The Freak-Out Over the Bureau of Labor Statistics
The July jobs report landed with a thud—just 73,000 new payrolls, almost all in health care and social assistance. And then came the fireworks: President Trump fired Bureau of Labor Statistics Commissioner Erika McEntarfer, the press corps exploded, and every economic commentator from Cambridge to Stanford and from Chicago to New Orleans dusted off their prewritten essays on authoritarianism and statistical independence. But as usual, the outrage machine missed the real story.
The May and June revisions weren’t just large. They were historically large by any standard and especially by the standards of our post-pandemic data environment, where early estimates have admittedly grown less reliable. These weren’t “routine” corrections. They were the statistical equivalent of a trapdoor opening beneath our feet.
Let’s be precise. The BLS initially reported 147,000 jobs added in June. In the second estimate, that number was slashed by 133,000, leaving a final gain of just 14,000. May looked healthier at first—144,000 added—but two rounds of revisions later, the month closed out with just 19,000 jobs. The total downward revision across May and June: 258,000.
That two-month swing is the largest non-pandemic revision in modern history. And we’re not using “largest” in the rhetorical sense. We mean statistically, demonstrably, and unequivocally.
Revisions from Beyond the Known Planes of Reality
When we say the May and June revisions were extreme, we don’t mean “notable” or “somewhat surprising.” We mean statistically extraordinary—anomalies so far outside the bounds of normal variation that they belong not in a debate about survey noise, but in a diagnostic report on economic whiplash.
Take June. That 133,000 downward revision is a 3.6 sigma event from the historical median using robust z-scoring, which accounts for the fat-tailed nature of payroll data. In simpler terms, it’s the kind of outlier you’d expect maybe once in a generation. Using median absolute deviation (MAD)—a sturdier metric for this kind of data—the June revision lands over 5.3 MADs from the center of the distribution. Only one month in over 260 has come close.
And May? May’s second revision of −125,000 is, if anything, even more striking. More than half of the second revisions in the historical record are exactly zero. The median absolute deviation is literally zero, meaning May’s result exists in a vacuum of precedent. You can’t even apply a standard robust z-score because the distribution collapses. Using classical methods, May’s revision clocks in as an 11.29 sigma event. Not in ten years. Not in twenty. In the entire post-2003 historical dataset, nothing comes close.
Stack the two together and the May–June combination becomes a 6.8 sigma event, more than 10 MADs from the historical two-month median. In percentile terms, that’s the 0.0th percentile. A statistical black swan.
This isn’t just “these things happen.” These things don’t happen. And when they do, they are the signal, not the noise.
The Response Rate Excuse Doesn’t Explain the Revisions
Which brings us to the fallback explanation offered by many Trump critics: survey response rates are down, therefore revisions are bigger, therefore nothing to see here. And it’s true—response rates in the establishment survey have declined, dipping into the low-40s percent range. That has made first prints more volatile. But May and June aren’t just volatile. They’re exceptional even by recent standards. The post-pandemic baseline may be noisier, but these revisions are still outliers relative to the new normal.
So, what’s really going on? Four factors stand out.
First, massive federal layoffs. The federal government has been shedding employees at a rate unseen outside of post-census realignments—buyouts, attrition, direct reductions in force. The Trump administration is engaged in an unprecedented downsizing of the federal government. This is rarely visible in the monthly topline number but can create meaningful distortions in the revision process.
Second, monetary policy. The Federal Reserve quickly reversed course after Trump’s election, changing from a stance of easing policy to holding it at a restrictive level despite a months-long slowdown in core inflation. The real policy rate is now firmly in contractionary territory. It’s entirely plausible that the labor market weakened much faster than historical precedents because Jerome Powell’s Fed is acting more irresponsibly than any Fed in recent history—and the revisions are having to play catch up.
Third, immigration enforcement. The ongoing crackdown on illegal immigration and the associated “self-deportation” among unauthorized workers is likely suppressing payrolls. These workers don’t show up in unemployment claims—because they don’t qualify for benefits—but they do disappear from employer payrolls. It’s the kind of structural change that messes with the data even more than it does the narrative.
Fourth, tariffs. While probably the weakest of the four explanatory factors, uncertainty around tariffs can delay hiring, especially for small firms dependent on international inputs or equipment. It wouldn’t take much to tip their hiring plans from “maybe” to “not now.”
Panicans Gonna Panic
And so we return to the political controversy. Trump fired the BLS commissioner. Cue the hand-wringing predictions that our statistical data will quickly become as worthless as those produced in China. But critics insisting the revisions were routine are gaslighting the public. These were not normal revisions. They were among the largest, sharpest, and most statistically significant corrections on record. To deny that is to signal that your critical faculties shut down the moment you are offered a narrative that makes Trump look bad.
Moreover, Trump hasn’t installed a crony. He elevated William Wiatrowski, a longtime BLS veteran, as acting commissioner. This is not how you rig numbers. It’s how you preserve continuity.
The real scandal here isn’t that Trump lost confidence in a Biden-appointed commissioner. It’s that most of the commentariat has lost confidence in data analysis—whenever the numbers don’t conform to their narrative. They did not even bother to wonder if something might be wrong with the numbers when the easier, progressive-approved anti-Trump narrative was proffered.
The only madness here is pretending otherwise.
Read the full article here