I & I Editorial BoardYesterday, President Donald Trump nominated a high-profile critic of the Bureau of Labor Statistics, EJ Antoni, to be its new commissioner, promising on Truth Social that under his leadership “the Numbers released are HONEST and ACCURATE.”
If confirmed, Antoni, currently chief economist at the Heritage Foundation, has his work cut out for him.
Last week, we pointed out glaring mistakes the Bureau of Labor Statistics had made in recent years and asked whether its foul-ups were driven by politics or ineptitude.
The issue came to a head after President Donald Trump fired the head of the BLS in the wake of its massive revisions to the two previous months jobs reports, in which it admitted to overcounting the number of jobs created by more than 250,000. Trump responded by saying the numbers were rigged.
The press, naturally, rushed to defend the bureaucrats, saying that revisions are normal and there’s nothing to get worked up about. The BLS routinely revises its initial monthly jobs estimate – based on a survey of 100,000 employers – over the next two months as more data come in. No biggie.
Is that true? We decided to find out and reviewed the BLS’s monthly jobs data going back to 2009. What we found was deeply troubling.
We compared its initial report on jobs each month to the actual figures released months later. Our conclusion: its initial jobs reports have been absolutely and completely unreliable for years – often wildly optimistic or wildly pessimistic – regardless of who was president. (See the table below.)
In the 199 months we examined, the BLS’s initial estimate of jobs gained or lost missed the mark by an average of 49.6%!
Only 15 times did its initial estimate come within 3% of being right. (Given the huge sample size, you’d think its margin of error would consistently be tiny.)
Some months, the miss was staggering.
In August 2011, to cite one example, the BLS said no new jobs had been created. Zero. Turns out, 132,000 were created that month.
In September 2017, it first said that the economy lost 33,000 jobs – which made big news because, as Politico put it at the time, it was “the first time in seven years” that had happened. In fact, the economy had created 88,000 jobs that month.
In January 2021 – the last month of Trump’s first term – the BLS initially reported that the economy had added a mere 49,000 jobs. The actual number was 365,000.
In March 2023, the BLS claimed the economy created 236,000 jobs. The actual number was 85,000. (The Committee to Unleash Prosperity, where Antoni also serves as a senior research fellow, found that the BLS exaggerated the number of jobs created during the Biden administration by a staggering 1.5 million.)
Reuters reports that the BLS has become less reliable, with its second monthly revisions growing larger in recent months. The New York Times claims that one problem is that the percentage of businesses responding to its survey has dropped sharply since 2020, going from around 60% to just over 40%. Of course, Trump routinely gets blamed because he cut BLS jobs.
But there must be more to this story – because it gets worse.
Last August – when Biden was president – the BLS said that it had overcounted new jobs by an eye-popping 818,000 between March 2023 and March 2024. That’s a massive mistake that should have caused heads to roll.
But then six months later, it issued a revision to that revision, saying that it had only overcounted the number of jobs created in those 12 months by 598,000. That a difference of 27%.
No one at the BLS lost their jobs over that screw-up, either.
In the private sector, an organization that was as consistently and wildly wrong as the BLS would have been out of business long ago. In Washington, a catastrophic failure like this usually results in a bigger budget.
At least until now. We hope Congress confirms Antoni and that he cleans up the place. Trump is right that we deserve honest and accurate jobs numbers.
— Written by the I&I Editorial Board
shadow stats told us these things years ago. no one believed him.
The editorial fails to address one of the biggest pitfalls of BLS and other government statistical reporting, namely “seasonal adjustments.”
The editorial’s various numbers are the “seasonally adjusted” figures BLS has published. The raw (unadjusted) figures, aka the actual number of jobs gained/lost (i.e., BLS’s estimate of what ACTUALLY happened), while disclosed, are rarely reported.
It’s as if the government and policy wonks believe that the public isn’t mature enough to handle the real data, which of course fluctuates from month to month (e.g., more hiring during Christmas season, big post-holiday reductions in January, etc.). Arguments that the raw info isn’t useful are bogus; all one has to do is compare a given month to the same month a year ago, 2 years ago, etc., to have an idea of whether there has been improvement or deterioration.
Instead, each month is converted to a “seasonalized” figure which somehow puts things “in context,” and is then treated as if it’s the Gospel truth (which it isn’t).
The raw numbers deserve at least equal time with the seasonalized figures.
The survey response-rate problem also needs to be addressed. It’s ridiculous that it’s so low, given the relative ease of online surveys vs. pen/paper and mailing costs that had to be incurred years ago.
I would argue that the lower response rate indicates that companies are deciding whether or not to respond (or respond quickly enough for the first pass) depending on whether they think their news is “bad” or “good” for “their” party.
Send in DOGE and request the payroll companies on how many jobs were added this month and how many terminations. I did that other BLM job report is was all a guess.
Actually, ADP publishes a jobs report that is remarkably accurate (it’s first iteration ends up close to where BLS’s third iteration lands). US could use ADP numbers for the initial and let BLS only publish its third revision number.
Errors this large means the BLS has no real purpose. Pick a homeless person on the street, ask them to make a wild guess…same probability…zero cost to taxpayers.
It would be interesting to add a little additional information. Did the overcount or undercount benefit or hurt the sitting president and was that POTUS a Democrat or Republican. My bet is that errors benefited the Democrats more than the Republicans.
Note to the adults…
Only liberals agreed with these numbers.
And the fact that the person in charge was a liberal FEMALE…and we all know how good chicks are at math and liberals are at anything requiring intelligence..says all that an adult needs to know.
Carry on
I wish I had confidence that Trump’s pick is up to the job. He has a track record of sloppy economic analysis – and seems to sometimes not understand economic statistics. He has published almost nothing in major economic journals. Trump picked him because he’s a Trump booster on social media.
I wish I had time to look it up – but the initial jobs report and the later revisions are really not the same figures at all. The initial report is based on surveys of employers. The surveys are generalized, and that number is reported. Employers might make mistakes, they might lie, they might miss a month, or two, or three. But BLS just reports what the surveys say. The “revised” figure is based on real data that is collected after the fact – like payroll taxes. It should be – and is – much more reliable.
So the “BLS Blows it Every Time” chart above really just shows that the surveys as they now are conducted are useless. The “initial” release really needs to be reconsidered. Who needs this earlier data? And why? Would it be better to delay the release for another week or two? Or just drop the survey altogether? Or change how the data is collected? Or are Democrats just gumming up the data for nefarious purposes? On the last question, I think no. The initial data is so randomly unreliable that it doesn’t seem to show intentional manipulation.
The government has given itself the power to track almost any transaction of any size.
Just track those transactions.