait and release Quarterly numbers on employment? Seems to me the main beneficiaries of the current system are Wall St. firms.
That would be one way of doing it, and it would cause a huge ruckus. The other way, which would likely cause less of a ruckus, would be to adjust the survey data quarterly, instead of annually. And then also figure out a way to reduce these two big apparently systematic errors in the survey data: the response error, and the nonresponse error. And there are probably many other ways to make the survey data more accurate. But at least, adjust it quarterly to the actual payroll data.
I recently got a Nielsen survey in the US Mail asking about my TV viewing habits. I threw it in the trash because I basically don’t watch TV. The following week I got another letter which I set aside unopened. Then I got a third letter. I then opened those last two letters and each had a crisp one dollar bill. That made me feel guilty as I put the money in my wallet, so I filled out the survey. And mailed it. I refused to go to their website. At the end of the survey they promised you $5 if you give them an email address. I passed on that opportunity since I hate email spam.
Anyway, my point being that the BLS could pay (even nominally) for survey responses, measure accuracy of responses, and dump out from future surveys the junk responses. That would get more compliance and more serious responses perhaps.
BTW, I run a small business LLC and have only received one federal job survey request in the last 7 years.
“and have only received one federal job survey request in the last 7 years”
Yes, that’s why I stopped responding. I didn’t get any for maybe 10 years, and suddenly I got a flow of them. One of them is for a “panel” that I was anointed to be on, and they send me the same survey regularly. I’m just tired of them, got other things to do.
I used to think the BLS existed to protect and inform the working public. I tend to lean in your direction that the main beneficiaries are wall street insiders now.
QCEW reports employment and wages, but not self employed. If working hours are down 2h/ week the spread between Nonfarm and QCEW will rise. QCEW isn’t more accurate. It distorts employment.
The unincorporated self-employed and some other categories are picked up in the final revisions in February 2026, which is why the final revisions were less bad than these preliminary revisions, as I said. In February 2025, the final revision was -655,000, so less bad than the preliminary revision of -818,000.
The under-the-table paid-cash workers are not picked up at all in the payroll data (establishment data), but are part of total employment in the household survey.
How much of what is happening in the job market has to do with bleeding the system of money supply and killing demand? This is rhetorical question, but also wanting to know? So many working parts ATM: killing off “illegal labor” supply, balance sheet action, fed funds rate…to name but a afew.
I used to work for a sizeable trade association, representing small mom & pop $5-$10 million firms, up to $100’s of billions multinationals. The data and membership guys tried to run surveys.
All of the companies lied. Even if economic data and trends were terrible, salvation was ALWAYS just one quarter away.
Where does farm labor fit in. Is it all contract labor reported on 1099s?
All labor, including farm labor, is included in the Household survey, total employment. It was adjusted UP by over 2 million last February.
Incorporated contract labor/self employed are included here in the nonfarm data. Non-incorporated self-employed are included in the finial revision of the nonfarm data (to be released next February).
Dow, S&P and Nasdaq book new record highs despite downward revision to jobs data
The interesting thing I’ve noticed is that no one, even the relatively unsophisticated in finance, seems to even think that the equity markets have anything to do with the economy anymore. It wasn’t that long ago that people did.
BLS has a birth/death model for jobs, but it is remarkable how BLS always talks about job *creation* but never job *destruction*. It’s like the BLS mainly exists to give politicians some numbers to brag about.
I found “total wages paid” and “labor participation rate” (from FRED) to be interesting measures, as well as “All Employees, Total Nonfarm”. Not at all sure how timely and accurate those numbers are, though.
So this is net job growth here: all the nonfarm payroll jobs now minus all the nonfarm payroll jobs before.
Yes, the death part is from companies shutting down and laying everyone off. There are several hundred thousand companies that do that every month. Most of them are 1-employee outfits. Maybe the owner retires or starts something new, whatever. There are also lots of small restaurants in that group. But some are bigger. And since no one there responds to surveys, the data cannot be captured by surveys, and it has to be estimated. But the payroll data will show it eventually.
There are also about 400,000-500,000 companies being created every month, most of them 1-employee companies, but bigger ones too. The Census Bureau tracks this via the IRS EIN applications. I write about it occasionally.
Startups are huge job creators because there are so many startups. If you start a company, and you’re the only employee, you created 1 job. If 300,000 people do that in one month, they created 300,000 new jobs that month. For many people, it’s their dream, to be their own boss. It’s an essential part of the dynamism of the US economy. Many big old companies don’t create any new jobs. They may be shedding jobs.
All the data you mention are from the BLS or the BEA. FRED just makes them easy to find and see.
“Total wages paid” are part of the BEA national account series, on which GDP is based. This is quarterly data.
“Labor participation rate” is from the BLS household survey of the monthly jobs report. It’s unrelated to the nonfarm data here, which is based on employer data.
Look at the “prime-age labor participation rate.” It’s very high. It excludes the impact of the large number of boomer retirements over the past 15 years.
“All employees nonfarm” is the data here that was revised today by -911,000. But as I said, the revision will not be applied to the monthly numbers. Instead, it’s the final revision, to be released in February, that will be applied to the monthly numbers going back to the beginning of 2024.
Thanks, Wolf. Good stuff.
If we assume the same rate of overestimation from March 2023 to March 2025 (-600K to -900k) has continued, then every jobs report from March 2025 could almost certainly be 50k to 75k too optimistic.
That would mean we could easily have already had 4 consecutive months of negative job creation, with July being the only possibility of remaining in positive territory.
Yikes.
Great summary, as always.
And yes, 0.9M on 156.6M is a 0.6% change in total non-farm employment, i.e. not a massive change in the overall employment picture.
But 0.9M on 2.35M in 12 month job growth is huge, nearly a 40% reduction in 12 month job growth. Nine or ten of those were Biden months (not that it matters, just pointing out that it’s non-partisan, unlike myself) and all of it took place before tariff focused magnifying glasses were even invented.
We can (and should) complain about data collection protocols, revisions of this type are simply unacceptable. But to me the key point here is that we live in a country where monthly job gains are routinely revised down by ~70k/month (see Wolf’s report on same subject around this time last year). That’s one thing when the numbers are dropping from 242k/mo to 171k/mo (12 months ending Mar-24) or from 196k/mo to 120k/mo (12 months ending Mar-25). But right now we’re sitting on a 3 mo moving avg of 29k/mo. We don’t have 70k/mo to lose.
In the meantime keep your eye on continuing claims. That’s a weekly series that is “continuously” revised and therefore immune to most of this hand wringing. It also has a fair track record as a recession indicator (credit to our host), and we’re a long way from the trigger of 2.6-2.7MM.
Beyond that, I can’t begin to imagine how our leaders could engineer a recession in the face of $2T in deficit spending. That, my friends, would cement a lasting legacy for the class of 2025.
Wolf – Do you still believe we’re not already in a recession (or stagflation)?
You need to have widespread DECLINES of the economy and in the labor market. The economy is growing, and the labor market is still growing, but more slowly.
A recession is not slower growth; it’s widespread DECLINE.
Wolf, I share your assertion that nothing goes to heck in a straight line.
Our economy would not go directly from meteoric growth to a steep Decline. There would be a deceleration first.
Is there enough evidence to say our economic/labor markets are decelerating? (Growing, yes, but growing ever slower)
Defined as two consecutive quarters of negative growth GDP.
I kept telling you the jobs were gone.
They’re not “gone.” They grew less than previously estimated.
A federal appeals court has blocked the firing of Lisa Cook from the Federal Reserve Board of Governors and she will be one of the 12 voting members of the FOMC when it meets next Wednesday.
So what? Let’s not confuse the issue. If, in her private life, she committed mortgage fraud (and listing multiple properties as you “primary residence” on multiple mortgages is indeed fraud), then she should be prosecuted accordingly. Has nothing at all to do with her performance at the Federal reserve. Other governors have stepped down when insider trading came to light. Personally, I’d rather people are actually prosecuted for breaking the law and go to prison. Let’s be honest, at the core of why America is in the mess it is in is the fact that the system has been rewarding bad behavior for 50+ years. That shit needs to change, and it will, either through a rational intelligent mechanism or by guillotine. History is very clear on this point.
Interesting times.
One of the issues raised in court is that the allegations refer to acts BEFORE Cook was appointed to the Fed.
The insider trading occurred WHILE these people were at the Fed and had advance knowledge of what the Fed would do in the future, and they traded on that advance knowledge.
That is not to say that the Cook allegations should not be pursued in court, along with the Paxton allegations, and all the others. But that’s not the issue here.
The issue is “for cause” in a termination case. They’re saying that “for cause” means you committed a wrong on the job that was worthy of getting fired. And she didn’t do that.
That definition of “for cause” is one of the big things to be decided by SCOTUS in this case.
Well, I think the “for cause” will be ruled as applied to being on the job.
Regardless, my point still stands. Well past time bad behavior was punished, not rewarded. It will happen, one way or another.
Good! Apparently this practice is incredibly common, and many, many members of both parties would get swept up, including the parents of Bill Pulte, the guy who brought these accusations in the first place!
Correct me if I am wrong, but it appears to me that neither this latest revision nor anything else changes the apparent fact that the labor market is actually quite healthy in that the employment-population ratio for 25-54 yrs is near an all-time high. As noted in Wolf’s comment under his September 4 article.