Political polarization in western societies

Here is what I think is the most important thing to say before we get into the weeds: the feeling that democracy is breaking is not an illusion, but the diagnosis most people reach for is probably wrong. We have done our research on this one, and the facts are genuinely surprising. Here is what the evidence actually shows. Affective polarization — the raw emotional hostility between partisan camps — has seen a five-fold increase in the United States between two thousand and twenty twenty, measured by the share of partisans giving the opposing party the coldest possible rating on a standard survey scale. That is not a media narrative. That is a measurement. And yet polarization in the United States was rising steadily from nineteen eighty, well before the internet existed, and it is highest among Americans over seventy-five — the cohort least likely to be on social media at all. So the social media explanation, which dominates public conversation, runs directly into those two facts. That does not mean platforms are innocent. It means the story is structurally more complicated than the headline version. The research maps three serious competing explanations. First: social media as amplifier of pre-existing fractures, not their cause. Second: long structural cycles driven by economic inequality, elite incentives, and institutional decay going back to the nineteen eighties. Third: what several non-Western political theorists are now calling institutional atrophy — the erosion of the shared meaning-making frameworks that historically converted conflict into collective renewal, before it curdled into pure tribal hostility. Those three camps disagree not just about mechanism but about timeframe and about what kind of problem this actually is. So here is the question I want to put to the panel: if the social media thesis is insufficient — and I think the evidence forces us to at least seriously entertain that — what is the load-bearing structural explanation, and does it have a solution?

The age inversion puzzle isn’t just a curiosity—it’s a structural X-ray. The 75+ cohort didn’t grow up with social media, but they did grow up with local party machines, union halls, and church basements that functioned as conflict-conversion mechanisms. Those institutions didn’t just mediate disagreement; they ritualized it, turning raw division into negotiable grievance. Their collapse left a vacuum, and what we’re seeing now is the sound of that vacuum being filled—not by platforms, but by the absence of anything else to absorb the energy of disagreement. The question isn’t why older Americans are polarized; it’s why we stopped building rooms where disagreement could be resolved before it became identity warfare.

The cross-national pattern forces a sharper distinction than the platform story allows. Internet uptake tracked closely across Western democracies after two thousand, yet affective hostility rose sharply only in the U.S., UK, and Canada while remaining flat or declining elsewhere. That selective effect survives uniform technological conditions, which means the variable doing the work is whatever each system retained or lost in its capacity to route disagreement through arenas that could impose negotiated costs on elites and voters alike. The rooms that once performed that function are not evenly distributed, and their absence shows up in the data before any algorithm does.

The deeper question is sequence. We still don’t know whether voters’ affective hostility drives elites toward extremes or whether elite incentives reconfigure voter emotion. That is not an academic quibble; it determines which system you rebuild. If politicians polarize first, the fix is institutional—change primary rules or coalition mechanics so incentives reward compromise. If emotion polarizes first, the remedy is cultural—rebuild spaces that humanize the other side. At the moment, both coils seem to tighten together. Until we can disentangle who moves first, every proposed cure is just pushing on one end of a feedback loop without knowing which direction the current runs.

The cross-country divergence is where the platform thesis frays. Internet adoption tracked identically across Western democracies, yet raw partisan dislike spiked in the United States while holding steady elsewhere. Rwanda saw internet usage grow four hundred twenty percent between twenty fifteen and twenty twenty three, yet only twelve percent report deep divisions, versus sixty two percent in the United States. I read that difference as institutional: whether a system maintains explicit arenas to route conflict into negotiated outcomes. Yogendra Yadav’s mobilization versus polarization distinction clarifies this. What looks like pathological division may be democratic energy with nowhere to go. That explains why information remedies stall. Social psychology tracks a specialist-known pattern called the numeracy backfire effect, where analytically skilled citizens harden their views when shown corrective data, using that capacity to rationalize existing loyalties. Platforms did not break the civic architecture. They filled the vacancy.

The numeracy backfire effect deserves more weight than it's getting here, because it isn't just a curiosity about smart people being stubborn. It is a structural indictment of the entire policy logic driving most platform reform proposals. Fact-checking, media literacy curricula, algorithmic transparency — all of these rest on an information-deficit model: citizens are polarized because they lack accurate data, and supplying it will help. The social psychology literature suggests that for analytically capable citizens, the opposite may be true. They deploy that capacity to defend existing loyalties more effectively. That doesn't make information remedies worthless, but it does mean the mechanism assumed by most reform proposals may be genuinely broken — and we should be honest that the evidence on this remains contested rather than pretending we've solved it.

The false polarization literature reveals something deeper than misperception—it exposes a structural flaw in how democratic disagreement is architecturally processed. When partisans consistently overestimate their opponents' extremism, they're not just wrong about the facts; they're responding to an institutional environment where disagreement has no productive outlet. The feeling-thermometer data shows this isn't about ideology—it's about the absence of mechanisms that convert raw conflict into negotiable outcomes. Rwanda's anti-divisionism law doesn't suppress disagreement; it channels it through Gacaca courts where resolution is mandatory before it becomes identity warfare. That's not about information—it's about creating spaces where disagreement is structurally impossible to avoid resolving. The U.S. has no equivalent. Every political dispute defaults to either litigation (which is adversarial by design) or protest (which is performative by necessity). That architectural absence is the load-bearing fact, not the content of the disagreement.

The absence of any federal conflict de-escalation architecture is not a neutral default. It is a policy choice that raises the transaction cost of every disagreement until only litigation or public performance remain rational moves. Ghana's District Peace Councils resolved ninety four percent of pre-election disputes in twenty twenty before they reached courts. The United States has no equivalent mechanism, so the same underlying cleavages generate higher returns to escalation. That difference in payoff structure, not the volume of speech, determines whether mobilization settles or simply circulates.

The real structural question may be why high-information citizens polarize faster. The numeracy backfire effect implies that polarization scales with analytical capacity, not ignorance. That means it’s not just about what people believe but how the system rewards belief defense. In a winner-takes-all political architecture, every numerical correction feels like a reputational loss, so the cognitively skilled deploy statistics as armor rather than evidence. That could make polarization an equilibrium outcome of adversarial design, not a flaw in citizen rationality. If that’s right, the core variable isn’t the content of digital feeds—it’s the absence of procedural incentives to treat correction as cooperative rather than competitive.

We are debating architecture while ignoring the causal hinge that determines which repairs actually work. The evidence map flags a critical gap: no longitudinal data establishes whether elite cuing precedes mass affective polarization or vice versa. This isn't a minor footnote. If primary incentives radicalized politicians first, then voter hostility is a lagging indicator, and cultural interventions are structurally doomed. A Han Fei reading of statecraft would warn that disorder flows from misaligned rewards at the top, never from the masses. Until we resolve this sequence, we are prescribing remedies for a mechanism we haven't isolated.

The elite-versus-grassroots sequencing problem is real, but I want to flag what it takes for granted. Both sides of that debate assume polarization is something happening to citizens — whether from above or below. Mutombo Nkulu-N'Sengha's challenge is more fundamental: he argues Western societies have misidentified what they're measuring. The noise we're analyzing as polarization may be what democratic energy sounds like when the covenantal frameworks that once gave it direction have dissolved. That reframe matters because it shifts the question from who moved first to what disappeared — and the answer to that second question points somewhere neither platform regulation nor primary reform can reach.

The cross-country divergence isn't just about whether institutions exist—it's about what kind of conflict they're designed to handle. Germany's constructive no-confidence rule doesn't prevent political disagreement; it forces the system to treat government collapse as a collective failure rather than an opposition victory. That single procedural difference changes the entire incentive structure: parties can't weaponize instability because they're on the hook to propose a replacement. The U.S. system, by contrast, contains no equivalent mechanism. Every dispute defaults to either a legal battlefield (where adversarialism is the point) or a performative one (where outrage generates attention). That structural difference explains why identical internet adoption produces different polarization outcomes—the variable isn't technology, but whether the architecture assumes conflict is something to be resolved or something to be exploited.

The numeracy backfire effect implies that any institutional redesign still relying on updated beliefs among high-capacity citizens will widen gaps rather than close them. Those actors already deploy stronger reasoning to protect priors, so new procedures get treated as fresh terrain for advantage-seeking. The brief's cross-national cases succeed where the rules change the immediate cost of non-settlement itself, not where they improve the quality of information. Without longitudinal evidence on whether elite or mass affective shifts lead, that design choice stays an untested assumption rather than a settled mechanism.

There’s a missing institutional dimension we haven’t touched: time. Most of the systems we’re comparing—Germany’s coalition rules, Brazil’s participatory budgeting, Rwanda’s Gacaca—bake delay into conflict resolution. They force opposing actors to stay in the room long enough for preferences to shift. The U.S. system does the reverse: primaries shorten the political time horizon, social media shortens the communicative one, litigation compresses negotiation into a single verdict. When every channel rewards speed over settlement, polarization isn’t ideological—it’s temporal. Rebuilding de-escalation architecture may start with rebuilding patience into procedure.

We are treating institutional design as a purely procedural puzzle, but rules only hold when the social substrate can bear their weight. Pratap Bhanu Mehta’s research shows inter-religious trust in rural India actually rose eleven percent between twenty fourteen and twenty twenty three, even as digital nationalism surged. That resilience did not come from constitutional clauses or feed adjustments. It came from dense, cross-cutting civic ties that make outright rupture socially and economically costly. The United States did not just lose procedural patience; it systematically hollowed out the local associations, labor networks, and municipal forums that once absorbed grievance before it reached the ballot box. You cannot legislate deliberation into a vacuum. If formal architecture is the frame, offline social infrastructure is the load-bearing wall. Platform regulation misses this because it treats a collapsed civic commons as a content problem.

Qwen names the collapsed civic commons as load-bearing, and that's right — but there's a prior question about what kind of civic infrastructure actually does the work. The false polarization literature offers a clue here that hasn't been connected to the institutional argument yet. Ordinary partisans dramatically overestimate how extreme opposing voters are. That perceptual distortion doesn't come from nowhere — it comes from the absence of the mundane cross-cutting encounters where you discover your neighbor holds three of your positions and one you find incomprehensible. That kind of friction — not information, not procedure — is what corrects the phantom. When local associations, mixed workplaces, and civic forums thin out, the phantom gets no correction, and the war against it intensifies. The architecture question and the false-polarization question turn out to be the same question.

The false polarization finding isn't just about misperception—it reveals something about how institutional grammar shapes what counts as a political claim in the first place. When partisans overestimate their opponents' extremism, they're not just wrong about positions; they're responding to an environment where disagreement has no shared language for resolution. Rwanda's anti-divisionism framework doesn't eliminate conflict—it forces political claims into registers where ethnic mobilization becomes legally risky. That's not about suppressing speech; it's about making certain forms of conflict structurally harder to articulate. The U.S. system has no equivalent grammatical constraint. Every dispute defaults to either litigation (where the rules reward adversarial framing) or protest (where the medium rewards outrage). That difference in institutional grammar—not the content of the disagreement—determines whether conflict becomes a negotiable difference or an identity war. The platforms didn't create this vacuum; they just amplified what was already structurally unviable to say in public.

The Science and Nature randomized trials on Facebook's twenty twenty election features directly test whether altering amplification changes affective outcomes. Switching millions of users to chronological feeds produced effect sizes below point zero two on the same feeling-thermometer scales used elsewhere in the literature. That result is hard to square with any mechanism that treats feed design as the primary channel. It instead points to deeper payoff structures that remain unchanged when the platform merely rearranges existing content.

A neglected variable here is scale mismatch. The democratic mechanisms that once converted conflict into governance—party caucuses, city councils, union locals—operated at human scale, where reputational interdependence made compromise rational. Social media and nationalized primaries push every dispute to a scale where that interdependence collapses; the marginal cost of outrage drops to zero. That suggests polarization grows not only from institutional decay but from institutional diseconomies of scale. The plausible policy lever isn’t just rebuilding civic forums but resizing them—restoring decision‑making layers where disagreement carries relational cost and resolution yields tangible reward. Otherwise, every reform remains an arena upgrade for the same game of performative antagonism.

The scale mismatch argument points to something real, but I think it misses the binding mechanism. Local forums only de-escalate when participation is tied to actual resource allocation. The Brazilian Workers Party’s documented experience with participatory budgeting fostered shared citizenship precisely because it routed fiscal decisions through inclusive assemblies. That structure converts moralized grievance into negotiable claims. The cross-national comparison suggests the same pattern: systems with constitutional power-sharing incentives show lower partisan animosity regardless of platform penetration. I read this as evidence that conflict nationalizes and hardens into identity warfare when local channels lose budgetary authority. We are not just missing patience or proximity. We are missing arenas where disagreement carries a direct material stake and settlement releases tangible resources. That causal link remains genuinely contested, but it explains why procedural reforms without fiscal authority keep producing performative antagonism instead of resolution.

The Brazilian participatory budgeting point lands hardest if you notice what it implies about information. The numeracy backfire effect shows that analytical capacity deployed in abstract political argument deepens entrenchment. But citizens in Porto Alegre weren't arguing about identity — they were allocating sewage budgets. The cognitive task was concrete, the feedback was immediate, and defection was visible to neighbors. That's not a better information environment; it's a different kind of reasoning altogether. Which suggests the deepest problem isn't that people reason badly about politics — it's that the institutional forms we've retained ask them to reason about politics in the one register where motivated reasoning is structurally guaranteed to win.

The participatory budgeting cases don’t just show that local fiscal authority matters—they reveal something about how conflict becomes legible to institutions. When citizens in Porto Alegre allocate sewage budgets, they’re not just making decisions; they’re producing a shared record of what counts as a valid grievance. That record becomes the institution’s grammar. The U.S. system lacks any equivalent mechanism for converting diffuse outrage into administrable claims. Instead, every dispute gets funneled into either legal adversarialism (where the rules reward maximal framing) or digital performativity (where the medium rewards moralized identity). The platforms didn’t create this grammar—they just inherited it. The real collapse isn’t of institutions, but of the shared language for what a political demand even looks like. Until that language exists, every reform will be a new arena for the same untranslatable fights.

The Brazilian and Ghanaian cases succeed less because of concrete tasks than because they impose repeated bargaining under resource constraints that participants cannot externalize. When settlement directly unlocks budgeted funds that local actors must live with, the incentive to defect weakens regardless of prior beliefs. National platforms and primaries let actors shift costs to diffuse audiences instead, so the same cognitive capacities that produce numeracy backfire now serve signaling to distant coalitions. Replicating the outcome requires arenas where exit is costlier than compromise, not merely smaller ones.

The material-stakes argument matters because it reframes polarization as a coordination failure, not just an empathy deficit. But that failure emerged from a policy choice: the fiscal centralization of the 1980s stripped municipalities of discretion over roughly sixty percent of their budgets. Once local governments lost the ability to trade resources for consensus, politics reverted to signaling festivals where outcomes were decoupled from bargaining. What needs testing now is whether partial fiscal devolution—say, devolving specific social or infrastructure funds with deliberative conditions—can measurably reduce affective polarization. If it can’t, the institutional thesis falters; if it can, the polarization story becomes a ledger of lost local fiscal capacity rather than moral decay.

Fiscal devolution only works if the rules make concession socially survivable. Material stakes change the calculus, but they do not rewrite the honor code. In systems where compromise is structurally dignified rather than treated as moral capitulation, conflict gets absorbed instead of amplified. Indonesia’s musyawarah tradition, backed by Pancasila, does not erase disagreement; it ritualizes negotiation so that yielding ground preserves standing rather than destroying it. Germany’s constructive no-confidence rule operates on the same logic at the parliamentary level: you cannot tear down a government without simultaneously building the next one, which forces coalition maintenance into the architecture itself. The United States has done the opposite. Primary challenges and litigation frameworks reward maximalist positioning and punish settlement as betrayal. Devolve budgets into that environment and you simply shrink the arena while keeping the zero-sum scoring. The missing variable is not just money; it is institutional design that makes compromise honorable rather than fatal.

The honor-code argument and the fiscal argument are both right, and the tension between them points at something neither fully addresses. Qwen is correct that devolving budgets into a zero-sum honor system just shrinks the arena. But the cases where institutional design actually shifted the honor code — Rwanda, Ghana, Brazil — share a feature that hasn't been named: each involved an explicit, publicly legible moment of institutional founding. Citizens knew a new set of rules had been declared. That declaration itself changed what compromise signaled. The U.S. reform conversation keeps trying to achieve that effect through incremental adjustment — tweaking primaries, adjusting feeds — which never produces the legibility moment. The question worth sitting with is whether incremental reform can ever generate the kind of public grammar shift that the comparative cases suggest requires a visible constitutional or civic rupture.

The comparative cases don’t just show that fiscal devolution works—they reveal how institutional grammar shapes what counts as a valid political claim. In Porto Alegre, participatory budgeting didn’t just allocate funds; it created a public record of trade-offs where grievances had to be framed in terms of measurable outcomes. That’s not about money—it’s about forcing conflict into a register where compromise becomes legible. The U.S. system lacks any equivalent mechanism. Every dispute defaults to either litigation (where the rules reward adversarial framing) or digital outrage (where the medium rewards moralized identity). The platforms didn’t create this vacuum; they just amplified what was already structurally unviable to resolve. The real question isn’t whether to rebuild institutions, but whether we can rebuild them in a grammar that makes settlement feel like progress rather than surrender.

The Yadav framing reveals the underlying incentive landscape: what registers as polarization is actually mobilization without settlement mechanisms, sustained by elite fragmentation that followed neoliberal deregulation of local authority. In that vacuum, platforms supply coordination at zero marginal cost while primaries select for perpetual escalation. Any devolution experiment would need to restore not just budgets but repeated bargaining under visible resource trade-offs that make continued gridlock more expensive for participants than concession. Absent that shift in payoffs, the energy simply finds new outlets.

The comparative cases hint at a deeper operating rule: institutions that channel conflict into bounded sequences—where competition ends before execution begins—sustain pluralism better than those that fuse contest and governance. Germany’s constructive no‑confidence rule makes replacing a government inseparable from forming the next one; Ghana’s Peace Councils separate adjudication from campaigning. That structural sequencing prevents permanent campaigns. In the U.S., primaries and litigation collapse those stages, turning every decision into a continuation of conflict. Polarization, in that sense, reflects institutions that never declare a season over. The repair question becomes procedural, not cultural: how to reintroduce clear hand‑off points between struggle and settlement so politics can resume after the fight.

The age-inversion finding is the quiet hinge here. Survey data shows polarization peaks among the seventy-five-plus cohort, the demographic least likely to use social media. I read that pattern as a generational map of institutional withdrawal rather than a platform anomaly. This cohort aged precisely as cross-cutting civic associations and local party chapters hollowed out. They did not radicalize in algorithmic feeds. They radicalized in the civic silence that replaced those rooms. You do not need a constitutional rupture to shift the grammar of compromise. You need to rebuild the mundane, repeated encounters where disagreement carries a shared material consequence. The comparative cases worked because they forced adversaries into the same space with a ledger. Incremental reform can replicate that mechanism if we stop treating local deliberation as cultural programming and start wiring it into mandatory administrative procedure.

The sharpest tension in this conversation was between two things that both turned out to be true: platforms probably aren't the primary driver, but the structural alternative — collapsed civic infrastructure, missing settlement mechanisms, zero-sum institutional design — is itself genuinely contested. The age-cohort data and the cross-national divergence both complicate the platform story, but they don't resolve it. The concrete takeaway: the numeracy backfire effect means more information won't fix this. The question is what kind of institutional design makes compromise survivable. On the original question — social media as primary cause remains genuinely unresolved. The structural arc is the stronger hypothesis, but it's a hypothesis. Thank you for listening. As it happened; as it is.