Democratization by Choice

What might induce authoritarian rulers to surrender or share power? Various arguments suppose they do so deliberately. In one approach, democracy emerges from a bargain, usually between rich and poor. Some see democratic institutions as the enforcement mechanism (Acemoglu and Robinson Reference Acemoglu and Robinson2006). The poor wish to expropriate the rich. Usually, they cannot coordinate, but occasionally some shock such as an economic crisis galvanizes them into revolt. The wealthy would like to coopt the rebels by promising future income. However, both sides know the rich will be able to renege. Establishing a parliament and enfranchising the poor creates a mechanism to enforce such promises, preventing revolution.

This account has several observable implications. Democratization should follow or coincide with anti-elite mobilization—protests, strikes, other mass actions—motivated by economic or redistributive demands. In response, incumbents should incorporate the poor into politics, for instance, giving them voting rights. These reforms should, in turn, increase redistribution from rich to poor and prompt the poor to demobilize. And—since democracy supposedly makes the commitment credible—the rich should not cancel redistribution or reimpose autocracy soon afterwards.Footnote ⁸

Acemoglu and Robinson (Reference Acemoglu and Robinson2006, 38–9) also suggest two variants featuring the middle class. If the threat comes from the middle class—or the poor require its leadership—it may suffice to enfranchise the former, creating “partial democracy.” Alternatively, if redistribution to the middle class also benefits the poor, allowing the middle class to choose policy may appease the poor without alienating the rich.

A second bargain concerns national defense. When their country is attacked, the rich need citizens to take up arms. By granting political rights, they motivate the poor to fight—and, afterwards, to put their weapons away. Democracy triumphed in Europe, Weber wrote, because the elite needed “the cooperation of the non-aristocratic masses and hence put arms, and along with arms political power, into their hands” (Ferejohn and Rosenbluth Reference Ferejohn and Rosenbluth2017; Ticchi and Vindigni Reference Ticchi and Vindigni2008; Weber Reference Weber1927, 324–5). This requires that democratization occur around the time of war or significant threat of it. The elite must consciously choose to extend rights—and to those needed to fight, or, later, those it seeks to demobilize.

Another approach traces democracy not to compacts between rich and poor but to splits within ruling circles. Lizzeri and Persico (Reference Lizzeri and Persico2004) view democratization as a way growth-oriented factions incentivize officials to buy support with public goods rather than patronage. Since public goods are non-rival, their usefulness for attracting votes increases with the electorate’s size. Broadening the franchise prompts politicians to substitute public goods for transfers. Evidence might include claims by reformers that democratization would reduce patronage or corruption.

Alternatively, political reforms might reflect partisan competition. One party might enfranchise new voters to boost its support against a rival (Collier Reference Collier1999; Ghosal and Proto Reference Ghosal and Proto2009; Llavador and Oxoby Reference Llavador and Oxoby2005; Przeworski Reference Przeworski, Schmitter, Whitehead and O’Donnell1986, 56). In this case, the record should show reformers expected to benefit electorally.

A third approach sees democracy as a peace-making device. It emerges, according to Rustow (Reference Rustow1970, 352–5), as a “great compromise” after “prolonged and inconclusive political struggle” between social forces. These might be elite and masses, social classes, or ethnic groups. In some cases, groups conclude an explicit “pact” that defines rules to resolve conflicts (O’Donnell and Schmitter Reference O’Donnell and Schmitter1986, 37–9). Observable implications include a history of “prolonged and inconclusive” conflict and reconciliation between factions that coincides with democratization.

In all these accounts, democracy is consciously chosen by the ruling group or some subset of it. But is that how democratization typically occurs? Consider another possibility.

Unintended Democratization: Previous Literature

Another tradition sees democratization not as a choice of the incumbent elite but as something that occurs after the old elite is overthrown. From Marx to Barrington Moore, writers have described parliamentary rule as resulting from a bourgeois revolution that displaced—or at least weakened—an agrarian aristocracy.Footnote ⁹ More recently, Huntington (Reference Huntington1991) distinguished between, first, transitions the elite initiated or negotiated with outsiders, and, second, “replacements,” in which democracy followed the incumbents’ forcible eviction.

I document that such replacements did occur in up to 49% of democratizations. I also examine why they did. Class theorists relate this to economic development and the evolving social structure. While these may matter, they rarely explain the exact timing and mode of transition. I explore how incumbents’ errors contribute.Footnote ¹⁰ Although structural factors may make revolution more likely and determine what comes after, leaders’ missteps often light the fuse. Indeed, among democratizations that followed the incumbent’s overthrow, my data suggest the revolt was almost always triggered or rendered lethal by a mistake.

A second literature, although presenting democratization as a deliberate choice, emphasizes the confused environment in which it occurs. O’Donnell and Schmitter (Reference O’Donnell and Schmitter1986, 3–4) noted the importance of “elements of accident and unpredictability, of crucial decisions taken in a hurry with very inadequate information.” I document how right these authors were. Still, in their account, ignorance and uncertainty are primarily obstacles to democratization rather than catalysts of it. Regime reformers must overcome such problems to unite with moderate outsiders. It takes “the talents of specific individuals (virtù)” to defeat the vagaries of fortuna (O’Donnell and Schmitter Reference O’Donnell and Schmitter1986, 5). In my argument, fortuna is not an obstacle but the hero of the story: reform occurs when fortuna overcomes the virtù of autocracy’s defenders.

In recent work, Weyland (Reference Weyland2008; Reference Weyland2012; Reference Weyland2014) shows how bounded rationality, cognitive heuristics, and loss aversion explain various political decisions. For instance, Louis-Philippe’s overthrow in 1848 France led citizens elsewhere to expect similar outcomes in their quite different settings. They took to the streets—only to be crushed. Weyland shows how citizens’ mistakes caused democratization attempts to fail, in both 1848 and the Arab Spring. I show here how leaders’ mistakes explain democratization successes.

The closest parallel to my argument is Przeworski (Reference Przeworski1991, 63–6), who describes two paths to democracy based on “mistaken assumptions.” In both, incumbents introduce partial reforms that then get out of hand. In the first, hardliners misread their reformist allies, thinking they favor preserving the regime when in fact they prefer full democratization. In the second, (reformist) incumbents overestimate their ability to halt a revolution midway by threatening repression. These correspond to two of the 10 types of mistakes I identify: empowering secret democrats within the regime and making inopportune concessions that lead to a “slippery slope.” I show that the first probably or very probably occurred in 7–12% of democratizations, and the second in 8–12%.Footnote ¹¹

I build on these various works. While authors such as Przeworski have noted particular errors, I situate these in a broader spectrum of miscalculations and missteps and propose a typology rooted in theories of authoritarian governance. I document the contribution of 10 varieties of mistake, across settings and eras, and show how the balance has shifted over time. The results suggest that in most cases—and far more than prominent deliberate-choice arguments can explain—rulers’ missteps were critical to the mechanism that produced regime change.

Democratization by Mistake

But first, what constitutes a mistake? By this I mean the choice of a course of action or inaction, the expected payoff of which is less than that of another feasible course, where expectations are based on objective probabilities. In short, a mistake is a non-optimal choice.

Mistakes come in two main forms. A mistake of information occurs when actors choose non-optimal actions because they have incorrect or imprecise beliefs. The actor’s reasoning may be perfect. “We must not say,” Cicero wrote, “that every mistake is a foolish one” (Taylor Reference Taylor1918, 481). But this reasoning is based on faulty information. A mistake of calculation occurs when, despite accurate and precise beliefs, actors optimize incorrectly. They fail to see that a different action yields a higher expected payoff.

Not all actions with undesired outcomes are mistakes. One may lose a gamble that was, nevertheless, optimal ex ante. Or, faced with only bad options, one may choose the “lesser evil.” Conversely, an action may be a mistake even if all options were bad. All that is required is that one other feasible course have a higher expected payoff. It might seem odd at first to call a choice that is logical given current beliefs a “mistake.” Hindsight, in the cliché, comes with 20/20 vision. But “mistakes of information” do, in fact, conform with common usage and dictionary definitions.Footnote ¹² The sentence “If I’d known X, I wouldn’t have made that mistake” makes perfect sense.

Democratization by mistake occurs if the regime becomes democratic because of one or more mistakes by the authoritarian incumbents.Footnote ¹³ Of course, democratization results from a series of actions rather than a single choice. To say a mistake led to democratization is not to say that all the ruler’s actions, including his last as ruler, were non-optimal—only that at least one important action in the chain was.Footnote ¹⁴ Finally, I do not assume leaders always prioritize retaining power. If they democratize because they expect to dominate under a competitive order (Slater and Wong Reference Slater and Wong2013) or in order to avoid bloodshed, that is not a mistake.

Thus, transition can occur by three paths: a deliberate choice by the incumbent—deliberate democratization, a failure by the incumbent to prevent democratization despite choosing optimal actions—unintended but unavoidable democratization, and a failure to prevent it after at least one non-optimal action—democratization by mistake. Empirically, it is hard to distinguish unintended but unavoidable democratization from deliberate democratization unless the incumbent was overthrown before political reform began. If all reform announcements came after the incumbent’s overthrow, reforms did not occur because he intended them. If, additionally, he made no clear mistakes, the democratization can be considered not just unintended but also from his perspective unavoidable. If, on the other hand, the incumbent was not overthrown and made no clear mistakes leading to democratization, I conservatively classify the case as deliberate.

More concretely, what do such mistakes of authoritarian rulers look like? To identify departures from optimality, one must know what the optimal strategy for preserving an autocratic regime is. Recent literature decomposes this into three key tasks. First, the ruler must deter or defeat challenges from domestic outsiders; Svolik (Reference Svolik2012) calls this the problem of “authoritarian control.” Second, he must avoid conflict among insiders—the problem of “authoritarian power-sharing” (Svolik Reference Svolik2012). Third, he must manage relations with foreign actors. Mistakes are non-optimal actions in pursuit of these three tasks. Figure 1 disaggregates further.

FIGURE 1.

Typology of Paths to Democracy

The main tactics for controlling domestic outsiders are co-optation and repression (often combined with surveillance and censorship). Broadly speaking, incumbents can use these non-optimally in four ways. They may make excessive or poorly targeted concessions that in fact strengthen opposition (Rasler Reference Rasler1996). Partial reforms, in some settings, increase pressure for more, pushing leaders down the proverbial “slippery slope.” Or rulers may fail to make concessions that would have divided and weakened the opposition or co-opted new allies (Goldstone and Tilly Reference Goldstone, Tilly and Aminzade2001, 188). Likewise, excessive or poorly targeted repression can backfire, exposing the regime’s brutality and inciting the previously apathetic to revolt (Sutton, Butcher, and Svensson Reference Sutton, Butcher and Svensson2014). Or, rulers may fail to repress—or monitor, or censor—the opposition in ways that could have deterred or disrupted its challenges (Tilly Reference Tilly1978).

Besides misusing carrots and sticks, autocrats can discredit themselves through domestic policy failures—destabilizing the economy, hiking food prices, tolerating extreme corruption, and so on (for such policies to be a mistake, a superior alternative must have been feasible). Overconfident leaders may also call an election or referendum, only to find they are less popular than believed. Such revelations embolden the opposition and often prompt insider defections. A related mistake is to falsify results so brazenly that it provokes rebellion. Incumbents may have no better option. But if they could have increased their survival odds by postponing, canceling, or never calling the election; campaigning more effectively; or manipulating results more adroitly, then the action constitutes a mistake.

Besides controlling the population, rulers must control other insiders. Internal conflict often triggers regime breakdown. Mistakes by incumbents can cause this in three main ways. First, the leader may avoidably antagonize key members of the ruling coalition such as the Church or business elite, weakening his position. Second, he may alienate some in the armed forces or security services, reducing repressive capacity or prompting a coup. Third, a dictator may choose as his successor—or insiders may select as leader—a covert reformer who undermines the regime or an incompetent who fails to defend it.

The third challenge is to manage international relationships. A common mistake, in this regard, is to enter or initiate military conflict—and do poorly. This may end in invasion or, at least, discredit the ruler domestically, prompting elite defections and mass protests. As always, failure only indicates a mistake if avoiding the conflict or commanding more intelligently would have increased survival odds.

Since democratizations consist of processes rather than single events, each episode may contain more than one mistake. One may lead to another—for instance, alienating the army may result in the non-optimal use of repression—or an underperforming leader may err in multiple ways. Similarly, actors in deliberate choice cases may have multiple motives—perhaps seeking to end a long confrontation while increasing their faction’s support. Complicating matters further, mistakes may combine with elements of deliberate choice (hence, the dashed arrows in Figure 1). In Spain, Franco and his cronies entrusted future power to individuals disloyal to the regime (a mistake), who later democratized by forging a multilateral pact (deliberate choice). I still label an episode “democratization by mistake” if—as in the Spanish example—the incumbent deliberately chose reform only after prior mistakes.

Identifying Cases

How one defines democratization depends on how one thinks about democracy. One tradition places political regimes on a continuum from pure autocracy to pure democracy (Bollen and Jackman Reference Bollen and Jackman1989). Democratization is then any reform that moves the country a certain distance in the democratic direction (some label this “political liberalization”). I call this a directional definition. A second tradition sees democracy as a binary category—either a system is one or it is not; democratization is then a reform, however large or small, that moves the country across the definitional threshold (Przeworski et al. Reference Przeworski, Alvarez, Cheibub and Limongi2000). I call this a qualitative definition. A third hybrid concept combines the first two. Such definitions require both a certain movement in the democratic direction and the crossing of some qualitative threshold.

Which definition is appropriate depends on the purpose (Collier and Adcock Reference Collier and Adcock1999). Fortunately, we need not privilege one here. I adopt definitions of each type and show that similar conclusions follow whichever is used. I use two directional definitions: (a) an increase of six or more points on the 21-point Polity2 scale, completed within three years (P), and (b) an increase of 0.3 within three years on the Varieties of Democracy (V-Dem) electoral democracy index, which runs from 0 to 1 (V).Footnote ¹⁵ My first qualitative definition employs the data of Boix, Miller, and Rosato (Reference Boix, Miller and Rosato2013)(BMR). They define democracy as a system in which elections are free and competitive, the head of government is either directly elected or answerable to an elected parliament, and at least half the male population can vote.Footnote ¹⁶ Democratization is a change from non-democracy to democracy. Two additional definitions come from the Lexical Index of Electoral Democracy database (LIED), which identifies objective thresholds for different levels of electoral democracy (Skaaning, Gerring, and Bartusevičius Reference Skaaning, Gerring and Bartusevičius2015). Specifically, LIED4 records when a country began holding at least “minimally competitive, multiparty elections for legislature and executive” (movement from < 4 to ≥ 4 on the index); LIED5 records when a country began holding such elections with at least “full male suffrage” (movement from < 5 to ≥ 5).Footnote ¹⁷ Finally, my first hybrid definition adopts Polity’s concept of “major democratic transition.”Footnote ¹⁸ This occurs when (a) Polity2 rises by six or more points within three years and (b) Polity2 moves from [-10, 0] to (0, 10] or from (0, 6) to [6,10] (MDT). My second, more demanding, hybrid definition requires that (a) Polity2 rises by six or more points within three years, and (b) Polity2 end at 6 or higher (MDT6). A final hybrid definition consists of an increase of 0.3 within three years on V-DEM’s electoral democracy index that ends with a rating of “democratic” on V-DEM’s ordinal electoral democracy index (V+).

The broad coverage of these datasets permits the examination of all democratizations between 1800 and 2010 (BMR) or 2015 (Polity IV, LIED, V-DEM). Applying these definitions yields a preliminary list of 316 cases. For various reasons, 46 turn out not to coincide with any identifiable democratic reform (see “Notes on Excluded Cases,” in the online appendix). Excluding these leaves 270 episodes—74–153 fitting directional definitions, 131–183 qualitative definitions, and 51–139 hybrid definitions; as is evident, the cases overlap (Table A1).

Sources and Classifications

My method here was congruence analysis (Beach and Pedersen Reference Beach and Pedersen2016, 269–301; Blatter and Blume Reference Blatter and Blume2008; Møller Reference Møller2017). The goal was to check whether historical evidence in particular cases matched the observable implications—usually about causal mechanisms—of given theories. The main theories examined were the deliberate-choice theories discussed in the “Democratization by Choice” section. At the same time, I explored the role of incumbent mistakes. For each case, I examined the mechanism that led to democratization and evaluated whether available evidence suggested that this included only optimal incumbent choices or at least one significant mistake. If the latter, I classified the mistakes by using the typology of Figure 1.

For each episode, I prepared a synopsis of events and historians’ interpretations.Footnote ¹⁹ I assessed the evidence for each observable implication of each deliberate-choice theory. To increase analytic transparency, I articulated the reasoning behind each nontrivial judgment in an accompanying table, along with relevant quotations from sources (Moravcsik Reference Moravcsik2014). The same tables summarize evidence of mistakes. Based on the historical evidence, I classified cases into the three paths of Figure 1; Table 1 shows the criteria that were used.

TABLE 1.

Criteria Used to Classify Cases of Democratization Into the Three Paths

The more than 2,000 sources included history books and articles, newspapers, magazines, web publications, diplomatic cables, biographies, memoirs, diaries, and published interviews of participants, in multiple languages and formats. I sought a range of sources for each case, following leads suggested by the sources themselves to locate missing information or resolve apparent disagreements. Each type is vulnerable to particular biases. Historians may be influenced by their political beliefs or worldview. Newspapers and magazines target specific audiences. Memoirs and interviews, while offering unique insight into actors’ motivations, are distorted by self-justification (because admitting mistakes is embarrassing, this should work against the conjecture that mistakes mattered). Private diaries are less self-serving but may be partial, episodic, or esoteric. Temporally proximate primary sources are more likely to contain direct observation and unfiltered evidence of actors’ thinking (Lieberman Reference Lieberman2010, 41). Still, heeding Kreuzer’s (Reference Kreuzer2010, 383) call to “take not only history, but also historians seriously,” I also studied historical interpretations.

I adopted standard techniques to combat potential biases (e.g., Howell and Prevenier Reference Howell and Prevenier2001). First, for each source I considered, based on identity of producer and context, what type of bias was plausible and discounted evidence that aligned with that bias, often noting this in the synopsis source notes. Second, I used multiple sources to cross-check facts and compare interpretations.

To minimize confirmation bias, Møller and Skaaning (Reference Møller and Skaaning2018, 4) recommend “giving less weight to historical works of other social scientists and to work by historians with similar theoretical claims as the one under consideration.” For each source, I therefore recorded in a table whether it embraced some social science theory—and, if so, how this related to those I examined. I relied mostly on atheoretical sources and discounted (or excluded) those committed to a theory similar to the one I was examining. I also heeded Møller and Skaaning’s recommendations to attend to possible differences in the meaning of concepts across eras and contexts and to prioritize sources that incorporated updated evidence.

As suggested by Lieberman (Reference Lieberman2010, 45), I graded the quantity and quality of sources available for each case and incorporated these judgments into the analysis. Specifically, I classified the materials as (a) “a lot of information—no serious disagreement on relevant points,” (b) “moderate amount of information—no serious disagreements on relevant points,” (c) “significant disagreement or ambiguity among sources on some relevant points, or key pieces of information lacking,” or (d) “very little information.” I then checked whether results would differ excluding cases with inferior source material.

Beyond the quantity and quality of source material, the evidence may be clearer in some cases than others. Therefore, I also recorded an estimate of my confidence in each conclusion (on a five-point scale: 1 = “very probably no,” 2 = “probably no,” 3 = “maybe/unclear,” 4 = “probably yes,” and 5 = “very probably yes”). Presenting results, I focused on the proportion graded 4 or 5 (“probably” or “very probably” yes), but in the tables I also list the proportions for narrower (just 5) and broader (≥3) criteria.

Throughout, I sought to trace mechanisms that linked incumbents’ actions to the outcome. But the analysis also employed counterfactual reasoning. To call an action a mistake, in my definition, implies that some alternative course would have increased the regime’s survival odds. Counterfactuals are sometimes considered controversial because it is hard to evaluate evidence for and against something that did not happen. Yet, as scholars increasingly recognize, it is almost impossible to discuss history in causal terms without entertaining counterfactuals (Bunzl Reference Bunzl2004; Tetlock and Belkin Reference Tetlock and Belkin1996).Footnote ²⁰ The question is rather how to distinguish “appropriate” or “grounded” counterfactuals from less pertinent ones.

Some have suggested a few principles. To be useful, counterfactuals should clearly define the starting point, outcome, and mechanism connecting the two (clarity or transparency). This mechanism should be consistent with well-established theoretical and empirical generalizations (theoretical and empirical consistency). The posited change should be possible in isolation (feasibility of manipulation). It should be consistent with background conditions and mechanism (cotenability) as well as with established facts about the historical setting (minimal rewriting). Finally, the counterfactual starting point and outcome should be close together in time (temporal proximity) (Bunzl Reference Bunzl2004; Fearon Reference Fearon1991; Levy Reference Levy2015; Rohlfing Reference Rohlfing2012; Tetlock and Belkin Reference Tetlock and Belkin1996).

In this project, the counterfactual always refers to the incumbent choosing the optimal feasible action (O) instead of the mistake (M). Let D denote democratization and C all other contextual factors. The counterfactual outcome is the achievement of a lower probability of democratization: P(D|O,C), where P(D|O, C) < P(D|M,C). Here, clarity requires a precise statement of the mistake, counterfactual, and mechanism. In each synopsis table, I provide these. Some of the other principles will hold almost automatically here, for several reasons. Since the types of “mistakes” were themselves derived from prevailing theories and empirical generalizations about how dictators maintain power, the mechanisms are rooted in these same theories and generalizations (theoretical and empirical consistency). Other principles are implicit in my definition of mistake. If an alternative could not be chosen in isolation, or if background conditions precluded it, then it could not be considered “feasible” (feasibility of manipulation, cotenability).

The final two criteria are less automatic. Minimal rewriting rejects counterfactuals that imply major reconstruction of antecedent facts and processes. When the mistake is a simple choice by the leader, imagining him choosing differently rarely requires much alternative prior history. However, when the mistake is a broader failure of the ruling elite to implement some policy, rewriting may not be so minimal. For roughly 7% of putative mistakes, I judged the counterfactual might require significant changes to preceding history; I coded these as not mistakes. The temporal proximity requirement says the counterfactual antecedent (here, O) and the counterfactual outcome (P(D|O,C)) should be close together in time. In fact, the effect here is instantaneous: the probability of democratization changes from P(D|C) to P(D|O, C) as soon as O occurs. So, under a literal interpretation, temporal proximity always holds. The change is also instantaneous between the actual antecedent (M) and actual outcome—from P(D|C) to P(D|M,C). However, I adopted an intuitive adaptation of the criterion which required that the mistake, M, and actual democratization, D, be close together in time. As a rule of thumb, I called a case proximate if the start of regime change occurred within three years after the mistake and checked the results excluding all nonproximate cases.Footnote ²¹

While I often refer to a single “incumbent,” authoritarian regimes are not unitary. Scholars generally distinguish between a leader and other members of a ruling elite group. While most mistakes are made by the leader, with or without other elite members, a few are made by the elite alone (e.g., when the elite appoint a suboptimal leader). In the synopses, I noted which case applied, and I examined how patterns change excluding mistakes made without the leader’s involvement.

The appendix includes the coding instructions. Throughout, I sought to be conservative, recording “maybe/unclear” when in doubt. However, limited inferences often seemed reasonable. Various incumbents were deposed—and killed or exiled—by coups that a little more surveillance or attention to military grievances could have prevented. In other cases, “emperors” advertised their “nakedness,” as when Ceausescu’s state television showed demonstrators booing the dictator days before his overthrow (broadcasting live was reckless). Some leaders openly admit error. Mikhail Gorbachev, asked in 2011 whether he had made any mistakes in office, listed five (Steele Reference Steele2011). Turkey’s General Evren later lamented not having created a pro-military party after the 1980 coup. “We made a mistake,” he confessed. “We’ve walked this sheep to market and now we are presenting the meat to somebody else” (Pope and Pope Reference Pope and Pope1997, 155). Quick reversals of policies that had proved costly also provide prima facie evidence of error. Meanwhile, accounts of contemporaries offer insight into incumbents’ thinking. Tocqueville portrays Louis-Philippe as obtusely blundering into revolution. Louis-Philippe’s son, the Prince of Joinville, described his father as “at an age at which a man no longer accepts criticism” (Bourgeois Reference Bourgeois1919, 285–6). Such sources sometimes identify feasible alternatives. For instance, Tocqueville warned of rising popular anger and proposed concessions he thought would preserve stability (Tocqueville Reference Tocqueville1964, 32–3).

Major redundancy should increase confidence in the coding. Among incumbents who “probably” or “very probably” made at least one mistake, about two-thirds made two or more and one-third three or more. Thus, I could have rejected one or even two “mistakes” for many cases without invalidating the overall classification. It would take a comprehensive reinterpretation of multiple episodes to challenge the conclusion that missteps were important.

As an additional check, I had a research assistant code a randomly selected 5% of the cases (14 episodes), using the same sources but without seeing my synopses or results. Table A2 shows the correspondence between our ratings. Of these 14 cases, I had coded 12 as “democratization by mistake,” one as “unintended but unavoidable,” and one as “deliberate democratization.” The second coder also classified 12 as “democratization by mistake,” but judged both of the other two “deliberate democratization.” Intercoder agreement measures were high, ranging from 0.79 to 0.93. Agreement was sometimes lower for particular types of mistake, but this was usually because the second coder saw more mistakes than I had, which suggests that my classifications were, indeed, relatively conservative.

The ultimate test, as with any historical research, will be examination by critical readers. To maximize transparency, the synopses (covering more than 2,300 pages) include lengthy quotations from sources. Sources are fully annotated with page numbers and web addresses where possible. To illustrate, Table A3 and Box 1 in the appendix show the coding—and associated reasoning—for one example.

Aggregate Patterns

Table 2 shows how the cases break down among deliberate democratization, unintended but unavoidable democratization, and democratization by mistake. Here, I focus on the percentage of cases in which a given classification “probably” or “very probably” fit available evidence. Depending on the democratization indicator, from 14% to 31% of the cases were deliberate and from 1% to 6% were unintended but unavoidable. However, the vast majority—almost three-quarters on average—followed at least one significant mistake.

TABLE 2.

Paths of Democratization (Percentage of Cases)

Note: Source—author’s assessments. See online appendix for historical sources. P: increase of six or more points on the Polity2 scale, completed within three years; MDT: major democratic transition under Polity definition; MDT6: major democratic transition ending at Polity2 ≥ 6; BMR: Boix, Miller, Rosato (Reference Boix, Miller and Rosato2013) definition; LIED4: transition to minimally competitive, multiparty elections (LIED); LIED5: transition to at least full male franchise (plus minimally competitive, multiparty elections) (LIED); V: increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years; V+: increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years, ending as “democratic” (e_v2x_api_5C ≥ 0.75). Deliberate democratization: incumbent was not overthrown before reforms announced and democratization was not “very probably” or “probably” a consequence of some significant mistake or mistakes by the incumbent. Unintended but unavoidable: incumbent was overthrown before reforms announced and democratization was not “very probably” or “probably” a consequence of some significant mistake. Democratization by mistake: incumbent “very probably” or “probably” made at least one significant mistake that increased the odds of democratization.

Figure 2 shows how the breakdown changed across the three waves of democratization. I use Huntington’s (Reference Huntington1991) periodization: 1st wave: 1828–1926; 2nd wave: 1927–1962; 3rd wave: from 1963.Footnote ²² Two observations suggest themselves. First, the frequency of deliberate democratization has fallen over time, while that of democracy by mistake has increased. Second, in all periods, democratization by mistake accounted for a large majority of the cases.

FIGURE 2.

Paths of Democratization Over Time (Percentage of Cases)

Note: Source—author’s assessments. See online appendix for historical sources. Percentage of cases (averaged across eight democratization measures). Deliberate democratization: incumbent was not overthrown before reforms announced and democratization was not “very probably” or “probably” a consequence of some significant mistake or mistakes by the incumbent. Unintended but unavoidable: incumbent was overthrown before reforms announced and democratization was not “very probably” or “probably” a consequence of some significant mistake. Democratization by mistake: incumbent “very probably” or “probably” made at least one significant mistake that increased the odds of democratization.

Deliberate Choice

In 23% of the cases on average, the incumbent appeared to choose democracy deliberately (Table 2). These do not correspond exactly to instances of the five deliberate choice theories discussed in the “Democratization by Choice” section, for two reasons. First, one of the five theories applies in only about one-third of the “deliberate democratization” cases in Table 2. In two-thirds, leaders democratized deliberately for other reasons—foreign occupation, international pressure, among others. Second, the five deliberate choice arguments sometimes do apply in cases where the incumbent stumbled into democratization after a mistake (19% of the cases on average). For instance, Chile’s General Pinochet, after his humiliating referendum defeat, spent the next years negotiating protections for the military. At least one of the five deliberate choice theories “very probably” or “probably” fit the evidence in 29% of the cases, averaging across the eight democratization indicators (Figure 3 and Table A4).

FIGURE 3.

Democratizations in Which Deliberate Choice Arguments “Probably” or “Very Probably” Fit Historical Evidence (Percentage of Cases)

Note: Source—author’s assessments. See Democratization Synopses for historical sources. Polity: increase of six or more points on the Polity2 scale, completed within three years; MDT: Major Democratic Transition under Polity definition; MDT6: Major Democratic Transition ending at Polity2 ≥ 6; BMR: Boix, Miller, Rosato (2013) definition; LIED4: Transition to minimally competitive, multiparty elections (LIED); LIED5: Transition to at least full male franchise (plus minimally competitive, multiparty elections) (LIED); V: Increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years; V+: Increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years, ending as “democratic” (e_v2x_api_5C ≥ 0.75). The number shown is the percentage of democratizations that “probably” or “very probably” fit the given argument, averaged across the eight democratization indicators.

Democratization by Mistake

How often did democratization follow from incumbents’ errors? Figure 4 (and Table A8) summarize the findings. In 74% of the episodes on average (67–85%), the path to democratization “probably” or “very probably” included one or more incumbent mistakes.

FIGURE 4.

Democratizations “Probably” or “Very Probably” Caused in Part by Incumbents’ Mistakes (Percentage of Cases)

Note: Source—author’s assessments. See Democratization Synopses for historical sources. Polity: increase of six or more points on the Polity2 scale, completed within three years; MDT: major democratic transition under Polity definition; MDT6: major democratic transition ending at Polity2 ≥ 6; BMR: Boix, Miller, Rosato definition; LIED4: Transition to minimally competitive, multiparty elections (LIED); LIED5: Transition to at least full male franchise (plus minimally competitive, multiparty elections) (LIED); V: Increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years; V+: Increase of at least 0.3 on the V-DEM electoral democracy index (v2x_polyarchy) within 3 years, ending as “democratic” (e_v2x_api_5C ≥ 0.75). The number shown is the percentage of democratizations to which the given mistake “probably” or “very probably” contributed, averaged across the eight democratization indicators.

Change over Time and Robustness

Mistakes were important in all waves, and their importance generally increased over time (Table A9, Figure A3). On average, democratization by mistake increased from 59% of the cases in the first wave to 79% in the third. In the first wave, leaders blundered most in foreign policy (usually through reckless bellicosity), domestic policy, and in failing to conciliate domestic actors. Later, the focus shifted to alienating the armed forces, counterproductive violence, and electoral miscalculations. The role of mistakes is slightly higher—81% on average—focusing on just high information cases (Table A10) and about the same—77% on average—among democratizations that proved to be permanent (Table A11).

Table A12 excludes the roughly 10% of mistakes—most often domestic or foreign policy failures—that were not temporally proximate (occurring within three years of regime change). Democratizations by mistake fall from 74% to 71% of the cases. Table A13 counts only mistakes in which the central leader himself participated, excluding, for instance, cases in which the elite selected a covert democrat as leader. Again, aggregate results change little. One reason such recodings have little effect is the noted redundancy in errors. Blunders were often multiple. In 50% of the cases, the incumbent made two or more mistakes; in 10%, the incumbent made four or more (Figure 5).

FIGURE 5.

Number of Mistakes per Democratization Episode

Note: Source—author’s assessments. See Democratization Synopses for historical sources. Percentage of cases in which mistakes “probably” or “very probably” contributed to democratization (averaged across eight democratization measures).

Finally, some draw a sharp distinction between “democratizations,” which end in democracy, and “political liberalizations,” which merely soften an authoritarian regime. While the MDT6, BMR, LIED4, LIED5, and V+ definitions already capture just “democratizations” in this sense, the Polity, MDT, and V definitions combine democratizations with political liberalizations. Table A14 shows results for just political liberalizations (that is, significant, concentrated increases on the Polity2 or V-DEM electoral democracy scale that do not cross the democracy threshold). Mistakes are slightly less frequent than for democratizations—61–63%—but still fit far more often than deliberate choice arguments—16–17%.