Growth Picked a Side. Decline Didn't.

Take every county with a complete population series and a complete presidential series — 3,103 of them — sort by residents gained or lost between 2000 and 2024, and cut the list into five equal groups. The politics come out in order. The fifth that shrank fastest moved 24.23 points toward Republicans relative to the nation, on average. The next fifth, 23.07. Then 18.19. Then 12.64. The fifth that grew fastest moved 1.55. Five buckets, five steps down, no reversals.

That is tidier than we expected and tidier than we are comfortable with. A monotone gradient invites a mechanism, and the mechanism everyone reaches for is decline — that emptying places resent being emptied and vote accordingly. We think that reading is wrong, and the same file that produces the gradient is what kills it.

The gradient is entirely on the growth side

Split the file at zero. Among the 1,749 counties that gained residents, the correlation between population change and partisan shift is 0.269 — modest, real, pointing the way the table implies. Among the 1,354 that lost residents, it is 0.023. Knowing a county shrank 30 percent rather than 3 percent tells you nothing about how far right it went.

So the first two rows of that table are not a step. They are one undifferentiated mass averaging R+23.42, and the ordering inside them is noise. Whatever the gradient measures, it is happening among places that are filling up, not among places that are emptying out.

West Virginia and Mississippi lost people the same way and voted the opposite way

Group the shrinking counties by state and the spread runs from R+45.59 to R+3.22 — a 42-point range across places sharing the one characteristic the decline story calls decisive.

Mississippi's shrinking counties lost 16.1 percent of their residents on average — a steeper decline than West Virginia's 12.8 percent or Kentucky's 9.9 — and 25 of the 56 moved toward Democrats. Across the 172 shrinking counties of West Virginia, Kentucky, Arkansas and Oklahoma, that number is zero.

Two counties carry the whole point. McDowell County, West Virginia went from 27,099 residents to 17,186 — down 36.6 percent — and moved 93.81 points toward Republicans. Washington County, Mississippi went from 62,657 to 41,218 — down 34.2 percent — and moved 20.41 points toward Democrats.

If decline were the mechanism, these two counties would be in the same place. They are 114 points apart.

The growth side breaks too, in both directions

The gradient is not clean where it is strongest either. Trousdale County, Tennessee grew 64.8 percent — 7,395 residents to 12,189 — and moved 92.08 points toward Republicans, the eighth-largest Republican shift in the file. Ten of the twenty largest Republican movers gained population.

Among the twenty fastest-growing counties, fifteen moved toward Democrats and five moved hard the other way: Flagler County, Florida at R+31.12, McKenzie County, North Dakota at R+25.19, Sumter County, Florida at R+24.22, Pinal County, Arizona at R+18.96 and Osceola County, Florida at R+2.94. The fastest-growing fifth is close to a coin flip — 303 of its 621 counties moved toward Democrats, 49 percent.

What a correlation of 0.34 is actually worth

Across all 3,103 counties, population change and partisan shift correlate at r = 0.341. Squared, that is about 12 percent of the variance. Eighty-eight percent of why a county moved is somewhere else, and this piece cannot see it.

Worse for our own framing: county size beats population change on the same test. The log of 2024 turnout correlates with shift at 0.391, above growth's 0.341. What size does not do is line up in order — its quintile means run R+20.45, R+26.61, R+22.67, R+14.12, then D+4.17, reversing at the second step. A county's 2000 lean barely correlates at all, at −0.111, and also reverses. Population change is the only one of the three that is monotone, and monotone is a presentational virtue, not an explanatory one. We are pointing at our own headline table on purpose.

What would have to be true

There is no forecast in this piece. What we will say is what would have to hold for the gradient to be more than description.

• The correlation inside the shrinking half would have to be something other than 0.023. The decline story has no gradient to stand on.
• Mississippi's 56 shrinking counties would have to look like West Virginia's 44. The gap is 42 points wide.
• The ten growers among the twenty largest Republican movers would need an explanation the table can supply. It cannot.

Until then the honest version is narrower than the headline. Inside the growing half, the more a county grew the more it moved toward Democrats. Counties that shrank moved right almost everywhere except Mississippi, and how far they shrank did not matter. Those are two findings that a quintile table makes look like one.