Svolik's The Politics of Authoritarian Rule (2012) splits the puzzle in two: the problem of control (how to handle the majority excluded from power) and the problem of power-sharing (how to make elites who rule jointly keep their bargains with each other). The second is harder, because no independent third party can enforce agreements among the elite — the role courts and elections play in a constitutional order. Engineering analogy: a distributed system with no trusted coordinator, where an agreement holds only if it is incentive-compatible.
Svolik's count of dictators who lost power by non-constitutional means between 1946 and 2008: 205 — more than two-thirds — were removed by regime insiders, while only 32 were toppled directly by popular uprising.
Geddes (1999) classified regimes by how elites are organized (averages, 1946–1998):
| Type | Mean duration | Mechanism and cost |
|---|---|---|
| Single-party | ~23 years | The party is both a queue for distributing posts and a channel for information flowing upward; the cost is that it may grow into a rival power center |
| Personalist | ~15 years | Fast decisions, few internal checks; the cost is succession with no rules to fall back on, and information distorted at every layer |
| Military | ~9 years | Shortest-lived, but holds an exit option — returning to barracks leaves the institution intact, making voluntary handover easier |
Mexico's Institutional Revolutionary Party governed continuously for 71 years (1929–2000). Magaloni (2006) argues the key was not banning opposition but combining growth, patronage networks, and overwhelming electoral margins to tie a great many people's interests to the regime's survival.
Institutions matter (Gandhi and Przeworski, Magaloni): legislatures and elections inside authoritarian regimes are not decoration. They let a ruler make credible commitments to elites — concede today, and there is still a channel for your share tomorrow. Without that channel, elites can only cash in through a coup.
The endogeneity sceptics: perhaps institutions do not make regimes stable; perhaps only already-stable regimes can afford institutions. Using the same set of survivors both to define resilience and to explain it invites survivorship bias.
The trade-off: the first yields operational mechanisms at the risk of overstating what institutions independently do; the second is methodologically tighter but offers almost no positive explanation.
"Authoritarian rule is just guns." The most common route out of office is an insider coup, not an uprising, so a better description is "the continuing consent of insiders." A second is reading resilience as "destined to last": when Nathan introduced the term in 2003 he said explicitly that he was not predicting the system would never fail, only explaining why it had not yet collapsed — explaining survival and predicting survival are different tasks.
Wintrobe's dictator's dilemma: the more a ruler relies on coercion, the less public expression reflects what people actually think, and the less the ruler knows how much support exists. Rising uncertainty then drives more coercion — a positive feedback loop, not an equilibrium.
Svolik adds a harder constraint: repression must be outsourced. Only the military or security services can suppress an organized challenge, and whoever holds that capacity is precisely who is best placed to remove the ruler. Coup-proofing (splitting commands, promoting for loyalty rather than competence) is the hedge, at the cost of degrading both repressive and war-fighting capability. Engineering analogy: handing root to a process you cannot audit, then bolting on rate limits — security and availability drop together.
Co-optation scales with the number of people who must be placated, but creates no new armed agent. The selectorate theory of Bueno de Mesquita and colleagues identifies the key variable as the size of the winning coalition: with a small coalition, private goods (posts, licenses, rents) are most efficient; only with a large one does public provision become the better buy.
| Resource base | Practice | Cost |
|---|---|---|
| Resource rents | Ross (2001), using data from 113 countries, proposed the "rentier effect": where revenue comes from resource exports, heavy taxation is unnecessary and citizens' incentive to demand accountability falls | Fiscal capacity tracks commodity cycles; the same rents can also fund an expanded coercive apparatus |
| Patronage networks | Posts, transfers, and local projects bind individual interests to the regime's survival | Requires growth to sustain; when growth stalls, the patronage network turns hostile before the opposition does |
| Coercion-led | Most effective in the short run against an already-organized challenge | Coup risk rises and information quality steadily degrades |
"Repression works": against an intact and loyal coercive apparatus, spontaneous protest rarely succeeds; discussing long-run costs presupposes surviving the short run.
"Repression backfires": it raises both the cost of mobilizing and the grievance driving it, so the net effect is indeterminate. It also leaks information — Lohmann's study of the 1989 Monday demonstrations in Leipzig argues that the authorities' response itself became the signal from which the public inferred the regime's strength.
The trade-off: coercion buys immediate control at the price of agency risk and distorted information; co-optation buys willing compliance at the price of fiscal dependence, and those bought keep raising their price.
Treating the two as either/or. What is usually observed is layering: distribution aimed at elites, targeted coercion at those with organizing capacity, informational tools at everyone else. A second misreading is "co-optation means handing out money": most of it manufactures switching costs — embedding careers, property, and schooling in the existing arrangement.
"Propaganda" collapses three goals with entirely different success criteria: persuasion (changing what you believe), distraction (changing what you think about), and blocking coordination (changing what you assume others will do). The third matters most: collective action requires common knowledge — not only that I am dissatisfied, but that I know you are too, and you know I know. Sever that layer and dissatisfaction can be widespread while action never materializes. Engineering analogy: not tampering with the data, but preventing the nodes from reaching consensus.
Two studies by King, Pan, and Roberts supply testable evidence. Their 2013 article in the American Political Science Review found that censorship of Chinese social media does not target criticism as such, but content with collective-action potential, whether favorable or hostile. Their 2017 article in the same journal estimates that government-linked accounts produce roughly 448 million posts a year, dominated by positive themes and subject-changing rather than argument with critics.
Guriev and Treisman's Spin Dictators (2022) proposes a trend claim: rule based on visible violence is being displaced by rule based on information and procedural legitimacy. On their coding, the "spin" type rises from roughly 13% of non-democratic leaders in the 1970s to roughly 53% in the 2000s.
The transition thesis: manipulation is the better bargain — cheap, deniable, martyr-free, and compatible with integration into the world economy. Miss the shift and you cannot explain why conventional human-rights indicators increasingly fail to capture repression.
The sceptical reply: the dichotomy rests on coding judgments, regimes commonly use both at once, and the shift in proportions may reflect measurement rather than reality; moving the emphasis onto manipulation also risks understating physical coercion that continues.
"The point of propaganda is to make people believe it." Most of these mechanisms do not require your belief, only that you believe others believe, or believe others will not act. The fact that "nobody really buys it" therefore does not weaken its effect. A second misreading equates censorship with deletion: under flooding, not one item is removed — it is buried. Availability is not discoverability.
Vulnerability is not evenly distributed. It concentrates at three structural nodes:
Engineering analogy: a phase transition. Once the observable (public expression) decouples from the state variable (private preference), the stability reading from outside stays entirely normal right up to the moment of collapse.
| Path | Case | Features |
|---|---|---|
| Cascading collapse | Eastern Europe, 1989 | Almost nobody predicted it; each country supplied the others with a demonstration that the cost was bearable |
| Local trigger | Tunisia, 2010–2011 | A leader of 23 years left roughly four weeks after protests began — long stability and rapid ending are not in tension |
| Transition led from within | South Korea and Taiwan, late 1980s | Opened and negotiated by incumbents; slower, but the ruling group could keep competing under the new rules |
Structuralists (in Skocpol's line): collapse is determined by structural conditions — fiscal capacity, elite splits, external pressure. The street is the expression of structure, not its cause.
Coordination theorists (in Kuran's line): threshold cascades make outcomes acutely sensitive to small initial conditions, so identical structures can yield collapse or stability depending on who moves first.
The trade-off: structuralists explain where trouble is likelier but cannot supply timing; coordination accounts explain suddenness at the cost of weak ex ante prediction. The two are really a division of labor: structure gives the probability, coordination gives the moment.
"There are always clear warning signs before a collapse." What preference falsification implies is precisely that the signs are systematically suppressed — observed stability and actual stability are not the same thing. A second misreading equates "the leader fell" with "the system changed": most non-constitutional exits are an internal replacement, with the regime type unchanged.