America Didn’t Wait for the Unsorted to Die — They Switched

In 1972, one placed partisan in five was at odds with their own side. Among Democrats and Republicans (leaners counted with their party) who put themselves on the seven-point liberal–conservative scale, 19.6 percent were cross-pressured — conservative Democrats or liberal Republicans. By 2020 the figure was 5.2 percent; in 2024, 6.3 percent. A creature that once defined whole regions of American politics is now a rounding error.

The die-off is steepest where the species was most common. Conservative Democrats were 24.6 percent of placed Democrats in 1972; by 2020 they were 6.6 percent (8.2 percent in 2024). Liberal Republicans fell from 13.1 percent of placed Republicans to 3.7 percent, then 4.4. The mirror image grew in their place: fully sorted partisans — liberal Democrats plus conservative Republicans — rose from 43.6 percent to 69.3 percent, and the weighted correlation between seven-point party ID and seven-point ideology more than doubled, from 0.32 in 1972 to 0.74 in 2020 and 0.72 in 2024.

So far this is the canonical sorting story, the one Matthew Levendusky documented through the early 2000s and that the last two decades have only sharpened. The interesting question is the mechanism. The version that dominates popular commentary is demographic: the old conservative Democrats — Southern, New Deal–loyal, set in their ways — never changed; they died, and sorted young people took their place in the electorate. Call that replacement. The alternative is conversion: the same human beings, alive across the whole period, brought their party and their ideology into line midstream.

A repeated cross-section like the ANES can’t follow individuals. But it can follow birth cohorts, and that is enough to separate the two stories. Under pure replacement, each cohort carries its youthful cross-pressure rate through life: the lines stay flat, and the aggregate falls only because flat-high lines exit the population and flat-low lines enter it. Under conversion, the lines themselves bend down.

That is not what flat lines look like. Start at the left edge: in the 1970s, every birth cohort old enough to measure was equally unsorted. Pooled cross-pressure ran from 17.6 to 19.7 percent across the four cohorts with usable cells — people born before 1910 and Boomers fresh out of school were statistically indistinguishable. Being young in 1974 did not make you sorted. Nobody was sorted.

Then everything fell at once. The 1943–1958 Boomer cohort — the same birth years, sampled decade after decade — went from 17.6 percent cross-pressured in the 1970s to 5.0 percent in the 2020s. The 1927–1942 cohort, in its eighties and nineties by the end of the series, fell from 18.7 to 10.0 percent. The cohorts born later start lower and fall further still. The lines move together, in parallel, regardless of birth year: a period effect that hit everyone at once, which is what conversion looks like in cohort data.

That counterfactual is the accounting version of the eyeball test. Take the 2020s cohort composition, assign each cohort the cross-pressure rate it (or its 1970s elders, for cohorts born too late to observe then) showed in the 1970s — an average entry rate of 18.5 percent — and pure replacement gets you 18.3 percent, barely below where the country started. Closing the rest of the gap, from 18.3 down to 5.6, required people who were once conservative Democrats or liberal Republicans to stop being either: to switch their party to fit their ideology, or their ideology to fit their party.

The sorting ran down the education ladder

If conversion was driven by elite cues — voters noticing that the parties’ elites had polarized and updating to match, the mechanism Levendusky proposed — you’d expect the most politically attentive to convert first. Education is the bluntest available proxy for attentiveness, and the gradient is exactly as predicted.

College graduates were already down to 15.1 percent cross-pressured in the 1970s and reached 5.2 percent by the 2010s — roughly where the whole country would arrive a decade later. Grade-school-educated partisans started at 23.3 percent and were still at 28.6 percent in the 2000s, against 6.9 percent for college graduates in the same decade. The gap between the bottom and top of the education ladder widened from 8.2 points in the 1970s to 21.7 points in the 2000s, then narrowed (14.8 versus 3.6 percent in the 2020s) as the signal finally reached everyone. Sorting diffused like a technology: early adopters with the most exposure to elite political discourse first, everyone else on a lag.

The South is the conversion engine made visible

Nowhere is “the same people switched” more concrete than the South. In the 1970s, 48.6 percent of self-described conservative Southern partisans called themselves Democrats — the living remnant of the one-party Solid South, conservatives wearing a New Deal label their parents had handed them. That arrangement did not quietly age out of the population. It collapsed in plain view, decade by decade.

By the 2020s, 9.3 percent of conservative Southern partisans identify as Democrats — statistically on top of the non-South rate of 8.0 percent. Region, once the single best predictor of holding a mismatched party label, now predicts nothing. The regional gap didn’t shrink because Southern conservative Democrats were uniquely mortal; it shrank because they re-labeled, and their children never picked the label up.

What this does and doesn’t show

Honesty requires three caveats. First, the ANES is a repeated cross-section, not a panel, so “within-cohort change” bundles true individual conversion with compositional churn inside cohorts — differential mortality and differential survey response could each move a cohort’s line without anyone changing their mind. But composition stories strain to explain why every cohort fell in parallel, including cohorts in their thirties and forties whose mortality is negligible. The parallel collapse is the signature of a period shock, not attrition.

Second, the denominator matters. These figures cover partisans who place themselves on the ideology scale; respondents who say they haven’t thought about it are excluded — Kinder and Kalmoe’s warning that scale-placers are the more ideologically attentive slice of the public applies in full. Placement rates also moved: 59.7 percent of partisans placed themselves in 1972, just 35.6 percent in 2000, and 84.9 and 85.8 percent in the online-heavy 2020 and 2024 waves. The long-run collapse dwarfs anything plausible from denominator drift, but precise year-to-year comparisons should be read loosely.

Third, the recent waves are mixed-mode and rely on post-stratification weights, so conventional standard errors are too small. That matters for the kicker in the series: 2024 shows the first rise in cross-pressure on record, from 5.17 to 6.33 percent — a gain of 1.16 points, about 1.8 design-naive standard errors. Read it as a plateau, not a turnaround, until another wave weighs in.

The replacement story was always more comfortable: it asks nothing of anyone’s convictions, just patience and actuarial tables. The data tell a less flattering, more interesting story. Americans of every generation looked at a political system whose elites had pulled apart, and — educated first, South most dramatically, everyone eventually — resolved the contradiction themselves. The unsorted didn’t die out. They surrendered.