Tweet
Email
Link
Large downward revisions to monthly jobs data in recent weeks put a gigantic spotlight on an incredibly typical procedure for economic data.
They also spurred something highly atypical, with President Donald Trump firing the Bureau of Labor Statistics commissioner over baseless allegations that the data was “rigged.”
Now, there’s an even bigger revision coming: One that could imply there were somewhere between 475,000 and 900,000 fewer jobs added between April 2024 and through March 2025.
Some potential factors driving the expected downward revision include weaker-than-inferred job creation at new firms; sampling errors from declining response rates; and, to some extent, adjusting for undocumented workers.
On Tuesday, the BLS will provide preliminary data for its annual effort of gaining a near-complete employment count by squaring past jobs data from business surveys (more timely but not as accurate) with comprehensive unemployment insurance quarterly tax filings (highly accurate but significantly lagged in timing).
Tuesday’s release marks the first step in an annual review called benchmarking, a previously innocuous process that’s been applied to BLS jobs data in some shape or form for 90 years.
“It is not a bug, it’s a feature; it makes the [Current Employment Statistics] more accurate,” said Erica Groshen, a former BLS commissioner who now serves as senior economics adviser at the Cornell University School of Industrial and Labor Relations. “And that happens every year.”
But in an age when economic data is frequently weaponized and now more recently has been itself put in the crosshairs, this BLS standard procedure feels anything but standard.
This time last year, the preliminary estimate inferred that there were likely 818,000 fewer jobs added in the 12-month period through March 2024. The initial not seasonally adjusted estimate — if it held firm come February when the final benchmark revision was released — would have been one of the largest in recent history.
After the preliminary data was released, then-candidate Trump told a fabricated story about how the government had been planning to announce this downward revision “after November 5th,” Election Day, but was forced to do so before the election because of “a whistleblower”— “a patriot leaker.” He later claimed the job gains under former President Joe Biden’s administration were “a fraud.”
(There was no whistleblower. The Bureau of Labor Statistics regularly releases the preliminary benchmark data this time of year, and it had disclosed the precise date of this particular data release — August 21 — weeks in advance.)
The final annual benchmarked seasonally adjusted data for the 12 months ending in March 2024 was a negative 589,000, or 0.4% of overall employment (-598,000 not adjusted for seasonality), which was the largest downward revision since March 2009’s negative revision of 902,000 and just above the negative 514,000 revision in March 2020 (for the year ended March 2019) that occurred during Trump’s first term.
These large swings happen in significant economic transitions (the Great Recession then and the pandemic now). In addition to pandemic-era economic shifts that knocked tried-and-true models out of whack, the recent surge in immigration likely played a role, economists have said.
While last year’s revision was larger than expected, it basically knocked down average monthly job gains by about 50,000 per month. Despite that, the labor market was still running strong during that April 2023-March 2024 period, economists told CNN last year.
If Tuesday’s revision is on the high end of the estimate, at 900,000 jobs, that would break down to a downward revision of 75,000 jobs per month.
While Tuesday’s data won’t shift the current labor market outlook, a large revision could carry deeper implications, Stephanie Roth, chief economist at Wolfe Research, told CNN.
“This just adds to the narrative of concerns around the labor market and questions about the reliability of the data. To the extent this data does feed into that narrative, we’re worried about the reaction and headlines associated with it.”
Federal data is fluid and frequently subject to change as more detailed and accurate information becomes readily available. The BLS’ monthly jobs report is meant to provide a higher-frequency look at employment trends, but that timeliness comes with a cost to accuracy.
To get the monthly payroll estimates, the BLS surveys about 120,000 US employers, accounting for 600,000 work sites (roughly one-third of employment) and is based upon survey responses from employers across a wide swath of industries. Those respondents are given three opportunities to report their payroll gains and losses for any given month.
That’s why the initial payroll estimates (drawn from a survey when roughly 60% have responded) are revised twice further (when north of 90% have).
Every year, the BLS conducts a revision to the data from its monthly survey of businesses’ payrolls, then benchmarks the March employment level to those measured by the Quarterly Census of Employment and Wages program.
The QCEW provides a more comprehensive read on the number of businesses, employees and wages at the state, regional and county level because it derives that data from quarterly tax reports submitted by businesses to their states. Given that process, the QCEW comes with a significant lag: The data for the first quarter of this year also will be released Tuesday.
Tuesday’s preliminary benchmark revision won’t change the existing monthly employment data for now, but they can provide a sense as to how the labor market was faring last year and heading into this year.
“Yet, with the benchmark revision likely to show a weaker pace of job growth through March, a loss of momentum at the beginning of the year would cast a shadow on the true strength of payroll growth since then,” Wells Fargo economists wrote in a note last week about the benchmark revision. “In short, even as the labor market is still standing, its footing is becoming more tenuous.”
Probable causes of a shifting job landscape
The expected downward revision for this 12-month period likely stems from three primary factors, economists say:
First, a “hefty chunk” — perhaps half — is likely attributable to the BLS modeling of business creation (called the birth-death model) overestimating the jobs added at new businesses, Pantheon Macroeconomics economists wrote in a recent note.
Groshen, the former BLS commissioner, told CNN that this potential factor is a hangover from the pandemic era.
“We have had a spurt in business formations since the end of the pandemic,” she said. “If this unusual spurt in business formations is populated not by the same kinds of firms that were being created before, but by firms that are in some ways more marginal, smaller, more likely to die after formation, then the birth-death model might be overestimating the jobs created by the number of formations.”
Low survey response rates also could be making for more challenging estimations of the monthly data, Wells Fargo economists wrote last week.
“The establishment survey’s overall response rate averaged 43% in the 12 months through March 2025, down from 59% in 2019,” they wrote. “The falling response rate leaves greater room for firms who are responding to be systematically different from the non-respondents.”
And the third primary factor is likely tied to asylum-seekers and undocumented workers. The QCEW data is drawn from tax filings for unemployment insurance, and those likely exclude undocumented workers who are not eligible for benefits.
“And so, when we benchmark those people are going to disappear,” Ron Hetrick, a former BLS employee and analyst who now is the labor economist at Lightcast, told CNN.
However, given the slowing of immigration that started in the middle of last year and accelerated this year, the revision might better truly reflect the labor market, he added.
“Last year, I didn’t like it because I felt that we were downward revising a number that probably wasn’t a true reflection of how many people were working, because, whether you like it or not, they’re on payrolls, I want to know how many people are working,” Hetrick said. “It’s probably going to get more accurate, because now we know these people probably are off of payrolls, so we probably know these numbers are a little too high.”