Open any map of the 2024 election and the country looks like a settled argument: a sea of red counties, blue islands at the cities, and a thin band of purple counties where the two are supposed to meet. The purple counties carry a lot of meaning. They are read as the moderate middle — the places that still mix, that could go either way, where a campaign might change minds.
Most of that purple is an artifact of arithmetic. A county's margin is a mean, and a mean hides its spread. Centre County, Pennsylvania split almost evenly in 2024 — Harris by 2.9 points. Its precincts did not split evenly at all: the precincts around the Penn State campus broke for Harris by margins north of 70 points while the surrounding ridge-and-valley townships broke for Trump by as much, and almost nothing sat in between. Centre County is not a purple place. It is a deep-blue dot in a deep-red field, and averaging the two paints a moderate that does not exist.
Do this for every county and a clear pattern emerges: the precinct map is more polarized than the county map, not less. Counties are the coarsest honest unit of American election data, and they systematically launder the sort.
Two maps of the same country
The familiar choropleth on the left colors each county by who won it and by how much — the map every outlet published in November. Flip the switch and the same counties are recolored by something the first map cannot show: how divided each county is on the inside, measured as the spread of its own precincts' margins. The deepest purple counties are not the ones that split down the middle. They are the ones whose precincts split hardest while the county average lands near zero.
How a county voted — and how divided it is inside
2024 presidential two-party result. Switch between each county's margin and the vote-weighted standard deviation of its precincts' margins (its internal division, in points). Hover any county for both numbers. Counties with fewer than 5 reported precincts are drawn in beige.
The two maps disagree most exactly where it matters. A county can be a coin flip and a war zone at the same time: Saginaw, Michigan finished within three points but its precincts stand 37 points apart on average; Tarrant County, Texas, the country's largest perennial swing county, is redder in its average than it is moderate in its parts. This is the ecological fallacy that William Robinson warned about in 1950 — the group average is not the member — rendered as cartography. Jonathan Rodden's Why Cities Lose makes the same point structurally: Democratic votes are packed into dense precincts so efficiently that the geography itself, not just the electorate, manufactures the divide. You cannot see packing on a county map. Packing happens below it.
Where the votes actually pile up
Line up every vote by the margin of the precinct or county it was cast in, and the difference between the two resolutions is stark. At the county level the distribution still has a soft middle. At the precinct level the middle caves in and the tails fatten: 18.7% of voters live in a competitive precinct against 23.8% in a competitive county, while 64.8% sit in a 20-point precinct landslide against 56.8% at the county scale. Same election, same ballots — a more sorted country the closer you look.
The middle is thinner up close
Share of the two-party vote cast in places of each margin, 2024 presidential.
The averaging is easiest to see one county at a time. Below are six counties the November map painted purple or nearly so — from a college town to metro Phoenix. Each dot is a precinct, sized by its votes and placed by its margin. A genuinely moderate county would cluster its dots around the center line. These do the opposite: they hollow out the middle and bunch at the edges, two electorates sharing a courthouse.
Inside six "purple" counties
Each dot is a precinct, placed by its two-party margin and sized by its two-party votes. The line marks an even split; the county's own margin is printed above each panel.
The vanishing competitive precinct
Bill Bishop's The Big Sort made its case in counties: Americans, he argued, increasingly live in landslide counties where they rarely meet the other side. Counties were the finest data he had, and they undersold his own thesis. At the precinct level — the scale of an actual neighborhood polling place — nearly two in three 2024 voters cast a ballot in a 20-point blowout, and fewer than one in five lived anywhere close to a tossup. In 1,275 precincts a major-party nominee did not lose a single two-party vote; the other side simply was not there.
None of this requires anyone to have moved. It is what the country already looks like when you stop letting the county line do the averaging. The competitive precinct — the place where neighbors actually disagree at the ballot box — is the genuinely endangered American geography, and it is rarer than even the sorted county map suggests.
What this shows — and what it doesn't
This is a single 2024 cross-section, not a trend: precinct returns are not harmonized across years, so the "more sorted than ever" story belongs to the county record, while this page shows how deep the 2024 sort runs once you leave the county behind. A precinct is an administrative unit, not a neighborhood, and its size varies enormously — California reports roughly 578 precincts per county, North Dakota about 7 — so a precinct in one state is a coarser or finer lens than in another; every figure here is vote-weighted to keep big and small precincts comparable. The internal-division map is drawn at county resolution and colored by a precinct-derived statistic; it does not plot precinct boundaries, which would require precinct shapefiles this analysis does not use. And this is geography, not behavior: a precinct margin describes a place, not why anyone there voted as they did, and the ecological fallacy cuts both ways — a deep-red precinct still contains Democrats. What the precinct lens establishes is narrow and solid: the county map systematically understates how separated the two electorates already are.