Partisans Finally Got Constraint

The most durable null result in the study of American politics has expired. For decades the same finding kept turning up: voters were sorting into the “correct” parties, but their opinions refused to organize. Knowing where someone stood on abortion told you almost nothing about where they stood on defense spending. Philip Converse called the mass public “innocent of ideology” in 1964, and in 2008 Delia Baldassarri and Andrew Gelman confirmed that, through 2004, not much had changed — party and issue positions were lining up, but the issues themselves still barely correlated with one another. They titled the paper Partisans without Constraint.

The American National Election Studies cumulative file now runs through 2024, and the two most recent waves are the largest ANES has ever fielded. Recomputing the classic quantity — the average pairwise correlation among six long-running issue scales, covering guaranteed jobs, aid to Black Americans, government health insurance, government services and spending, defense spending, and abortion — confirms the original result inside its window and then breaks it decisively outside. The mean correlation across all 15 issue pairs was 0.172 in the 1980s, 0.199 in the 1990s, and 0.234 in the 2000s: the modest drift the original paper described. Then it surged — 0.352 in the 2010s and 0.443 in the 2020s, 2.57× the 1980s level. Every one of the 15 pairs, without exception, is more correlated in the 2020s than it was in the 1980s.

01The matrix fills in

“Constraint” is Converse’s word for the degree to which knowing one of a person’s political opinions lets you predict the others. The cleanest way to see it is the correlation matrix itself: six issues, 15 pairs, one panel per decade, drawn on an identical scale. In the 1980s the matrix is pale — outside a cluster of economic items, most pairs sit near zero, and abortion is statistically unrelated to everything. Decade by decade the grid darkens, and by the 2020s there is no pale region left.

The tightening is not just the loose social–economic seams closing. Even the historically coherent core got more coherent: the average correlation among the four economic scales was 0.314 in 1972, and reached 0.590 in 2020 and 0.547 in 2024. The economically consistent welfare-state liberal or small-government conservative — once a minority pattern even within their own domain — is now closer to the default.

02Where the old story ended

Baldassarri and Gelman’s argument rested on a contrast between two trend lines. Issue partisanship — the correlation between each issue position and party identification — was clearly rising through 2004. Issue constraint — the correlation among the issues themselves — was not. Their conclusion: parties were re-sorting voters they already had, not forging ideologues. Plotted year by year through 2024, the chart now tells a different story. The two lines that diverged in their data have since risen together, steeply.

The alignment side of the ledger kept climbing too. The mean correlation between issue positions and party ID went from 0.204 in the 1970s to 0.498 in the 2020s — with aid to Black Americans moving from 0.156 to 0.560, government health insurance from 0.225 to 0.529, and guaranteed jobs from 0.232 to 0.527. Self-placement on the liberal–conservative scale, once a noisy signal of partisanship, now nearly is partisanship: its correlation with party ID rose from 0.394 in the 1980s to 0.731 in the 2020s.

Researchers watching the post-2004 data saw this coming. Daniel DellaPosta documented an accelerating “oil spill” of belief alignment through 2016, and Kozlowski and Murphy, revisiting the partisans-without-constraint thesis directly, found the flat line bending over the same window. What the 2020 and 2024 waves add is the mature form: a 2020s mean of 0.443, nearly double the 2000s value of 0.234 that the original debate was fought over.

03Abortion learned its party

No issue illustrates the re-bundling like abortion. Pooling the 1980–1988 surveys — 9,129 respondents — the weighted correlation between abortion attitudes and 7-point party ID was −0.005. Literally nothing. Pro-life Democrats and pro-choice Republicans were unremarkable, and abortion views were orthogonal-to-negatively related to economic conservatism: −0.047 with the guaranteed-jobs scale, 0.016 with services and spending, 0.075 with defense.

By the 2020s, abortion correlates 0.469 with party ID and has been annexed into a cross-domain package: 0.362 with services and spending, 0.353 with guaranteed jobs, 0.327 with defense spending. Its average correlation with the economic scales runs −0.017 in 1980, 0.034 in 2000, 0.148 in 2012, 0.227 in 2016, and 0.384 in 2024 — the steepest cross-domain fusion of any issue in the battery, with 58.9% of the total 1980–2024 movement coming after 2012. Abortion did not merely polarize; views on a constitutional and moral question became predictive of views on the welfare state and the military.

04Sorted more than converted

So did a hundred million Americans build genuine ideologies? Here the honest answer is: only partly. The national correlation mixes two things — coherence inside individual heads, and separation between two increasingly distant camps. Recompute the same 15-pair average within each party and the number collapses: 0.251 among Democrats and leaners, 0.242 among Republicans and leaners, against 0.443 for the country as a whole. The gap between the national line and the within-party lines — an upper-bound estimate, since splitting by party also restricts the range of the scales — reaches 44.5% of the national figure in the 2020s.

That is the part of Baldassarri and Gelman’s mechanism that survives: parties did much of the bundling. But the amendment matters as much as the confirmation. The within-party floor itself has risen sharply. Republicans in the 1980s had a mean within-camp issue correlation of just 0.103 — essentially Converse’s unconstrained public, wearing red — and Democrats sat at 0.165. By the 2020s both camps are near 0.25, roughly double. Genuine belief-system tightening is happening inside the parties too, especially after 2012; it is simply smaller than the headline national number makes it look.

None of those caveats rescues the null result. The era in which an American’s opinions could be treated as a loose bag of unrelated commitments — the era Converse described and Baldassarri and Gelman certified as still in progress through 2004 — is measurably over. The parties sorted first. The belief systems followed.