Ask someone if they support democracy and almost everyone says yes. That’s not a soft finding — it’s one of the most stable results in political science, replicated across countries, decades, and survey houses. Put the same people in front of two candidates in a simulated election, though, and let one of them propose cutting polling stations in opposition areas, ignoring court rulings, and prosecuting journalists, and that abstract love of democracy mostly stops predicting who they pick. That’s the gap at the center of Voting for Democracy? Testing Political-Cultural Constraints on Democratic Backsliding, a candidate-choice experiment run on 5,951 voters across the United States, Poland, and South Korea by Christopher Claassen of the University of Glasgow, published in Public Opinion Quarterly on Sep 27, 2026.

Alexis de Tocqueville would not have been surprised, and he’d have been irritated it took a conjoint experiment to prove something he wrote about in the 1830s. Touring a young United States, he became less interested in whether Americans loved democracy — they plainly did, to the point of obsession — and more interested in what stood between that love and its worst version. His answer wasn’t more enthusiasm for equality. It was the boring machinery around it: a free press, independent courts, town governments people actually showed up to. Strip that machinery away, he argued, and the raw passion for democratic equality doesn’t protect a country from tyranny. It becomes the tyranny, wearing a majority’s face. Claassen’s study is, almost accidentally, a dated empirical test of exactly that distinction.

The Quick Verdict

Question	What we found
What was studied?	Whether specific democratic orientations — support for liberal-democratic institutions, generic democracy-vs-autocracy preference, institutional trust, political tolerance, emancipative values — predict voters choosing a prodemocratic candidate over an antidemocratic one
Who ran it?	Christopher Claassen (University of Glasgow)
Where and when?	Public Opinion Quarterly, published Sep 27, 2026; covered the same day by PsyPost
Sample	5,951 voters — 1,983 in the US, 1,967 in Poland, 2,001 in South Korea — each judging 12 simulated candidate matchups
The antidemocratic proposals	Cutting polling stations in opposition areas, ignoring court rulings, prosecuting journalists
Core finding	Support for liberal-democratic institutions specifically — not a generic preference for democracy over autocracy — was the most consistent predictor of choosing the prodemocratic candidate
The number	A one-standard-deviation increase in institutional commitment moved the prodemocratic vote by 2 to 12 points, depending on country and model
Where partisanship still won	In Poland and South Korea, support for the prodemocratic opposition candidate often stayed below 50% even among voters who scored high on democratic commitment
What Tocqueville already argued	Democracy in America (1835–40): raw love of democratic equality, unmoored from specific mediating institutions, curdles into majority tyranny rather than guarding against it

The Quick Version

A candidate-choice experiment on 5,951 voters in the US, Poland, and South Korea found that generic love of democracy barely predicts whether someone rejects a candidate who proposes suppressing opposition votes, ignoring courts, and jailing journalists. What did predict it — modestly, and inconsistently across countries — was specific commitment to liberal-democratic institutions: free elections, an independent judiciary, equal rights. Even then, partisan loyalty usually won anyway. Tocqueville argued in the 1830s that the passion for democratic equality needs mediating institutions to keep it from turning into majority tyranny. This study is a 2026 test of that exact claim, and the institutions mostly held — barely, and only sometimes.

What the Study Actually Found

The design wasn’t a poll asking people whether they value democracy. It was a candidate-choice conjoint experiment: participants in three countries facing real backsliding pressure — the US, Poland, South Korea — saw twelve pairs of hypothetical candidates and picked between them, with the candidates’ positions on democratic norms randomly varied alongside partisan and policy cues. One candidate in each pair, on some trials, proposed cutting polling stations in opposition-leaning areas, ignoring adverse court rulings, or prosecuting journalists who criticized the government. The question was simple to ask and hard to answer well: does anything about a voter’s prior commitments predict whether they punish that candidate for it, especially when the candidate shares their party?

Claassen tested six separate measures of political culture against that question — generic support for democracy over autocracy, trust in political institutions, political tolerance, emancipative values, and, distinct from all of those, support for the specific institutions of liberal democracy: free and fair elections, judicial independence, press freedom, equal political rights. Only one of the six held up with any consistency. Generic democratic preference — the “yes, I support democracy” answer nearly everyone gives — showed weak and inconsistent relationships with resisting antidemocratic candidates. Institutional trust barely moved the needle at all. Support for liberal-democratic institutions specifically was the one measure that predicted prodemocratic choice across all three countries, and it kept predicting it even in what Claassen calls cross-pressured scenarios — trials where picking the prodemocratic candidate meant voting against your own party or ideology.

That “even then” is doing real work, and it’s worth sitting with the actual size of the effect rather than rounding it up. A one-standard-deviation increase in institutional commitment moved the probability of choosing the prodemocratic candidate somewhere between 2 and 12 percentage points, depending on the country and the specific model. That’s a real, statistically consistent effect. It is not a large one. And in Poland and South Korea specifically, support for the prodemocratic opposition candidate often stayed under 50% even among voters who scored high on institutional commitment — meaning the single strongest predictor Claassen found still lost to party loyalty more often than it won, in two of the three countries tested.

What Did Tocqueville Mean by the Tyranny of the Majority?

Tocqueville’s tyranny of the majority describes what happens when a democratic majority faces no check on its will — not just in law, but in public opinion, which he thought could crush a minority view more completely than any king. His worry wasn’t that Americans loved equality too much. It was that love of equality, alone, generates no internal limit on what a majority feels entitled to do to whoever disagrees with it.

He didn’t leave it there as a warning with no exit. In the chapters on what mitigates the tyranny he’d just described, Tocqueville pointed to specific structures doing specific work: township governments that kept power too dispersed for a national majority to reach into every local decision, a legal profession he called the closest thing America had to an aristocracy — one committed to precedent and procedure over popular mood — and a press so fragmented across thousands of local papers that no single outlet could manufacture a unanimous public opinion. None of these institutions ran on love of democracy. They ran on procedures that didn’t care how popular the majority’s mood was that week.

Why Doesn’t Loving Democracy Stop People from Backing Antidemocratic Candidates?

1. A generic preference for democracy is compatible with almost any specific vote. Nearly everyone endorses “democracy” in principle, which means the preference carries almost no information about what a person will do when a real candidate from their own side threatens a specific democratic norm.
2. Institutional commitment is a narrower, harder-to-fake belief. Valuing judicial independence or press freedom by name forces a person to evaluate a concrete practice, not a feel-good abstraction — which is likely why it was the one measure in Claassen’s study that kept predicting behavior.
3. Institutional trust measures faith in current officeholders, not commitment to the rules. Someone can distrust every sitting judge and journalist and still value judicial independence and a free press as institutions — which may be exactly why trust performed so much worse than institutional support in this study.
4. Partisan identity operates upstream of most other political judgment. The cross-pressured trials existed specifically to isolate cases where party loyalty and democratic principle pointed in different directions — and party loyalty won often enough, in Poland and South Korea, to keep prodemocratic support under 50% even among the institutionally committed.
5. Tocqueville named the gap this study is measuring, without the data to prove it. He argued that democratic sentiment needs mediating institutions to avoid curdling into majority tyranny. This study didn’t test townships or a free press directly — it tested whether valuing those kinds of institutions predicts resisting an antidemocratic candidate. It does. Just not by much, and not reliably enough to bet a country on.

The Gap Between Loving Democracy and Committing to Its Machinery

Here’s where the study quietly sharpens Tocqueville rather than just confirming him, and I think that’s the more useful reading.

Tocqueville wrote as if the mediating institutions did the protective work directly — townships absorbing shocks, a free press fragmenting opinion, courts slowing down popular fury. He wasn’t really writing about what individual voters believed about those institutions. He was writing about whether the institutions existed and functioned. Claassen’s study measures something one step removed: not whether courts are independent, but whether a given voter values judicial independence enough to punish a candidate who threatens it. Related questions, but not the same one — and the gap between them is where the study’s most uncomfortable finding lives.

Valuing the institutions turns out to be necessary but nowhere near sufficient. It’s the strongest lever Claassen found among six, and it still lost to party loyalty most of the time in two of the three countries he tested. Tocqueville assumed — reasonably, for a man watching functioning townships and a genuinely chaotic free press in the 1830s — that if the institutions existed and people valued them, the system would hold. This study suggests a voter can value judicial independence sincerely, in a survey, and still hand power to a candidate who proposes ignoring courts, if that candidate is wearing the right jersey. The institutions Tocqueville trusted to check the majority’s mood have to survive contact with a mood he underestimated: partisan loyalty strong enough to override a value the voter actually holds.

Why the Study and the Philosophy Rhyme

The connection here isn’t “a nineteenth-century Frenchman predicted a 2026 conjoint experiment.” It’s narrower, and it holds up better for being narrower: Tocqueville’s specific claim was that generic democratic sentiment needs particular institutional commitments to avoid becoming majority tyranny. This study tested six candidate beliefs against real candidate choices and found that the generic sentiment barely worked and the specific institutional commitment worked better — which is close to a direct vindication of the distinction he was drawing, a century and a half before anyone had the data to isolate it.

It sits close to what James’s theory of emotion, tested against a 2026 disgust study, found about belief that never gets examined — that a reaction can shape political identity without passing through the kind of conscious appraisal that would let a person catch it doing so. Generic democratic preference looks like that kind of unexamined belief: something almost everyone holds and almost no one interrogates until a specific candidate forces the question. Institutional commitment requires the interrogation to have already happened.

It also connects to Noelle-Neumann’s argument that silence isn’t neutral — that people read which opinion is winning inside their own group and go quiet about the one that isn’t. Partisan loyalty overriding institutional commitment looks like that same pressure operating on a vote instead of a comment: a voter who privately values judicial independence may simply not want to be the one in their coalition who broke ranks over it, especially when the candidate threatening it is otherwise on their side.

There’s a useful parallel, too, in Durkheim’s argument that shared anger alone doesn’t organize a street into a movement — that raw collective sentiment needs structure to become durable collective action. Tocqueville was making the same claim about democratic sentiment specifically: love of equality, alone, is raw material, not a safeguard. It needs the equivalent of Durkheim’s structure — townships, a free press, independent courts — to hold a majority’s worst impulses in check.

And it’s worth reading against Habermas’s case for genuine public deliberation, which assumes reasoned exchange between citizens legitimizes democratic outcomes. Habermas’s account has an uncomfortable dependency this study exposes: deliberation about democratic norms only checks backsliding if people let its conclusions override their partisan attachments when the two conflict. Claassen’s cross-pressured trials measured exactly that willingness, and found it was often absent.

What This Study Doesn’t Show

It’s worth being precise about the size of this, because the framing invites more confidence than the data earns.

This is a candidate-choice conjoint experiment, which means the candidates’ proposals were genuinely randomized — that part supports real causal claims about how specific proposals affected vote choice. But the six measures of democratic orientation were not randomized. Nobody manipulated anyone’s commitment to liberal-democratic institutions and watched their voting shift in response; those relationships are correlational, observed within the same survey. And three countries, however well chosen for facing real backsliding pressure, are three countries. The effect sizes — 2 to 12 points per standard deviation — vary enough across them that a single average number would flatten something the study itself found to be inconsistent.

It also isn’t a finding that institutions or elections don’t matter, or that belief is worthless. Institutional commitment was the one measure that held up, across contexts, under real partisan pressure — a meaningfully positive finding on its own. The uncomfortable part isn’t that nothing works. It’s that the thing that works, works modestly, and loses to partisanship more often than it beats it.

And it says nothing about what happens outside a simulated election. A hypothetical candidate matchup in a survey isn’t a real primary or a real choice made under the weight of a candidate’s full record, media coverage, and social pressure from friends and family. This study measured what happened when democratic commitment met partisan pull in a controlled, twelve-trial experiment. Real elections have more inputs than that, in both directions.

What Actually Follows From This

1. A “yes, I support democracy” answer is close to meaningless as a predictor of behavior — check the specifics instead. The useful question isn’t “do you value democracy,” it’s “do you value this specific court, this specific press freedom, this specific election rule” — that’s the belief that did the work in this study.
2. Notice when your own institutional commitments are losing to your partisan ones. The cross-pressured trials isolated exactly the moment worth watching for in yourself: does your support for judicial independence survive contact with a candidate from your own side threatening it, or does it quietly get outvoted by loyalty before you notice the trade?
3. Institutions built to survive bad moods matter more than beliefs built to survive good ones. Tocqueville trusted townships and courts because they didn’t depend on anyone’s momentary enthusiasm for democracy. This study is a reminder that individual belief, even sincere belief, is a weaker safeguard than a functioning institution — an argument for protecting the institutions directly, not just for holding better opinions about them.

Tocqueville never ran an experiment and never had a sample size. He had a notebook, a nine-month tour, and an argument he was confident enough in to build two full chapters around. What Claassen’s study offers, nearly two centuries later, is the data Tocqueville didn’t have and the distinction he was right to insist on anyway: loving democracy in the abstract protects almost nothing. Valuing its specific machinery protects a little more. And even that, this study found, is still fighting an uphill battle against whichever side a voter is already on.

---

This post discusses a 2026 candidate-choice conjoint experiment published in Public Opinion Quarterly and an 1835–40 political text. The study’s causal claims apply to the randomized candidate proposals within the experiment; the relationships between democratic orientations and vote choice are observed correlations within that same experiment, not independently manipulated ones. Findings come from 5,951 voters across the US, Poland, and South Korea and describe patterns in a simulated election, not a universal rule for every democracy or every real-world vote.