Counting the Unemployed: The Messy Truth Behind the Jobs Report - Part 2

If you haven't read yesterday's blog post, please read it first before continuing here.

The Numbers Keep Changing

If there's one thing that should make you more skeptical of economic headlines, it's this: the first jobs number you see is almost never the final number. The initial report is based on whatever employer survey responses have come in by the deadline. It gets revised the next month when more responses arrive, revised again the month after that, and then revised once more during the annual benchmark process, when the BLS compares its survey estimates to a near-complete count of all jobs from state unemployment insurance tax records.

That benchmark revision is where the biggest surprises happen. In September 2025, the BLS revealed that employers had added 911,000 fewer jobs than previously reported over the twelve months from April 2024 through March 2025. It was the largest downward benchmark revision on record. When the final numbers were incorporated into the February 2026 jobs report, total job growth for 2025 was slashed from 585,000 to just 181,000, an average of roughly 15,000 jobs per month. Several months that originally showed gains were revised into outright contractions. Throughout 2025, policymakers, investors, and businesses had been making decisions based on data that significantly overstated job creation. The Fed was calibrating interest rate decisions on it. And for three consecutive years, 2023, 2024, and 2025, initial job numbers were revised downward, with cumulative overstatements estimated at around 1.5 million jobs.

Why the Revisions Have Gotten Worse

Several factors are at play. Survey response rates have dropped sharply, from about 60% pre-pandemic to below 45% recently. The rate at which new businesses agree to participate has fallen even further, from over 70% in 2015 to between 25-35%. On top of that, the BLS uses a "birth-death model" to estimate net job creation from businesses opening and closing. It works reasonably well in stable times, but tends to miss turning points, exactly when accurate data matters most. And then there's the media problem: the first number gets the headline. The revision gets a footnote, if it gets anything at all.

A Long History of "Is This Data Rigged?"

Accusations that the jobs data is cooked are nothing new. Both parties have attacked the BLS when the numbers are inconvenient. In 1962, President Kennedy convened a formal commission after a Reader's Digest article alleged political manipulation of employment data. The commission dismissed the charges. In 1979, another commission found no manipulation but criticized the survey's methodology. In 2025, President Trump fired BLS Commissioner Erika McEntarfer after a weak jobs report, calling the data "rigged."

There is an important distinction here. The legitimate critiques are real, including declining response rates, a birth-death model with known weaknesses, and benchmark revisions that have been unusually large. These are serious methodological challenges. But the claim that career civil servants are deliberately falsifying numbers for political purposes is not supported by evidence, and has been contradicted by every independent review conducted over the past sixty years.

A Few More Things Worth Knowing

Seasonal adjustments are invisible but powerful. Almost every jobs number you see has been adjusted to strip out predictable seasonal patterns (holiday hiring, summer teacher layoffs, etc). These adjustments are updated annually and can retroactively change previously published data.

Historical comparisons are trickier than they look. Definitions, age thresholds, survey questions, and industry classifications have all changed over the decades. Comparing today's unemployment rate to the 1970s requires more caveats than most commentators bother with. International comparisons are even trickier. Different countries define "actively seeking work" differently, use different age thresholds, and employ different survey methods. Cross-country unemployment comparisons are rarely apples-to-apples.

The data is always backward-looking. The jobs report describes a reference week from the prior month. In a fast-moving economy, it is a rearview mirror - not a windshield.

The Bottom Line

The monthly jobs report is powerful, consequential, and deeply imperfect. It is built on surveys with declining participation, definitions that exclude millions of struggling workers, and a revision process that routinely rewrites history. In 2025, the largest benchmark revision ever, 911,000 jobs, revealed a labor market that had been far weaker than anyone realized in real time. None of that makes the data worthless. It makes it something that deserves to be read carefully rather than taken at face value.

The next time the jobs number flashes across your screen on a Friday morning, resist the urge to draw sweeping conclusions. Ask what the revisions say. Ask who isn't being counted. And remember that the most important thing about any single data point isn't the number itself - it's understanding how it was made.