The Bellwether Test Is Broken. Here Is a Better One.

Every four years a county gets its press cycle. It has picked the winner in every presidential election going back decades, so reporters drive out to its diners and ask what the country is thinking. The credential sounds like it should mean something. It mostly does not, and the reason is arithmetic rather than cynicism.

The winner-picking streak is a sign test. It asks one question per election: did this county land on the same side of the line as the nation? Yes or no. It does not ask by how much. In 2008 the country broke D+7.4 in two-party terms. A county that voted D+7.4 that year and a county that voted D+40 both cleared the bar identically. One of them was describing the United States. The other was describing itself, loudly, and happening to face the same direction.

So we grade differently. For each of the seven presidential elections from 2000 through 2024, take a county's two-party margin, subtract the national two-party margin, and keep the absolute value. That is the gap — how far the county sat from the country that year, in points. Average the seven gaps. Lowest average wins. By that standard the closest county in the country is Sumter County, Georgia, at an average gap of 1.85 points — but Sumter cast 12,049 votes in 2024, and Baldwin County, Georgia is second at 2.07 on 18,825. Among counties casting at least 25,000 votes, the leader is Montgomery County, Ohio, at 2.09 points, followed by Bucks County, Pennsylvania, at 2.30. Everything below is that larger-county ranking.

The one-bit test

Here is the baseline every county is measured against. Seven elections, one number each.

Notice how narrow that band is. The widest result in the set is D+7.4; the widest the other way is R+2.5. In a country decided by margins that small, a great many counties can sit on the winning side by accident. And in the one genuine wave here — 2008 — the sign test was nearly free: every county that voted Democratic at all, by two points or by forty, got credit for calling it.

The streak test has two other problems. It is a survivorship contest: run a coin-flip filter across thousands of counties seven times and some will come through clean on luck alone, and those are precisely the ones that get written about. And it is absurdly brittle in the wrong direction. A county that misses by two-tenths of a point loses its title forever, while a county that called every winner by thirty-point blowouts keeps its title and its diner interviews.

A county that voted D+40 in 2008 picked the winner exactly as well as one that voted D+7.4. One of them described the country. The other described itself.

A better question: how far off were you?

The gap measure throws nothing away. It uses magnitude, not just direction, and it keeps all seven elections rather than collapsing them into a pass/fail badge. A county with a small average gap is not necessarily good at predicting anything — nobody is — but it is a place whose result you can read as a rough stand-in for the national result. That is the actual job description of a bellwether, and it is a measurable one.

What the list actually looks like

It is not a collection of quaint rural counties. It is mid-size metros and their suburbs: Dayton in Montgomery, the Philadelphia suburbs in Bucks, Shreveport in Caddo, South Bend in St. Joseph, Rockford in Winnebago, Manchester and Nashua in Hillsborough, New Hampshire, Tampa and St. Petersburg in the two Florida entries, Peoria, Reno in Washoe, Albuquerque's northern suburbs in Sandoval, and Portland's Washington-side suburbs in Clark. Twelve counties spread across ten states, together casting roughly 3.1 million votes in 2024 — from 79,138 in Peoria to 671,993 in Hillsborough, Florida, a more than eightfold range. Resembling the country is not a function of size.

It is also worth saying plainly that most of these places do not matter to the Electoral College. Pennsylvania and Nevada were contested in 2024; Louisiana, Indiana, Illinois and Washington were not. The counties that best describe America mostly live in states where the presidential outcome was settled in advance. That is not a paradox — it is what happens when you measure resemblance instead of consequence.

Working the arithmetic backward on 2024: with the nation at R+1.5, Montgomery's D+2.0 lean puts its own margin at roughly D+0.5, and Bucks's D+1.4 lean puts it at roughly R+0.1. Two counties that finished, in effect, tied.

2024 tilted the whole board

Look down the 2024 lean column and something jumps out: ten of the twelve leaned Democratic relative to the nation, and only the two Florida counties leaned Republican. The average lean across the list was D+3.2. These were not random errors scattered around zero. Anyone treating this list as a live thermometer on election night in 2024 would have read the country as roughly three points bluer than it was.

Seven of the twelve posted a 2024 gap larger than their own seven-cycle average, and four of them at least doubled it: Clark (8.7 against an average of 3.46), Sandoval (7.4 against 3.19), Peoria (6.3 against 3.05) and Caddo (6.2 against 2.35). Whatever moved the country in 2024 did not move those four with it.

Montgomery and Bucks did the opposite. Both came in tighter than their own career averages — 2.0 against 2.09, and 1.4 against 2.30. They lead this ranking not on a long-ago reputation but because they tracked the country in the most recent cycle too.

One housekeeping note before anyone builds a theory on it: the average size of the national margin over these seven elections was 3.24 points, and the average 2024 lean across these twelve counties was 3.2 points in the Democratic direction. Those two numbers have nothing to do with each other. It is a coincidence, not a finding.

How much this buys you

Not as much as a clean ranking implies, and we would rather say so than let the table do the overselling. The whole list spans 1.37 points, from 2.09 at the top to 3.46 at the bottom. The distance between first and fifth is 0.40 points. Hillsborough and Pinellas, Florida are tied outright at 2.86. These are seven-observation averages; a single unusual cycle could reorder the middle of this table entirely, and 2024 already showed four of these counties drifting by more than double their historical gap.

So the honest claim is narrow. Montgomery County, Ohio and Bucks County, Pennsylvania have, over seven elections, sat closer to the national result than any other county we measured — Montgomery closer, on average, than the national margin itself typically was to a tie. If the pattern holds, they are the two places whose returns are worth reading early as a rough proxy for where the country is landing. If it does not hold, that will be visible in the same number, the same way, four years from now. That is the point of scoring distance instead of handing out streak medals: the measure keeps working even when the county stops being right.