DOGE Contract Cuts Tracked Which Counties Opposed Trump

By September 2025, the federal government’s self-styled cost cutters had posted a public list of 13,231 canceled contracts, a tally meant to read as pure arithmetic. A new peer-reviewed analysis finds that arithmetic wasn’t so pure: the single strongest predictor of which contracts got cut wasn’t their price tag or their agency, but how strongly the county receiving them had voted for Donald Trump in 2024. Counties that gave Trump his weakest support lost contracts at rates the study’s numbers make hard to wave away. The agency built to hunt waste, it turns out, left an unusually clean partisan fingerprint behind.

That agency is DOGE (the Department of Government Efficiency), created by executive order in the first days of Trump’s second term and charged with rooting out what the administration called bureaucratic waste. You’d expect a purely technocratic exercise, run by algorithm and spreadsheet, to be blind to how a county voted; that expectation is exactly what the numbers challenge.

DOGE’s public list only ever tracked one slice of what it did: cutting federal contracts, alongside mass layoffs of federal workers and new oversight of the civil service it left in place. Housed directly inside the Executive Office of the President and free of any fixed savings target, it had wide latitude to decide which contracts lived and which got canceled. Political scientists already had a name for presidents steering money toward friendly territory: presidential particularism, the tendency for federal benefits to cluster in places that vote for the president. What hadn’t been tested, until now, was whether that same particularism could run in reverse: not who gets rewarded, but who gets punished.

That’s the question Lukas K. Alexander, a political scientist publishing in the American Political Science Review, set out to answer with the DOGE data. He built a dataset of nearly 700,000 federal contracts and checked which ones DOGE had canceled against where the money was headed politically.

Seven Hundred Thousand Contracts, County by County

The bones of the analysis are procurement records, not talking points: every contract active in the Federal Procurement Data System when Trump took office in January 2025, dating back to fiscal year 2004, matched against DOGE’s own list of cuts. DOGE had publicly claimed 13,231 canceled contracts as of September 2025; of those, 7,507 could be matched to verifiable federal procurement records and made it into the analysis. Each surviving contract then got geocoded to a county by way of its postal code, an address that (once matched against county politics) doubles as a data point in its own right, and stitched to how that county voted in the 2024 presidential race.

Plotted out, the pattern looks almost too clean for social science: a downward-sloping line running from counties that backed Trump hardest, which saw the fewest cuts, to counties that opposed him hardest, which saw the most. Not a subtle line, either. Was it just noise? The correlation between Trump support and logged contract cuts worked out to negative 0.48, hardly a coincidence in a field used to messier data.

Sorting Counties Into Core, Swing and Hostile

Alexander ran the numbers two ways, first contract by contract, then county by county, controlling for population, income, education levels, how many contracts a county typically pulled in, and whether its member of Congress happened to belong to Trump’s own political party, a possible confound in its own right. In both versions, higher Trump support meant fewer cuts: a 10 percentage point jump in a county’s Trump vote share was associated with roughly a 33% lower odds of any individual contract being cut, and a roughly 22% drop in the total number of contracts a county lost. He also sorted counties into three buckets echoing older work on presidential favoritism: core counties, where Trump cleared 55% of the two-party vote; hostile counties, where he didn’t reach 45%; and swing counties in between. Contracts in hostile counties were significantly more likely to be cut than contracts in core counties, at both the individual and county level; swing counties leaned the same direction but not reliably enough to call it more than a hint. None of it moved much once he checked whether a county’s own member of Congress shared Trump’s party, which stayed statistically flat across all four models.

That last result cuts against one easy explanation: that DOGE was simply protecting friendly incumbents in a way that happened to track voters too. The pattern also wasn’t a rerun of undoing Biden-era spending, counties with more contracts signed under the previous administration were, if anything, somewhat less likely to see cuts, not more.

None of this proves intent, and Alexander is careful to say so: the design is observational, built from DOGE’s own public accounting of what it cut, which by DOGE’s own description (its words, not the author’s) folds contract cancellations together with layoffs, asset sales and other savings it counts differently. He ran a check for whether Republican-leaning states might simply be underrepresented in DOGE’s public reporting, a gap that could hide cuts in friendly territory, and found no sign of it. A real limit, not a fatal one. It is also not a limit unique to this study: a Government Accountability Office audit separately found roughly $34.6 billion of DOGE’s claimed contract savings could not be substantiated or corroborated as terminated, and that about 27.8% of the contracts DOGE listed as canceled lacked enough identifying information to check against federal procurement records at all.

A Pattern With Later Echoes

Existing research on presidents and federal money has mostly gone the other direction: which places get rewarded, not punished, for their votes. Alexander’s argument is that the same tools cut both ways, and that DOGE’s total lack of a fixed savings target made it an unusually easy instrument for turning administrative discretion into what he calls a punitive form of presidential particularism. He points to later episodes outside his own dataset: the Office of Management and Budget’s freeze of $2.1 billion for a Chicago transit project the administration called race based (its phrase, not the study’s), and its suspension of nearly $8 billion in climate spending concentrated in 16 states, every one of which voted against Trump in 2024. Whether those count as more of the same pattern, or coincidence, is exactly the kind of question a single study of one agency’s contract list can point toward without settling.

DOGE itself has mostly faded from headlines since the numbers in this dataset were collected, its website’s savings count moving on to other categories entirely. What it leaves behind, if Alexander is right, is a template: an agency built to look apolitical, behaving, county by county, like anything but.

FAQ