The real reason behind the stunning U.S. job revisions and why Trump's firing of the BLS commissioner is utter nonsense

Globe and Mail

a day ago

'For the FOURTH month in a row, jobs numbers have beat market expectations with nearly 150,000 good jobs created in June. American-born workers have accounted for ALL of the job gains since President Trump took office and wages continue to rise.' - White House Press Secretary Karoline Leavitt, July 3rd, 2025
'In my opinion, today's Jobs Numbers were RIGGED in order to make the Republicans, and ME, look bad.' - President Donald Trump, August 1st, 2025
What a difference a month makes. Strong leaders share the credit and accept the blame. Weak leaders take all the credit and lay the blame on others.
Talk about a classic case of shooting the messenger. If you don't trust the payroll data, then just go to the companion survey, which showed a huge 260,000 jobs decline in July and down 402,000 since the end of the first quarter (in the aftermath of all the tariff-related uncertainty if you are seeking out a culprit). And with no revisions to blame, either. What a sham. We are on a slippery slope, folks.
President Trump said BLS Commissioner Erika McEntarfer would be 'replaced with someone much more competent and qualified,' claiming in a social- media post the government's jobs numbers were manipulated. What utter nonsense, but nary a peep from Congress who worry about being primaried.
Never mind that Ms. McEntarfer wasn't merely nominated to the post by then President Joe Biden, but she was confirmed by the Senate 86-8 in January 2024 – and Vice President JD Vance, then a senator, was among those voting for her! Did she all of a sudden become incompetent?
Hard to fathom.
I hardly would fire a BLS commissioner because of the headline or revisions to the data, which are normal – in fact, the sort of downward revisions we saw in the last two months, while very large, is hardly without precedent. We have seen revisions close to this no fewer than two dozen times back to 1980. Nobody else ever got fired over it. This was a large two-month downward revision, to be sure, but that is only because the numbers in May and June were grossly overstated and every other employment statistic showed that it was nonfarm payrolls was the odd man out. And the revisions only corrected that anomaly.
The plain fact of the matter is that there is nothing insidious nor nefarious going on. No attempt to mislead and no sloppy usage of the data. No case for Erika McEntarfer, who has been a government statistician since 2002 which covers a span where Bush, Obama, Biden, and Trump were in the Oval Office, to be fired. This is one part ruse and one part deflection. That's all it is. The fact that this last two-month revision (-258,000) was so big only attests to how the Establishment survey was so out of sync with the other data which is why the consensus on the first release has been consistently below what came out initially. So, I ask: what is so difficult to figure out here beyond the sampling problem which the BLS did not create?
The issue is with the post-Covid plunge in the business 'response rate'.
This is not about the BLS which is forced to deal with the data that companies send in with respect to the initial release.
It seems completely lost in this discussion that the root of the problem is the historically low company response rate to the first round of the monthly survey – this is a survey that depends on business cooperation and the reality is that the response rate does not approach anything that can be considered reliable until that second revision comes in. Maybe the BLS should simply stop publishing the payroll data so quickly – think of the first release as something no more than an incomplete snapshot of the labor market because it is no easy task 'to get it right' in the days that follow a month in a market as complex and large as a 130 million workforce, and all the churning that goes on beneath the surface. What we gain in speed of delivery of the data we lose in the veracity given the naturally lower sample size once the response rate rises in the next two months.
The one thing to consider is that it is an entire employment report, replete with a wealth of information beneath the headline, even if incomplete at first. But there is typically a high error term in the first go-around and especially since the pandemic as a record low share of businesses 57% get in their responses now in time for the first payroll release. Pre-covid it was over 80% in terms of the response rate. By the time the third revision comes in, and the response rate goes to 94%, where it's always been in the past and it is only then that the BLS truly has enough information collected for anyone to get an accurate portrayal of what the labor market really looked like in the month of the first release.
It's really something that only now are people paying attention to the fact that first estimates get revised as more accurate information is received. This has been a fact of life… forever. Nobody was talking about it a month ago, funny enough. And there will be future benchmark revisions in the future as even more information comes in. Everyone who follows the data closely knows that there is a high error term in the initial release of everything from payrolls to retail sales to GDP. It is all written up each month in the detailed notes to the data releases. The price paid to receive information quickly is the accuracy, as it pertains to the initial report. Nobody is amazed that we got July data on the first day of August? And this number will get revised too, for sure. These are preliminary estimates only with a large error term only because the sample size with the first stab at the employment report is so small. Why is everyone so shocked?
It's not as if the BLS hides from the fact that the smaller the sample size, the larger the error term … this is taken right from the report (the range of possibilities is huge but is stated for the record):
'The confidence interval for the monthly change in total nonfarm employment from the establishment survey is on the order of plus or minus 136,000 … The precision of estimates also is improved when the data are cumulated over time … in the establishment survey, estimates for the most recent 2 months are based on incomplete returns; for this reason, these estimates are labeled preliminary in the tables. It is only after two successive revisions to a monthly estimate, when nearly all sample reports have been received, that the estimate is considered final.'
Maybe the way the BLS reports the data should be changed, but it is at behest of the companies reporting in their payroll on time and accurately. Maybe those in the trading pits should be forced to wait two to three months for the better estimate instead of being spoon fed something quick with a low sample size.
You just need to compare the business response rate of the first NFP estimate to the month containing the second revision – as aforementioned, from around 58% to 94% -- to see how the BLS is forced to make guesswork out of the 42% of the business universe that fail to report their headcount on time. The information trickles in the next two months. Maybe there should be a financial penalty applied to the firms who don't send in their information on time. I've been talking about this discrepancy for the past few years … and, in fact, the revisions have constantly been on the downside.
The next question is why have the revisions been squarely to the downside, even before last Friday's report? Prior to what we saw unfold on Friday, there were downward revisions to every month of the year, and they totalled 188,000. That was before the downward two-month revision of 258,000 in May and June. Ergo, this has been a pattern all year long and transcends what happened in the July report.
There is also the question as to why the data are constantly being revised lower. This is akin to asking why the prior payroll data were so artificially inflated. Once again, at the time of that initial release, the BLS is compelled to deal with whack load of guesswork. It must fill in the gaps from the fact that, once again, the initial response rate is historically so low. There is a huge information gap. The lower the sample size, the wider the confidence interval and the higher the error term – a basic premise of statistical analysis.
The issue is that since Covid, the small business sector, in particular, has been slow to send in their updated staffing level numbers to the BLS in time for that first survey. And we know for a fact that the small business sector (fewer than 50 employees) has created no jobs at all over the past six months and have on net fired -42k workers over the May-July period. The BLS very likely was extrapolating small business job creation that simply did not exist over the spring and into the summer and that anomaly was corrected last Friday. End of story.
Nobody from the White House discusses this, but what happened on Friday with the revisions is that nonfarm payrolls, which had been the odd man out, was brought into alignment with the vast array of other very soft labor market indicators of late. For example, the average private sector nonfarm payroll print of 51,000 from May to July now more closely approximates (actually a little higher) the ADP comparable of 37,000. Mr. President – it's not as if the BLS is any further away from telling the same story as ADP is. Do you want to know the name of the person who is president and CEO of ADP so you can dismiss here too (if you can)? Her name is Maria Black. Maybe she needs to be subpoenaed.
Over this same May-July period, the Fed's Beige Book showed half the country posting flat to negative job growth. All the payroll numbers did on Friday was reflect that. The University of Michigan consumer sentiment data on employment in July lined up as the fourth worst reading since the end of the Great Financial Crisis in mid-2009. The Conference Board's consumer confidence survey showed only 30% of those polled stating that jobs were 'plentiful', the lowest since April 2021 – surely households would have a pretty good idea of what their job situation is, don't you think?
But just in case you want to have the President and CEO of the Conference Board fired too, his name is Steve Odland, and I'm sure he is not too hard to find.
There are plenty of culprits around these days spreading bad labour market news.
David Rosenberg is founder of Rosenberg Research.