The biggest story in the federal workforce cuts isn't how many left, it's who wasn't replaced

The story of the 2025 federal workforce reduction has been told in headcount. In September 2025 alone, about 121,000 civilian employees left federal service, nearly six times the figure for the same month a year earlier, the bulk of them through a governmentwide deferred resignation program with an end-of-September deadline. The number is large, and the debate about whether the government is now too small or right-sized will continue. But headcount is the wrong number to watch. What determines whether agencies can still do their work is not how many people left. It's who got replaced.

I looked at public Office of Personnel Management data on federal separations and accessions, and I want to be precise about the window. This is a snapshot of one unusual month, the month the deferred resignation deadline fell, compared with the same month in 2024 to strip out the seasonality in federal hiring. It is not a yearlong trend. It is a close look at what happened during a single, intense contraction. For every occupation, I computed a simple replacement ratio: hires divided by departures. A ratio near 1 means departures are being refilled. A ratio near 0 means they are not.

In that month, replacement broke sharply along skill lines. The occupations that went barely replaced were the analytical, technical and acquisition roles. The ones that kept being refilled were frontline service roles. Behind the percentages are stark raw counts, all for September 2025. Among management and program analysts, 8,218 people left and 106 were hired, a 1% replacement rate. In information technology, 7,456 left and 181 were hired, a 2% replacement rate. Contracting, the function that runs federal procurement, saw 3,520 departures and 174 hires, a 5% replacement rate. Statisticians saw 319 departures and three hires. By contrast, nurses saw 1,185 departures against 731 hires, a 62% replacement rate. Medical officers were replaced at 58%, while criminal investigators were replaced at slightly above one-to-one.

How do we know this is unusual and not just how these jobs always hire? Because a year earlier, hiring followed a much more typical pattern. In September 2024, management analysts saw 497 departures against 608 hires, information technology workers saw 645 departures against 814 hires and statisticians saw 16 departures against 27 hires. In an ordinary month, these roles refill at or above replacement. The single-month comparison is one baseline, not a law of federal hiring, but it is a clean one, and it shows that what happened in 2025 was a break from the prior year, not the normal rhythm of these occupations.

A word on how I separated higher-skill from lower-skill work, because it matters. I ranked occupations using a simple composite of attributes the OPM data already report: share with a bachelor's or advanced degree, share in STEM or health fields, average tenure and average pay. The conclusion does not hinge on the exact recipe. The same split appears if you rank occupations by any one of those measures alone, for instance degree level or salary. The composite is a convenience, not a black box on which the finding depends.

Why did the break fall this way? Two policies acted at once, and here the data are clear about the mechanism even where they cannot prove intent. A near-total hiring freeze shut off replacement across the board, which is why hiring fell from hundreds to handfuls. At the same time, the deferred resignation program drove departures, and it drew most heavily on the technical occupations. Between 86% and 94% of the 2025 departures in management analysis, information technology, contracting and statistics came through that program, compared with roughly 20% for nurses and medical officers. Frontline roles, with statutory or operational staffing floors, continued hiring. An across-the-board instrument produced a sharply uneven result.

What the data show is the replacement gap. What they do not directly measure, but what reasonably follows, is the consequence. When a contracting shop loses most of its officers and refills almost none, the expected result is slower procurements and thinner oversight. When information technology replacement stops, modernization and security work are the likely casualties. When analysts and statisticians leave and are not refilled, leadership has less capacity to answer its own questions with evidence. These are inferences about capability, not measurements of it, and they should be read that way. But they are the natural reading of a contraction that hit hardest exactly where expertise is most concentrated and slowest to rebuild.

The narrower, defensible point is this. Headcount, vacancy rates and attrition all registered a smaller workforce. They did not show that the composition of what remained had shifted away from the hardest-to-rebuild skills. But that shift is measurable in close to real time with data the government already publishes. A replacement ratio computed by occupation and checked against the prior year is a simple early indicator that a chief human capital officer could act on through targeted hiring exceptions, retention or knowledge transfer while the expertise can still be recovered.

A workforce reduction is a policy choice, and reasonable people will disagree about its size. But there is a difference between deciding to be smaller and, without quite meaning to, deciding to be less capable. In this one episode, the numbers point to the second. That is the kind of thing worth measuring before it becomes the kind of thing you only notice in hindsight.

Ali Akram is a data scientist and independent researcher in Minneapolis who studies public sector workforce capacity and staffing. He previously served as the Human Resources Workforce Data Analyst for the City of Minneapolis. He holds an M.S. in Data Science from the University of St. Thomas.