Why some democracies fail in ten years while others last centuries

Between 1950 and 1990, across a dataset covering 135 countries, not a single democracy with a per capita income above $6,055 collapsed. Zero. The figure — measured in 1985 purchasing-power-adjusted dollars — comes from Adam Przeworski and his collaborators’ Democracy and Development, the most comprehensive empirical study of democratic survival ever conducted. The threshold is defined by its boundary case: Argentina’s income in 1975, one year before a military coup against Isabel Perón ended its democracy at the richest level of any breakdown in the dataset.

The finding should be reassuring. It isn’t. It tells you where democracies were safe within a specific forty-year window. It says nothing about what the money was actually purchasing — the mechanism underneath the correlation. And it was measured through 1990, before three decades of documented democratic erosion in wealthier states complicated what “breakdown” even means. The gap between that clean statistical fact and both of its problems — mechanism and scope — is where the interesting political science lives.

The poverty trap

Przeworski, Alvarez, Cheibub, and Limongi’s dataset is blunt in what it shows. Below the threshold, democracies died constantly — 39 of 69 poorer democracies collapsed during the study period. Above it, 32 democracies accumulated 736 country-years of survival without a single failure. The correlation is not subtle. It is among the strongest in all of comparative politics.

But correlation without mechanism is just a description wearing a lab coat. What is the money actually buying?

Przeworski’s proposed answer is not about civic virtue, democratic culture, or enlightened citizens. It’s about costs. In wealthy societies, the capital stock is large, interconnected, and dependent on institutional stability. A coup doesn’t just change who governs — it destroys the legal framework that protects contracts, investments, and property. The richer a society, the more actors have too much to lose from that destruction.

The military officer in a rich country isn’t inherently more loyal to constitutional order than his counterpart in a poor one. He faces a different calculation: seizing power means cratering the stock market, triggering capital flight, destroying the legal infrastructure that makes the economy function. In a poor agrarian society, the productive economy can survive a change of government. In a complex industrial one, it probably can’t. The coup calculus doesn’t shift because wealthy citizens are better democrats. It shifts because breakdown is prohibitively expensive.

Strip away the cost structure — through economic crisis, hyperinflation, the collapse of the productive economy — and the threshold means nothing. Argentina in 1975 sat precisely at that boundary, and when the economic chaos of Isabel Perón’s final year eroded the cost calculation, the military moved. The money hadn’t bought commitment. It had bought a temporary equilibrium. When the equilibrium broke, nothing underneath it held.

What counts as a democracy?

Przeworski's threshold finding depends entirely on how you define "democracy" and "breakdown." His dataset uses a dichotomous classification: a country either is or isn't a democracy in a given year. Freedom House uses a three-tier system — Free, Partly Free, Not Free. V-Dem uses continuous scoring on multiple dimensions. These aren't just academic quibbles. Hungary after 2019 looks like an exception to the wealth threshold or not depending entirely on which classification you use. Freedom House downgraded it to "Partly Free" in 2019. V-Dem's Liberal Democracy Index shows sustained decline from 2010. Przeworski's dichotomous coding would still classify it as a democracy — elections continued, parties competed. The same country, the same decade, three different verdicts. The threshold's power comes from its precision. That precision rests on a definitional choice that is less settled than the finding makes it appear.

The dataset also has a temporal boundary that matters more the further we move from it. Przeworski’s study ends in 1990. Subsequent work — notably Brownlee and Miao’s 2022 extension in the Journal of Democracy and the Boix-Miller-Rosato dataset covering 222 countries through 2020 — broadly confirms the wealth-survival pattern. But the increasing documentation of democratic backsliding in wealthier states has forced a question the original study didn’t face: does gradual erosion through legal means count as breakdown? Hungary post-2010, Turkey post-2016, suggest the threshold remains remarkably robust against classic coups but says nothing about a different species of democratic death — the slow, legal kind.

The threshold tells you where democracies are safe from tanks. It doesn’t tell you what makes them safe from their own elected governments.

Not just the money

Above the wealth threshold, or among countries at similar income levels, what else separates the democracies that survive from those that don’t? The institutional design literature has an answer — and its internal contradictions are more illuminating than its conclusions.

Juan Linz’s 1990 essay “The Perils of Presidentialism” made the case with three structural arguments. First, rigidity: a president serves a fixed term, and short of impeachment or a coup, there is no constitutional mechanism to remove a failing executive. Parliamentary systems have the vote of no confidence — an exit valve that lets the system adjust without rupturing. Second, dual legitimacy: when a president and legislature both claim separate electoral mandates and disagree, there is no constitutional mechanism for resolution. The system deadlocks, and deadlock in a young democracy invites extra-constitutional solutions. Third, zero-sum elections: the presidency is winner-take-all. Every election is existential. Losers get nothing, which gives them powerful incentives to refuse the result.

Linz observed that virtually every long-lived stable democracy was parliamentary. Britain, the Scandinavian countries, the Netherlands, Belgium, the older Commonwealth democracies — all parliamentary. The United States was the conspicuous exception — and he treated it as exactly that. An exception, not a refutation. A single case, however successful, that proved the system could work under specific conditions, not that it would work generally.

Then José Antonio Cheibub took the thesis apart. His 2007 study covering 199 countries from 1946 to 2002 argued that presidential democracies fail more often not because of their institutional design but because of where presidentialism happens to exist — predominantly in Latin America and sub-Saharan Africa, regions that are poorer, more volatile, and more recently democratised than the Western European countries where parliamentary systems dominate. Control for income and regional context, and the survival-rate gap narrows substantially. The higher failure rate isn’t an institutional cause. It’s a selection-bias artefact.

The impasse between Linz and Cheibub is not a problem for the argument. It is the argument. The same institution — presidentialism — produced stable democracy in the United States for more than two centuries and collapse in Chile after three years. If the institution were doing the explanatory work, that couldn’t happen. Something invisible in the formal constitutional text must be doing the actual work.

One structural note: federal systems provide additional veto points that can slow institutional capture. Hungary — unitary, parliamentary — demonstrated how rapidly a supermajority government can reshape institutions when no subnational level resists. But the relationship between federalism and democratic durability is genuinely ambiguous in the comparative literature; federal systems also produce regional autocrats and governance paralysis. The evidence doesn’t support pressing the claim further than this observation.

The measurement problem

Linz and Cheibub don't merely disagree about presidentialism. They disagree about how to count democratic death. Linz's analysis relies on historical comparison and structural reasoning; Cheibub's relies on statistical controls across a large-N dataset with specific coding rules for what constitutes a democratic breakdown. Different coding decisions — when exactly does a democracy end? does gradual erosion count? — produce different survival rates, which produce different conclusions about institutional effects. The methodological dispute is not peripheral to the substantive one. It is the reason the substantive dispute cannot be resolved by simply looking at more data.

If it isn’t the formal institution, what is it?

The rules beneath the rules

Steven Levitsky and Daniel Ziblatt’s answer, developed in How Democracies Die, identifies two informal norms that do the work formal constitutions cannot.

The first is mutual toleration: competing parties recognise each other as legitimate rivals with an equal right to govern. Not enemies. Not existential threats. Not traitors. Opponents you’d rather beat but can live with losing to.

The second is forbearance: the restraint politicians exercise in not deploying their institutional powers to their fullest legal extent. A governing majority with the votes to pack a court, gerrymander districts, defund opposition media, or appoint loyalists to every oversight body — and chooses not to. Not because it can’t. Because the norm says you don’t.

What makes both norms structurally distinctive is that they cannot be enforced. They’re not written down. Courts can’t compel them. No formal sanction follows their violation. They hold because every actor expects every other actor to hold them — a coordination equilibrium that works precisely until one side defects and forces the other to choose between unilateral compliance and response. The moment the first defection goes unpunished, the equilibrium shifts. What was unthinkable becomes merely aggressive. What was aggressive becomes normal.

The historical depth here is uncomfortable. In Britain, norms of parliamentary restraint accumulated over centuries of negotiated conflict between crown and parliament — an accommodation that calcified into professional habit precisely because neither side ever won decisively enough to simply dictate terms. The monarch couldn’t abolish Parliament; Parliament couldn’t quite abolish the monarchy. So they developed habits of coexistence that eventually became indistinguishable from principle. But the process took centuries, not decades, and it required a specific structural condition: stalemate.

In the United States, Levitsky and Ziblatt make the observation that most accounts of American democratic stability prefer to skip: the normative consensus held, in significant part, because it excluded Black Americans from political participation. The most politically volatile conflicts in the country were kept outside the system. Democratic norms weren’t tested on the hardest cases because the hardest cases weren’t admitted to the democracy. Bipartisan civility in Congress was achievable, in part, because both parties agreed on what didn’t need to be discussed. When the Civil Rights Act and Voting Rights Act finally democratised the country fully, they also polarised it — posing the greatest challenge to mutual toleration since Reconstruction.

Countries with shorter democratic histories haven’t had time to calcify these habits. That’s a structural disadvantage, not a cultural one. Nobody is born with forbearance. It forms through repeated practice over decades, sometimes centuries. And a norm that took 150 years to establish can be seriously tested in a single electoral cycle. Once tested without consequence, the test becomes the new baseline.

The asymmetry is the structural reason young democracies are more fragile. Not because their citizens value democracy less. Because the habits haven’t had time to form.

The pivotal moment

All three variables — wealth, institutions, norms — converge most dangerously at one specific point in a democracy’s life. Not its founding. Its first serious electoral defeat for an incumbent.

Przeworski’s research on democratic alternation, drawing on approximately 3,000 elections since 1788, shows that the average democracy takes nine elections before the first peaceful transfer of power from a governing party to an opposition party. Nine. Once the first transfer occurs, the second takes an average of 3.4 elections. The third takes 2.3. Habit, once established, accelerates faster and more durably than any constitutional provision could mandate.

But that first time is structurally unlike any other moment in a democracy’s life. An incumbent facing defeat for the first time cannot know whether the incoming opposition will comply with democratic norms once in power. The opposition cannot know whether the incumbent will actually yield. Neither side has evidence to rely on — only the formal rules, and formal rules alone have never been sufficient anywhere. The coordination game has never been played before. Both sides are guessing.

Think about it from the incumbent’s position. You have controlled the state apparatus — the police, the military, the intelligence services, the treasury. You are being asked to hand all of that to people you may genuinely believe will abuse it. Every norm of forbearance is untested. Every promise of mutual toleration is unproven. The formal constitution says you must go. But constitutions are pieces of paper enforced by the same institutions you currently control. The temptation to find a reason to stay — to declare the election fraudulent, to invoke emergency powers, to allow the military to intervene on your behalf — is not a character defect. It is a structural feature of the situation.

Democratic systems that make political defeat survivable — that insulate former officeholders from prosecution, preserve opposition party resources, prevent the winner from controlling the judiciary and media apparatus — are structurally more stable at this juncture. They lower the stakes of losing. Systems that make defeat total — where the winner controls prosecutorial machinery, the media landscape, judicial appointments — raise those stakes to the point where defection from democratic norms becomes rational even for actors who genuinely believe in them. The question isn’t whether the incumbent is a good person. The question is whether the system makes yielding power cheaper than holding it.

Founding periods, for all their drama, are actually safer. The actors involved have just collectively decided to try democracy and have incentives to demonstrate its functionality. The dangerous moment comes later. When habit hasn’t formed. When formal rules are untested. When one side is about to discover what losing actually means.

Three deaths

Three democracies died — each with a different configuration of the variables, each through a different mechanism. Two of them the framework predicts cleanly. The third breaks it.

Weimar Germany. The wealth variable predicted vulnerability correctly. By 1932, the Great Depression had driven registered unemployment to approximately 30 percent of the workforce — nearly six million people. Germany was well below any plausible version of Przeworski’s threshold. The economic catastrophe didn’t just impoverish the country; it destroyed the cost structure that makes democratic breakdown prohibitive. When the productive economy is already in ruins, a coup can’t make things much worse. The calculus flips.

The institutional failure was Article 48, the emergency-decree provision that was supposed to be exceptional. President Hindenburg invoked it sixty times in 1932 alone. By the time Hitler was appointed chancellor on January 30, 1933, governing by emergency decree had displaced parliamentary legislation as the routine mode of government. The formal exception had become the actual system. The Reichstag had effectively stopped governing Germany before Hitler arrived to finish the job.

The Enabling Act, passed March 23, 1933 — 444 votes to 94 — was a constitutional amendment adopted by constitutional supermajority. All 94 no votes came from the Social Democrats. Every other party present voted yes, including the Centre Party, whose leader Ludwig Kaas had obtained personal assurances from Hitler. The mainstream parties that provided the votes are the story, not just the party that used them.

The norm failure was multilateral. Mutual toleration had collapsed across the entire party system, not only among the Nazis. The Centre Party’s vote for the Enabling Act was what Levitsky and Ziblatt’s framework predicts: a rational calculation by actors who had already accepted that opponents were not legitimate rivals but threats. Accommodation seemed survivable. Resistance did not.

Chile 1973. A different configuration entirely. Also below the wealth threshold. Also an institutional failure — but a different mechanism. Salvador Allende was elected in September 1970 with 36.3 percent of the vote in a three-way race. A plurality, not a majority. The presidential system’s fixed term was the structural problem: there was no constitutional mechanism to adjust to his minority status. No vote of no confidence. No coalition arithmetic enforced by the system. No early-election provision.

When the Chamber of Deputies passed its August 22 resolution declaring the government in breach of the constitution — 81 votes to 47 — the resolution lacked the Senate supermajority needed for formal conviction. But it publicly delegitimised the government in terms the military would use seventeen days later to justify intervention. The gap between the resolution’s legal weight and its political function was the mechanism.

The norm failure in Chile was bilateral, and the standard narrative consistently suppresses this. Allende’s coalition pushed constitutional boundaries — land redistribution in advance of legislation, nationalisations beyond what Congress had sanctioned. His opponents refused to accept him as a legitimate president from election day and mobilised economic destabilisation, with documented CIA involvement confirmed in declassified State Department records from the FRUS series. Neither side treated the other as a legitimate rival operating within shared rules. Both sides treated the other as an existential threat that justified extraordinary measures.

Mutual toleration had failed on both sides before the first shot was fired on September 11, 1973. The coup remains what it was — a military seizure that killed thousands, inaugurated seventeen years of dictatorship, and cannot be equated with whatever constitutional violations preceded it. But analytically, Chile shows that democratic collapse rarely requires a single authoritarian villain. It requires a system in which mutual toleration has failed on enough sides simultaneously that the military stops looking like an intervention and starts looking like a settlement.

Hungary post-2010. The most important complicating case. Hungary in 2010 was an EU member state with a per capita income of approximately $13,190 in current dollars — well above Przeworski’s threshold by any reasonable comparison. The framework doesn’t cleanly predict what happened next.

Fidesz won 263 of 386 parliamentary seats in 2010 — a two-thirds constitutional supermajority — with 52.7 percent of the popular vote. A product of Hungary’s electoral system, not a proportional mandate. Under the Hungarian constitution, a two-thirds majority carries the power to amend the constitution itself. Within a year, Fidesz had written a new fundamental law, expanded and packed the Constitutional Court, forced sitting judges into early retirement by changing the mandatory retirement age, restructured electoral districts, and appointed loyalists to electoral oversight bodies. None of these moves was illegal. All were performed within the existing legal framework. By 2014, the gerrymandering had taken effect: Fidesz received 44.9 percent of the vote under the new system and maintained a near-supermajority with 133 of 199 seats.

What failed in Hungary was forbearance. A parliamentary supermajority represents the severest test of that norm, because it makes legal precisely the institutional moves constitutions are supposed to prevent. Levitsky and Ziblatt call it “executive aggrandizement” — not a coup, not a classic breakdown, but the incremental and legal accumulation of institutional advantage by an incumbent facing no normative check on what its majority permits. Freedom House downgraded Hungary from “Free” to “Partly Free” in 2019. Elections continued. Parties competed. But competition occurred on a playing field Fidesz had redesigned.

Hungary forces an extension of the framework. The wealth threshold protects against tanks. It does not protect against a government that captures institutions one legal amendment at a time. The mechanism of death is different — slower, quieter, dressed in constitutional procedure. No single action crosses the line that would trigger the binary coding of “democratic breakdown.” Each individual move is defensible; the cumulative effect is not. And the original dataset, measuring through 1990, never had to account for it because this species of democratic death hadn’t yet been documented at scale.

Three cases, quantified

V-Dem's Liberal Democracy Index scores illuminate what quantitative measures captured and missed. Germany's Weimar Republic scored approximately 0.5-0.6 in the late 1920s — a middling democracy by the index's standards — before collapsing toward zero by 1934. Chile scored higher in the early 1970s, reflecting longer institutional continuity, before its sharp drop in 1973. Hungary's trajectory is different: a gradual decline from approximately 0.7 in 2010 to below 0.4 by 2019, never hitting the sharp discontinuity that characterises coup-based breakdown. The quantitative measures captured Hungary's erosion in real time. They also show why dichotomous classifications — democracy or not — struggle with the case. Hungary's decline was continuous, not binary. The threshold model, built for binary events, sees nothing until the line is crossed. V-Dem saw the slope the entire time. Historical scores for Weimar and Chile use reconstruction methods with acknowledged limitations; Hungary's scores draw on contemporary expert surveys.

The April 2026 Hungarian election — in which Péter Magyar’s Tisza Party won 141 of 199 seats and Orbán conceded defeat after sixteen years — suggests democratic erosion is not always irreversible. But the mechanisms of erosion themselves are the case study. Whether one election can undo a decade of institutional capture is a question Hungary is now testing in real time.

What the data cannot answer

Return to that $6,055 threshold — the clean, reassuring number that promised to tell us where democracies are safe. By now, what the money is actually buying should be clear: not democratic commitment or constitutional perfection, but three interacting conditions. Development above the level that makes breakdown prohibitively costly. Institutional design that genuinely constrains power rather than merely distributing it on paper. And normative habits — forbearance, mutual toleration — that have survived enough stress tests to calcify into something approaching professional reflex.

When all three hold and reinforce each other, democracies tend to survive. Not because their citizens are vigilant or virtuous, but because the structural conditions make holding together cheaper than falling apart. When those conditions degrade, what follows is a script the cases make legible. Weimar: all three failed simultaneously under economic catastrophe. Chile: institutional rigidity and bilateral norm collapse below the wealth threshold. Hungary: forbearance eroded legally, above the wealth threshold, through mechanisms the original framework never anticipated.

The framework tells you where democracies are structurally safe. It identifies the specific conditions doing the work, and the specific moment at which those conditions face their severest test. What it cannot tell you — what no one’s dataset answers — is whether those conditions can be rebuilt once they’ve eroded.

There is a growing literature on democratic recovery. V-Dem tracks autocratization reversals; roughly half of all documented backsliding episodes since 1900 eventually turn around. But that work measures regime-level outcomes — scores going up or down. It cannot answer the harder question underneath: whether norms of forbearance and mutual toleration, once seriously damaged, can be deliberately reconstructed, or only grown again over generational time through the kind of repeated practice Przeworski’s alternation data describes. The asymmetry between formation and erosion is documented. The path back is not.

The data does not pretend otherwise. Neither should we.

Gen AI Disclaimer

Some contents of this page were generated and/or edited with the help of a Generative AI.

Media

John Trumbull ‘s painting, Declaration of Independence , depicting the five-man drafting committee of the Declaration of Independence presenting their work to the Congress. The painting can be found on the back of the U.S. $2 bill . The original hangs in the US Capitol rotunda . It does not represent a real ceremony; the characters portrayed were never in the same room at the same time. – Wikipedia

Key Sources and References

Adam Przeworski, Michael E. Alvarez, José Antonio Cheibub, and Fernando Limongi, Democracy and Development: Political Institutions and Well-Being in the World, 1950–1990, Cambridge University Press, 2000.

Adam Przeworski, “Acquiring the Habit of Changing Governments Through Elections,” Comparative Political Studies 48(1), 2015, pp. 101-129.

Juan J. Linz, “The Perils of Presidentialism,” Journal of Democracy, Vol. 1, No. 1, Winter 1990, pp. 51-69.

José Antonio Cheibub, Presidentialism, Parliamentarism, and Democracy, Cambridge University Press, 2007.

Steven Levitsky and Daniel Ziblatt, How Democracies Die, Crown, 2018.

Alfred Stepan, “Federalism and Democracy: Beyond the U.S. Model,” Journal of Democracy, Vol. 10, No. 4, October 1999, pp. 19-34.

Jason Brownlee and Kenny Miao, “Why Democracies Survive,” Journal of Democracy, Vol. 33, No. 4, October 2022, pp. 133-149.

Carles Boix, Michael Miller, and Sebastian Rosato, “A Complete Data Set of Political Regimes, 1800–2007,” Comparative Political Studies 46(12), 2013, pp. 1523-1554. Updated through 2020 via Harvard Dataverse.

Freedom House, Freedom in the World 2019: Hungary Country Report, February 2019.

United States Holocaust Memorial Museum, “Article 48,” Holocaust Encyclopedia.

Foreign Relations of the United States, 1969–1976, Volume XXI, Chile, 1969–1973, compiled by James McElveen and James Siekmeier, United States Department of State, Office of the Historian.

V-Dem Institute, V-Dem Dataset, University of Gothenburg. Available at v-dem.net.

World Bank, GDP per capita (current US$) — Hungary, World Development Indicators.