If a country, any randomly selected country, is to have a democratic regime next year, what conditions should be present in that country and around the world this year? The answer is: democracy, affluence, growth with moderate inflation, declining inequality, a favorable international climate, and parliamentary institutions.
This answer is based on counting instances of survival and death of political regimes in 135 countries observed annually between 1950 or the year of independence or the first year when economic data are available (“entry” year) and 1990 or the last year for which data are available (“exit” year), for a total of 4,318 country-years. 1 We found 224 regimes, of which 101 were democracies and 123 dictatorships, observing 40 transitions to dictatorship and 50 to democracy. Among democratic regimes, there were 50 parliamentary systems, 46 presidential systems, and 8 mixed systems. 2
Our definition of democracy is a minimalist one. We follow Robert A. Dahl’s 1971 classic Polyarchy in treating as democratic all regimes that hold elections in which the opposition has some chance of winning and taking office. When in doubt, we err in the direction of calling a regime dictatorial. Our classification is not idiosyncratic, but is closely related to several alternative scales of democracy. The rationale and the rules for classifying regimes are discussed in the Appendix below. [End Page 39]
Democracy. It may seem tautological to say that a country should have a democratic regime this year in order to have a democracy next year. We do so in order to dispel the myth, prevalent in certain intellectual and political circles (particularly in the United States) since the late 1950s, that the route to democracy is a circuitous one. The claim is that 1) dictatorships are better at generating economic development in poor countries, and that 2) once countries have developed, their dictatorial regimes will give way to democracy. To get to democracy, then, one had to support, or at least tolerate, dictatorships.
Both of the above propositions, however, are false:
1. While analyses of the impact of regimes on economic growth have generated divergent results, recent econometric evidence fails to uncover any clear regime effect. The average rate of investment is in fact slightly higher in poor democracies than in poor dictatorships; population growth is higher under dictatorships but labor productivity is lower; and investment is more efficiently allocated under democracies. Dictatorships are no more likely to generate economic growth than democracies. 3 Indeed, the 56 dictatorships with annual per-capita income of less than $1,000 when we first observed them simply failed to develop. 4 By the exit year, only 18 of them had made it (whether under democracy or continued dictatorship) to $1,000, only 6 to $2,000, and only 3 to more than $3,000. South Korea and Taiwan are exceptional: they are the only two dictatorships that started under $1,000 in 1950 and had annual per-capita income exceeding $5,000 by 1990. If we consider as “initially poor” those countries with less than $2,000, we find that among 98 dictatorships first observed below this level, by the exit year only 26 had made it to $2,000, 15 to $3,000, 7 to $4,000, and 4 to $5,000. These figures should be enough to dispel any notion that dictatorship somehow promotes economic growth in poor countries.
2. Democracies are not produced by the development of dictatorships. 5 If they were, the rate at which dictatorships make the transition to democracy would increase with the level of development: analyses of the survival prospects of dictatorships, however, indicate that this is not the case. Indeed, transitions to democracy are random with regard to the level of development: not a single transition to democracy can be predicted by the level of development alone. 6
Since poor dictatorships are no more likely to develop than poor democracies and since developed dictatorships are no more likely to become democracies than poor ones, dictatorships offer no advantage in attaining the dual goal of development and democracy. In order to strengthen democracy, we should strengthen democracy, not support dictatorships.
Affluence. Once a country has a democratic regime, its level of economic development has a very strong effect on the probability that [End Page 40] democracy will survive. Poor democracies, particularly those with annual per-capita income of less than $1,000, are extremely fragile: based on our study, the probability that one will die during a particular year is 0.12. This rate falls to 0.06 in the $1,000 to $2,000 range, to 0.03 between $2,000 and $4,000, and to 0.01 between $4,000 and $6,000. These numbers mean that a democracy can be expected to last an average of about 8.5 years in a country with per-capita income under $1,000 per annum, 16 years in one with income between $1,000 and $2,000, 33 years between $2,000 and $4,000, and 100 years between $4,000 and $6,000.
Whatever their theoretical and political differences, both Samuel P. Huntington and Guillermo O’Donnell claim that there is a level beyond which further development actually decreases the probability that democracy will survive. 7 Huntington argues that both democracies and dictatorships become unstable when a country undergoes modernization, which occurs at some intermediate level of development. O’Donnell, in turn, claims that democracies tend to die when a country exhausts “the easy stage of import substitution,” again at some intermediate level. Our finding, however, is that there is no income level at which democracies become more fragile than they were when they were poorer. Only in the Southern Cone countries of Latin America have authoritarian regimes arisen at the intermediate levels of development. Four out of the nine transitions to authoritarianism above $3,000 transpired in Argentina. Adding Chile and Uruguay, we see that the instances in which democracy fell at medium levels of development are to a large extent peculiar to the Southern Cone. 8
Above $6,000, democracies are impregnable and can be expected to live forever: no democratic system has ever fallen in a country where per-capita income exceeds $6,055 (Argentina’s level in 1976). Hence Seymour Martin Lipset was correct to assert that “the more well-to-do a nation, the greater the chances that it will sustain democracy.” 9 Once established in a developed country, democracy endures regardless of how it performs and regardless of all the exogenous conditions to which it is exposed.
Why democracies are more durable in more-developed countries has been the subject of extensive speculation. One reason, put forward by Lipset in Political Man, is that the intensity of distributional conflicts is lower at higher income levels. Another plausible hypothesis, suggested to us by Larry Diamond, focuses on institutions: political actors in more-developed countries may be more likely to adopt a superior institutional framework at the moment when democracy is established. Later, we will examine this hypothesis with regard to parliamentarism and presidentialism. First, however, we will take up consideration of our third condition for the maintenance of democracy: economic performance [End Page 41]
Economic performance. For some countries, therefore, the story ends here: once democracy is in place, affluence is a sufficient condition for it to survive regardless of anything else. But democracies can survive in poorer countries, if they generate economic growth with a moderate rate of inflation.
While Lipset, economist Mancur Olson, and Huntington all thought that democracy becomes destabilized when a country grows rapidly, they could not have been more wrong. 10 Rapid growth is not destabilizing for democracies (or for dictatorships): indeed, democracies are always more likely to survive when they grow faster than 5 percent annually than when they grow slower. In turn, the fragility of democracy at lower levels of development flows largely from its vulnerability in the face of economic crisis. 11 Poor democracies, those under $1,000, have a 0.22 probability of dying in a year after their income falls (giving them a life expectancy of less than five years) and a 0.08 probability (or an expected life of 12.5 years) if their income rises. Between $1,000 and $6,000—the middle range—democracies are less sensitive to growth but more likely to die if they stagnate: they die at the rate of 0.059 when they decline, so that their expected life is about 17 years, and at the rate of 0.027, with an expected life of about 37 years, when they grow. Thus Larry Diamond and Juan Linz are correct to argue that “Economic crisis represents one of the most common threats to democratic stability.” 12 Conversely, economic growth is conducive to the survival of democracy. Indeed, the faster the economy grows, the more likely democracy is to survive.
Inflation also threatens democratic stability. A democratic regime has a 0.023 chance of dying and an expected life of 44 years when the annual inflation rate is under 6 percent; a 0.014 chance and an expected life of 71 years when inflation is between 6 and 30 percent; and a 0.064 chance and an expected life of about 16 years when inflation is above 30 percent. Note that these results appear to confirm Albert Hirschman’s 1981 hypothesis that a moderate rate of inflation promotes democratic stability. 13
Economic performance, then, is crucially important for the survival of democracy in less-affluent countries. When the economy grows rapidly with a moderate rate of inflation, democracy is much more likely to last even in the poorest lands.
Income inequality. The study of the political effects of income inequality is hampered by the paucity and poor quality of the available data. The best collection of internationally comparable data, generated by the World Bank, includes 266 observations, scattered over time for 84 countries.
We tried to assess the impact of income inequality (measured by the income ratio between the richest and the poorest quintiles) on the [End Page 42] probability that a democracy will survive for three and for five years following the time for which data are available. Unlike Edward N. Muller, we could find no pattern. 14 Since income inequality tends to be lower in poor countries, where most of the labor force is employed in self-sufficient agriculture, and in wealthy countries, where most workers are wage earners, and since democracy is brittle in poor countries and impregnable in rich ones, no overall pattern emerged from this analysis. The scantiness of our data, moreover, prevented us from controlling for the level of development.
On the other hand, we did find that democracy is much more likely to survive in countries where income inequality is declining over time. For those democratic regimes for which we had more than one observation of income distribution, we calculated the probability that democracy would die should inequality either increase or decrease. We found that the expected life of democracy in countries with shrinking inequality is about 84 years, while the expected life of democracies with rising income inequality is about 22 years (these numbers are based on 599 democratic years, with inequality increasing during 262 and declining during 337). Note that these findings contradict any notion that distributional pressures threaten the survival of democracy: people expect democracy to reduce income inequality, and democracies are more likely to survive when they do.
International climate. Economic factors are not the only ones that matter for the durability of democracy. Indeed, international conditions predict regime survival better than does the level of development. While we cannot statistically distinguish different mechanisms by which the international climate becomes transmitted to particular countries, the proportions of other democracies in the region and in the world matter separately for the survival of democracy in any particular country: the larger the proportion of democracies on the globe and in the region during a particular year, the more likely is democracy to survive in any particular country. The global effect is about twice as large as the regional effect, but these findings indicate that contagion operates independently of the direct influence of Western governments and various international institutions.
Political learning. It is frequently argued—Russia is a favorite example—that the absence of democratic traditions impedes the consolidation of new democratic institutions and, conversely, that democracy is more stable in countries (like Chile) that have enjoyed it in the past. What this argument misses is that if a country had a democratic regime (note the past tense), it is a veteran not only of democracy but of the successful subversion of democracy. Political learning, in other words, cuts both ways. Democrats may find the work [End Page 43] of consolidation easier when they can rely on past traditions, but antidemocratic forces also have an experience from which they can draw lessons: people know that overthrowing democracy is possible, and may even know how to do it. If the failed Russian hard-liners’ coup of 1991 was more of a coup de théâtre than a coup d’état, it was perhaps because the coup plotters simply did not know what they were doing—an ignorance for which they were justly ridiculed by their more-experienced Latin American soul mates.
An overthrow of democracy at any time during the past history of a country shortens the life expectancy of any democratic regime in that country. To the extent that political learning does occur, then, it seems that the lessons learned by antidemocratic forces from the past subversion of democracy are more effective than the traditions that can be relied on by democrats. 15
The effect of institutions. Democracies are not all the same. Systems of representation, arrangements for the division and supervision of powers, and methods of organizing interests, as well as legal doctrines and the rights and duties associated with citizenship, can and do vary widely among regimes that are generally recognized as democratic. These differences, expressed in the details of institutions, generate effects that two millennia of reflection and investigation have still not enabled us to grasp fully. We are far from knowing any clear answer to the question that Rousseau posed in his Constitution of Poland: Which institutions have which effects under which historical conditions?
Should we expect democracy to last longer under one institutional system than under another? Our analysis is limited to only one set of institutional features, summarized as parliamentarism versus presidentialism (we leave “mixed” systems aside as presenting too small a sample to yield any robust estimate). We thus test the hypotheses of Juan Linz, who offers several reasons why parliamentary democracies should prove more durable than presidential ones. 16
One of Linz’s arguments is that the stakes are higher under presidentialism, since a race for the presidency can have but a single winner. Linz observes that a defeated presidential candidate has no official role in politics, and most likely will not even be a member of the legislature, while in a parliamentary system the defeated candidate for the premiership will be leader of the opposition. 17 Moreover, it is likely that the fixed term of office under a presidential system is longer than the expected term of office under a parliamentary system. Finally, under presidentialism the chief executive is at the same time the head of state, thus being able to portray the president’s partisan interest as the national interest and thereby undermine the legitimacy afforded to the opposition.
The second reason why presidential democracies may be less durable [End Page 44] is that they are more likely to generate legislative paralysis. Such paralysis can occur under either system: under parliamentarism when no majority coalition can be formed, and under presidentialism when the legislature is controlled by a majority that is hostile to the president but not large enough to override presidential vetoes routinely. Under presidential systems the executive, by virtue of the fixed term of office, can survive alongside hostile legislatures, leading to stalemates between the executive and the legislative branch. As the great nineteenth-century English political writer Walter Bagehot observed, “when a difference of opinion arises, the legislature is forced to fight the executive, and the executive is forced to fight the legislature; and so very likely they contend to the conclusion of their respective terms.” 18 In several contemporary presidential systems the main line of political conflict is between the president and the congress, rather than among political parties. Under such conditions, no one can govern.
Legislative majorities are more frequent under presidentialism than under parliamentarism: 57.9 percent of the time under the former and 49.0 percent under the latter. 19 But in 24.2 percent of the presidential years, the share of the largest party in the legislature was smaller than one-half and larger than one-third. Since the proportion needed to override a presidential veto is typically two-thirds, these figures indicate that the conditions for executive-legislative deadlock are common under presidentialism. The average number of effective parties is about the same under the two systems: 3.10 under parliamentarism and 3.05 under presidentialism. Yet extreme fractionalization—in which no party controls more than a third of the seats—is more frequent under presidentialism (occurring 18 percent of the time) than under parliamentarism (where it occurs only 8.9 percent of the time).
Linz is right about the durability of alternative institutional arrangements. During the period under consideration, 14 democracies (or 28 percent of the 50 cases) died under a parliamentary system. Only one (12.5 percent of 8 cases) died under a mixed system, and 24 (52 percent of 46 cases) died under presidentialism. 20 Among those democracies that died during the period under our scrutiny, the parliamentary systems lasted an average of eight years, while their presidential counterparts lasted nine. But the parliamentary systems that were still around as of 1990 were much older: on the average about 43 years, as compared with 22 for presidential regimes. The probability that a democracy would die under presidentialism during any particular year of our study was 0.049; the comparable probability under parliamentarism was 0.014. If this difference appears small, think in terms of expected lives: democracy’s life expectancy under presidentialism is less than 20 years, while under parliamentarism it is 71 years.
This difference in durability is not an effect of the levels of economic development at which parliamentary and presidential regimes [End Page 45] operated. While parliamentary systems are on the average found in wealthier countries, presidential democracies are less durable at almost every level. Excluding countries with a 1990 population of fewer than one million (many of which have parliamentary systems) changes nothing: the hazard rates—conditional probabilities that a regime would die given that it survived thus far—are exactly the same. Nor is this difference due to some hidden features of Latin America: in fact, presidential regimes in Latin America live much longer than those in other regions, the United States excepted. Hence presidential democracies are not shorter-lived because they are in Latin America.
Scott Mainwaring, like Alfred Stepan and Cindy Skach, is also correct: democracies are less likely to survive when they combine presidentialism with a fragmented party system. 21 Combining presidentialism with a legislature where no single party has majority status is a kiss of death: such systems can expect to live only 15 years. Presidential democracies in which a single party does have a legislative majority can expect to live 26 years. “Deadlock,” a situation in which the share of seats of the largest party is between one-third and one-half, is even more deadly to presidential regimes. They die at the rate of 0.038 (with an expected life of 26 years) when there is no deadlock and at the rate of 0.091 (with an expected life of 11 years) when there is. Furthermore, descriptive information on parliamentarism supports Scott Mainwaring and Matthew Shugart’s argument that “Parliamentary systems with disciplined parties and a majority party offer the fewest checks on executive power, and hence promote a winner-takes-all approach more than presidential systems.” 22 Single-party majorities are not conducive to the survival of parliamentary democracies: those in which one party had a majority of seats in the lower house of the legislature have an expected life of 55 years, while parliamentary systems without a one-party majority have an expected life of 111 years. Yet this difference is not statistically significant.
How good are the alternative institutional arrangements at coping with economic crises? When the economy declines during a particular year, parliamentary democracies die at the rate of 0.039: they can expect to live 26 years under such conditions. When the economy grows, their death rate is 0.007 and the expected life is 143 years. Hence parliamentary systems are vulnerable to economic crises. Presidential systems are less sensitive, but they die at much higher rates under any conditions. When the economy declines, they die at the rate of 0.064, with an expected life of 16 years. When the economy grows, they die at the rate of 0.042, with an expected life of 24 years. Democracy is vulnerable to economic crisis under either institutional system, but presidential systems are less likely to survive under good economic conditions than parliamentary systems are under bad conditions.
Statistical analyses provide even stronger evidence in favor of [End Page 46] parliamentarism. The expected life of presidential systems depends on the level of development, on economic growth, and on the presence of legislative majorities. Perhaps most startlingly, statistical analysis confirms that presidential systems are highly vulnerable to legislative-executive deadlocks. By contrast, in spite of the descriptive numbers cited above, for parliamentary systems neither the distribution of seats nor economic growth is a statistically significant predictor of the survival of democracy.
Statistics confirm as well that presidential regimes are less likely to survive in those countries that were not independent by 1950 (which is another way of saying “outside Latin America”), while parliamentary systems are equally likely to survive in either the “old” or the “new” countries. In turn, only parliamentary systems are sensitive to the ethnic fragmentation of the population. But this effect, while statistically significant, makes little difference for their expected lives. Thus presidential democracies are simply more brittle.
To summarize, the survival of democracies does depend on their institutional systems. Parliamentary regimes last longer, much longer, than presidential ones. Majority-producing electoral institutions are conducive to the survival of presidential systems: presidential systems facing legislative deadlock are particularly brittle. Both systems are vulnerable to bad economic performance, but presidential democracies are less likely to survive even when the economy grows than are parliamentary systems when the economy declines. The evidence that parliamentary democracy survives longer and under a broader spectrum of conditions than presidential democracy thus seems incontrovertible.
The choice of institutions. Since parliamentary democracies last longer, it is puzzling why so many democracies adopt presidentialism. What determines the initial choice of democratic institutions? Much of the answer can be gleaned from a casual glance at history. Countries that had monarchies but experienced no revolution transferred governmental responsibility from crown to parliament, ending up with parliamentary systems. Countries in which monarchy was abolished (France in 1848 and again in 1875, Germany in 1919) and colonies that rebelled against monarchical powers (the United States and Latin America in the late eighteenth and early nineteenth centuries) replaced monarchs with presidents. As Simon Bolívar once put it, “We elect monarchs whom we call presidents.” Countries that emerged from colonial domination after the Second World War typically inherited parliamentarism from the colonizers. Characteristically, however, these same countries instituted presidential systems if and when the initial democracy fell. Just as characteristically, democratizing dictatorships tended to retain presidentialism.
According to our count, among the 35 countries that democratized [End Page 47] between 1974 and 1990, 19 adopted presidential systems, 13 chose parliamentarism, and 3 opted for mixed systems. If the political stakes are indeed higher under presidentialism, it is hard to see why this system would emerge under conditions in which the political parties are perfectly well informed and not risk-prone. One explanation might be that the parties are unduly optimistic: each projects itself as a winner at the polls and assumes that it will gain the presidency. Still, we suspect that the choice of presidentialism is not just a decision of political parties.
Note that among the countries which were democratic at some time before the current transition to democracy, almost all chose the same system as the last time around. 23 This continuity, particularly in Latin America, may please those who like to find explanations in culture or traditions. 24 Yet it is more likely that it reflects the continuing political role of the military, which appears to have a preference for presidential regimes, perhaps because such regimes offer a clearer hierarchy. This reason is sufficient for the military to bargain for presidentialism when the issue of democratic institutions appears on the transitional agenda. The empirical patterns appear to support this expectation: while 10 of the 17 democratic regimes that emerged from civilian dictatorships went for presidentialism, an overwhelming 22 of the 28 democracies that surfaced from military dictatorships made the same choice. Thus presidentialism appears to be at least partly a legacy of military rule.
Once we learn that presidential systems are more likely to be adopted whenever the previous regime was military, the obvious question is whether all the findings concerning the longevity of presidential democracies are not spurious. Democracy may be more brittle under presidentialism precisely because this set of institutions is chosen where the military plays an active role in politics. To some extent this is the case: while the expected life of presidential democracies that emerge from civilian dictatorships is about 24 years, presidential systems that follow military dictatorships can expect to last only 17 years. Yet parliamentary democracies that follow military rule simply last much longer: 71 years. 25 Hence it would seem to be presidentialism per se that makes democracy more brittle.
Once we exclude the institutions inherited from the colonial rulers, the level of development at which the transition to democracy occurs does appear to have some impact on the institutions that are chosen: between 1950 and 1990, the average levels at which parliamentary institutions were chosen was $2,945, while presidential institutions were chosen at the average level of $2,584. The mode of transition, at least as indicated by strikes and other forms of social unrest (as coded by Arthur S. Banks), appears not to affect the choice of institutions. 26
We focus on the moment of transition since particular institutional frameworks tend to persist once established, as if “renegotiation-proof.” [End Page 48] As the recent Brazilian referendum rejecting a proposed change to parliamentarism shows, the difficulty of changing complex institutional arrangements is that the status quo, whatever it happens to be, is favored. If the proponents of change offer only a slogan, “parliamentarism,” then the defenders of the status quo can call for details of the new institutional arrangement; if the proponents of change offer such details, then the defenders can always find innumerable faults with the new system. During the entire period from 1950 to 1990, there were only three instances in which democratic regimes passed from one institutional system to another: France changed in 1958 from a parliamentary to a mixed system, while Brazil changed in 1960 from a presidential to a mixed system, only to return to presidentialism in 1963. Both cases occurred under exceptional circumstances. Countries that adopt presidential institutions when they transit to democracy are stuck with them. 27
Conclusions. Our central finding is the importance of economic factors in sustaining democracies. While the modernization theory was wrong in thinking that development under dictatorship breeds democracies, Lipset was correct to argue that once established in a wealthy country, democracy is more likely to endure. Indeed, we have found that once a country is sufficiently wealthy, with per-capita income of more than $6,000 a year, democracy is certain to survive, come hell or high water. And while international factors as well as political institutions are important for the durability of democracy in less affluent countries, economic performance does matter: indeed, democracy is more likely to survive in a growing economy with less than $1,000 per-capita income than in a country where per-capita income is between $1,000 and $4,000, but which is declining economically. Democracies can survive even in the poorest nations if they manage to generate development, if they reduce inequality, if the international climate is propitious, and if they have parliamentary institutions.
For a variety of reasons, however, this is not an optimistic conclusion. Poverty is a trap. Few countries with annual per-capita income below $1,000 develop under any regime: their average rate of growth is less than 1 percent a year; many experience prolonged economic decline. When poor countries stagnate, whatever democracies happen to spring up tend to die quickly. Poverty breeds poverty and dictatorship.
Institutional choice offers a partial escape from this trap: parliamentary systems in the poorest countries, while still very fragile, are almost twice as likely to survive as presidential democracies, and four times as likely when they grow economically. Yet since it appears that poor countries are more likely to choose presidentialism, little solace is offered by the possibility of institutional engineering. Equally little solace is offered by political learning. Most countries returning to [End Page 49] democracy usually go back to whatever constitution they had in the past, even if it never worked, as in Argentina, where the first democratic alternation in office under the revived 1853 Constitution already violated its letter.28
Finally, we find no evidence of “consolidation.” A democracy becomes “consolidated” if its aforementioned “hazard rate” declines with its age, so that, as Robert Dahl has argued, democracies are more likely to survive if they have lasted a while. 29 We find some evidence that this is true, but also that democracies are heterogeneous. Once we control for the level of development, the heterogeneity disappears and the hazard rates become independent of age, meaning that for a given level of development, democracies are about equally likely to die at any age. Since democracies are much more likely to survive when they occur in developed countries, these findings would indicate that hazard rates (uncorrected for the level of development) drop because countries develop economically, and not because a democracy that has been around is more likely to continue being around.
Clearly, we do not think that “consolidation” is just a matter of time, of some kind of “habituation” or mechanical “institutionalization.” 30 We discovered that democracies are more likely to survive at higher levels of development. But we also found that democracies survive if they generate economic growth and if they control distributional pressures by allowing some inflation and reducing income inequality. This is not to deny that institutions matter: in fact they do, and not just parliamen-tarism and electoral systems but others that we have left out of consideration because we lack data. Democracy’s ability to survive is a matter of politics and policy, as well as luck. Yet, conversely, if democracies become “consolidated” for whatever reasons, then we should observe that at any level of development the mere passage of time makes their demise less likely. This, however, we do not observe, and so conclude that “consolidation” is an empty term.
In sum, the secret of democratic durability seems to lie in economic development—not, as the theory dominant in the 1960s had it, under dictatorship, but under democracy based on parliamentary institutions.
Adam Przeworski is professor of political science at New York University.
Michael Alvarez is assistant professor of political science at De Paul University.
José Antonio Cheibub is assistant professor of politics at the University of Pennsylvania.
Fernando Limongi is assistant professor of political science at the University of São Paulo.
Appendix. Classifying Regimes
We define democracy as a regime in which governmental offices are filled as a consequence of contested elections. Only if the opposition is allowed to compete, win, and assume office is a regime democratic. To the extent to which it focuses on elections, this is obviously a minimalist definition.
This definition has two parts: “offices” and “contestation.” In no regime are all governmental offices filled as a consequence of elections. What is essential to consider a regime as democratic is that two kinds of offices are filled by elections: the chief executive office and the seats in the effective legislative body.
Contestation occurs when there exists an opposition that has some chance of winning office as a consequence of elections. We take quite literally Przeworski’s [End Page 50] dictum that “Democracy is a system in which parties lose elections”: whenever in doubt, we classify as democracies only those systems in which incumbent parties actually did lose elections. Alternation in office constitutes prima facie evidence of contestation.
Contestation, in turn, entails three features: 1) ex ante uncertainty, 2) ex post irreversibility, and 3) repeatability.
By “ex ante uncertainty,” we mean that there is some positive probability that at least one member of the incumbent coalition can lose office in a particular round of elections. Uncertainty is not synonymous with unpredictability: the probability distribution of electoral chances is typically known. All that is necessary for outcomes to be uncertain is that some incumbent party could lose.
By “ex post irreversibility” we mean the assurance that whoever wins elections will be allowed to assume office. The outcome of elections must be irreversible under democracy even if the opposition wins. The practical consequence of this feature is to exclude sham elections as well as periods of liberalization. Liberalization is typically intended by dictatorial regimes to be a controlled opening of the political space. When it fails—that is, when the opposition does win—a clampdown sometimes follows. Hence there is no certainty that the opposition would be able to celebrate its victory.
The final feature of contestation is that elections must be expected to be repeated. Whoever wins the current round of elections cannot use office to make it impossible for the competing political forces to win next time. Democracy, as Juan Linz once said, is government pro tempore. All political outcomes must be temporary: losers do not forfeit the right to compete in the future, to negotiate again, to influence legislation, to pressure the bureaucracy, or to seek recourse to courts. Even constitutional provisions are not immutable; rules, too, can be changed according to rules.
Operationally, a regime was classified as a democracy if it did not fail under any of the four rules listed below. (Our timing rules are as follows: We code the regime that prevailed at the end of the year, even if it came to power on December 31, as, for example, dictatorship arrived in Nigeria in 1983. Transitions to authoritarianism are signaled by a coup d’état. Transitions to democracy are dated by the time of the inauguration of the newly elected government, not of the election. In the few cases, like those of the Dominican Republic in 1963, where a democratic regime lasted six months, or Bolivia in 1979, where the situation changed several times, the information about regimes that began and ended within the same year is lost.) A regime is classified as a dictatorship if at least one of these conditions holds:
Rule 1: “Executive Selection.” The chief executive is not elected.
Rule 2: “Legislative Selection.” The legislature is not elected.
Rule 3: “Party.” There is no more than one party. Specifically, this rule applies if 1) there were no parties or 2) there was only one party, or 3) the current tenure in office ended up in the establishment of nonparty or single-party rule, or 4) the incumbents unconstitutionally closed the legislature and rewrote the rules in their favor. Alternation in office overrides the party rule: Jamaica, where a single party at one time held 100 percent of the seats in the legislature yet subsequently yielded office after losing an election, was classified as democratic during the entire period.
These three rules are not sufficient, however, to classify those regimes which repeatedly hold elections, allow varying degrees of freedom for the opposition, and always win. There are some regimes which cannot be unambiguously classified on the basis of all the evidence produced by history: we have no way of telling whether the incumbents would have held elections if they were not certain to win. In such cases we must decide which error we prefer to avoid: classifying as democracies regimes that may not be ones or rejecting as democracies regimes that may in fact be ones. Err we must; the only question is which way. We decided to err on the conservative side, disqualifying as democracies regimes that pass the previous three rules but not the following: [End Page 51]
Rule 4: “Type II Error.” The incumbents held office in the immediate past by virtue of elections for more than two terms or without being elected, and until today or the time when they were overthrown they have not lost an election.
Throughout this discussion, we have focused on democracy. We treat dictatorship simply as a residual category, perhaps better denominated as “not democracy.” Since we are often told that democracy “is” a continuous variable, here are the reasons we insist on dichotomizing political regimes: 1) While some democracies are more democratic than others, unless offices are contested, no regime should be considered democratic. Kenneth A. Bollen and Robert W. Jackman, in their 1989 American Sociological Review essay “Democracy, Stability, and Dichotomies,” confuse the argument that some democracies are more democratic than others with the claim that one can distinguish the degree of “democracy” for any pair of regimes. 2) The idea that we should, as Bollen and Jackman suggest in their discussion of “borderline cases,” place the cases that cannot be unambiguously classified given our rules into an “intermediate” category, halfway between democracy and dictatorship, strikes us as ludicrous. 3) “Borderline cases” constitute either systematic or random errors. Systematic errors can be treated by explicit rules, such as our Type II Error rule, and their consequences can be examined statistically. Once this decision is made, the classification is unambiguous. 4) In turn, some errors random with regard to the rules will remain and we have to live with them. But there are no a priori reasons to think that a more refined classification will have a smaller measurement error. A finer scale generates smaller errors but more of them, a rougher scale generates larger errors but fewer of them. If the distribution of true observations is unimodal and close to symmetric, a more refined classification will have a smaller error, but in fact observations on all the polychotomous scales tend to be U-shaped, which advantages a dichotomous classification.
Whatever the peculiarities of our rules, the resulting classification differs little from alternative approaches: the Coppedge-Reinecke scale for 1978 predicts 92 percent of our regimes, the Bollen 1965 scale predicts 85 percent, and the Gurr scales of autocracy and democracy for 1950–86 jointly predict 91 percent. The Gastil scale of political liberties, covering the period from 1972 to 1990, predicts 93.2 percent of our classification; his scale of civil liberties predicts 91.5 percent; and the two scales jointly predict 94.2 percent of our regimes. Hence there is no reason to think that our results are idiosyncratic in the particular classification of regimes.
Since the distinction between parliamentary and presidential systems is uncontroversial, we state it only briefly. In parliamentary systems, the legislative assembly can dismiss the government, while under presidential systems it cannot. This criterion coincides perfectly with the mode of selection of the government: by the legislature in parliamentary systems, by the voters (directly or indirectly) in presidential systems. Within each type of institutional design there are important differences. Most important among these differences is the electoral system, some varieties of which may or may not be prone to generate legislative majorities.
Some institutional arrangements, however, do not fit either pure type: they are “premier-presidential,” “semipresidential,” or “mixed,” according to different terminologies. In such systems, the president is elected for a fixed term and has some executive powers, but governments serve at the discretion of the parliament. These “mixed” systems are not homogeneous: most lean closer to parliamentarism insofar as the government is responsible to the legislature; others, notably Portugal between 1976 and 1981, grant the president the power to appoint and dismiss governments and therein lean closer to presidentialism.
Among the 135 countries which are included in our sample, there were 50 parliamentary democracies, 46 presidential, and 8 mixed. Outside the Americas, there were nine presidential democracies: Congo (1960–62), Ghana (1979–80), Nigeria (1979–82), Uganda (1980–84), Bangladesh (1986–90), South Korea (1988-present), Pakistan (1972–76), and the Philippines (before 1964 and then from 1986 to the present). [End Page 52]
In Latin America, the only parliamentary regimes were the short-lived attempt in Brazil, preceding the 1964 coup, and Suriname. Most West European countries have parliamentary systems, but parliamentary democracies can also be found in most other parts of the world.
* This essay is a revised version of a paper presented at a conference on “Consolidating Third Wave Democracies: Trends and Challenges,” held in Taipei, Taiwan, on 27–30 August 1995. The work was supported in part by National Science Foundation grant SES-9022605.
1. Most of the political data were collected by the authors, but some are taken from Arthur S. Banks, Cross-National Time-Series Data Archive (Binghamton, N.Y.: Center for Social Analysis, State University of New York at Binghamton, magnetic tape, 1993). They are described in Alvarez, Cheibub, Limongi, and Przeworski, “Classifying Political Regimes for the ACLP Data Set” (Working Paper No. 3, University of Chicago Center on Democracy, 1994). Most of the economic data are derived from Penn World Tables, version 5.6; other data are from the World Bank and the International Monetary Fund. We refer to this collection of data as the ACLP data base. Saudi Arabia and the five Persian Gulf states were excluded because oil revenues accounted for more than 50 percent of their GDP most of the time.
2. These numbers add up to 104 democratic institutional systems since there were three democratic regimes that changed their institutional framework without passing through a dictatorial spell.
3. Adam Przeworski and Fernando Limongi, “Democracy and Development” (paper presented at the Nobel Symposium on Democracy, Uppsala, Sweden, 27–30 August 1994). For divergent assessments of how regimes affect growth, see the overview presented in Przeworski and Limongi, “Political Regimes and Economic Growth,” Journal of Economic Perspectives 7 (1993): 51–69. For recent econometric evidence, see John F. Helliwell, “Empirical Linkages Between Democracy and Economic Growth,” British Journal of Political Science 24 (1993): 225–48; Robert J. Barro, “Democracy and Growth” (Working Paper No. 4909, National Bureau of Economic Research, Cambridge, Mass., 1994).
4. All figures for annual per-capita income are expressed in purchasing power parity (PPP) U.S. dollars in 1985 international prices, as given by version 5.5 of the Penn World Tables. In some cases, these numbers differ significantly from the 5.6 release, used in the remainder of this paper to measure the “level of development.”
5. The results reported in this paragraph are treated at length in Adam Przeworski and Fernando Limongi, “Modernization: Theories and Facts” (Working Paper No. 8, University of Chicago Center on Democracy, 1995).
6. After the fact, it may appear that development led to democracy. Suppose that we observe a dictatorship with a per-capita income of $2,000 a year in a country that grows at 2.5 percent per year. Assume further that at $2,000 any dictatorship faces each year the same risk of dying, equal to 0.025. If this dictatorship died exactly 28 years after its birth, at $4,000, we would be tempted to attribute its demise to development. But this dictatorial regime would have had a 50 percent cumulative chance of making it all the way to $4,000 even if the marginal chance of surviving (the hazard rate) was exactly the same at $4,000 as at $2,000. Conversely, take Spain, which we observe for the first time in 1950 at $1,953 per-capita income and which grew under the dictatorship at the average rate of 5.25 percent per annum, to reach $7,531 by 1976. Suppose that the Spanish dictatorship faced during the entire period a 0.03 chance of dying during each year, so that, assuming an exponential hazard function, it had about a 50 percent chance of not being around by 1974 even if it had not developed at all.
7. Samuel P. Huntington, Political Order in Changing Societies (New Haven: Yale University Press, 1968); Samuel P. Huntington and Joan Nelson, No Easy Choice: Political Participation in Developing Countries (Cambridge: Harvard University Press, 1976); and Guillermo O’Donnell, Modernization and Bureaucratic-Authoritarianism: Studies in South American Politics (Berkeley: Institute of International Studies, University of California, 1973).
8. In addition to the transitions in Argentina in 1955, 1962, 1966, and 1976, they occurred in Chile in 1973, Uruguay in 1973, Suriname in 1980, Turkey in 1967, and Fiji in 1987.
9. Seymour Martin Lipset, “Some Social Requisites of Democracy: Economic Development and Political Legitimacy,” American Political Science Review 53 (1959): 56. Our best guess is that the European countries which succumbed to fascism between the wars had per-capita incomes not higher than $2,000 in the 1985 international prices. See Przeworski and Limongi (1995).
10. Lipset, Political Man: The Social Bases of Politics (Baltimore: Johns Hopkins University Press, 1981 [orig. publ. 1960]), esp. 27–63, 459–76, and 488–503; Mancur Olson, “Rapid Growth as a Destabilizing Force,” Journal of Economic History 23 (1963): 453–72; and Huntington, Political Order in Changing Societies.
11. This finding parallels again the results of John B. Londregan and Keith T. Poole with regard to coups, which they found to be less likely when the economy grows. See Londregan and Poole, “Poverty, the Coup Trap, and the Seizure of Executive Power,” World Politics 42 (1990): 151–83.
12. Larry Diamond and Juan J. Linz, “Introduction: Politics, Democracy, and Society in Latin America,” in Diamond, Linz, and Lipset, eds., Democracy in Developing Countries: Latin America (Boulder, Colo.: Lynne Rienner, 1989).
13. Hirschman’s argument was that a moderate rate of inflation allows governments to pacify the most militant groups. See “The Social and Political Matrix of Inflation: Elaborations on the Latin American Experience,” in Essays in Trespassing: Economics to Politics and Beyond (New York: Cambridge University Press, 1981), 177–207.
14. Edward N. Muller, “Democracy, Economic Development, and Income Inequality,” American Sociological Review 53 (1988): 50–68.
15. Note again the parallel finding of Londregan and Poole in “Poverty, the Coup Trap, and the Seizure of Executive Power” that coups breed coups.
16. Juan J. Linz, “The Perils of Presidentialism,” Journal of Democracy 1 (Winter 1990): 51–69 and “The Virtues of Parliamentarism,” Journal of Democracy 1 (Fall 1990): 84–91.
17. Juan J. Linz, “Democracy: Presidential or Parliamentary—Does It Make a Difference?” (paper prepared for the Workshop on Political Parties in the Southern Cone, Woodrow Wilson International Center for Scholars, Washington, D.C., 1984). Linz’s claim is disputed by Scott Mainwaring and Matthew Shugart, “Juan Linz, Presidentialism, and Democracy: A Critical Appraisal” (Working Paper No. 200, Helen Kellogg Institute for International Studies, University of Notre Dame, 1993).
18. Walter Bagehot, “The English Constitution: The Cabinet,” in Arend Lijphart, ed., Parliamentary Versus Presidential Government (Oxford: Oxford University Press, 1992), 18. Woodrow Wilson’s 1884 essay “Committee or Cabinet Government?”—reprinted in the same volume—makes an argument similar to Bagehot’s. Also to be found in Lijphart’s collection is an analysis of the U.S. political structure done to mark the 1976 Bicentennial by the U.S. Committee on the Constitutional System. The Committee notes that “The separation of powers, as a principle of constitutional structure, has served us well in preventing tyranny and the abuse of high office, but it has done so by encouraging confrontation, and deadlock, and by diffusing accountability for the results.”
19. Note that throughout we refer only to the share of the largest party in the legislature, whether or not it has been the same as the party of the president. In the United States, since 1968, the control of at least one house of the Congress has rested in the hands of the party other than that of the president 80 percent of the time.
20. Mainwaring counted democratic breakdowns since 1945, finding 27 under presidentialism, 19 under parliamentarism, and 4 under other types. “Presidentialism, Multipartism, and Democracy: The Difficult Combination,” Comparative Political Studies 26 (1993): 198–228.
21. Mainwaring, “Presidentialism in Latin America,” Latin American Research Review 25 (1990): 157–79. Alfred Stepan and Cindy Skach, “Meta-Institutional Frameworks and Democratic Consolidation,” World Politics 46 (1993): 1–22.
22. Mainwaring and Shugart, in “Juan Linz, Presidentialism, and Democracy,” take issue with Linz: in their view it is majoritarian parliamentarism, rather than presidentialism, that increases the political stakes. Yet even if majoritarian parliamentary systems last shorter than minoritarian ones, parliamentary democracies of any kind last longer than presidential regimes. Whether this difference is due to the intensity of political conflicts, however, we do not know.
23. Only Pakistan went from parliamentarism in 1950–55 to presidentialism in 1972–76 and back to parliamentarism in 1988. Only Ghana, Nigeria, South Korea (which was a parliamentary democracy for one year in 1960), and Turkey chose a presidential system after having experienced parliamentary democracies. Lastly, only Suriname opted for a mixed system after having experienced democratic presidentialism.
24. This is the argument of the Nigerian Constitution Drafting Committee of 1976: “The tendency indeed of all people throughout the world is to elevate a single person to the position of ruler. In the context of Africa the division [of powers] is not only meaningless, it is difficult to maintain in practice. No African head of state has been known to be content with the position of a mere figurehead.” See the Committee’s report in Lijphart, Parliamentary Versus Presidential Government.
25. Only two parliamentary democracies emerged from a civilian dictatorship and died before 1991. Their expected life is 22 years, but the tiny number of countries involved greatly diminishes confidence in this number.
26. See Banks, Cross-National Time-Series Data Archive. Among the 35 transitions that occurred after 1973, parliamentary institutions were chosen in 13 cases at the average level of $3,414, presidential institutions in 19 cases at the average level of $2,591, making the effect even more pronounced.
27. This is not to argue that countries that have adopted presidential institutions during recent transitions to democracy should immediately attempt to move to parliamentarism. Whenever institutional choice is present on the political agenda, substantive conflicts, even minor ones, tend to spill over to institutional issues. Such situations are dangerous for democracy, since they signify that there are no clear rules by which substantive conflicts can be terminated. Hence having a clear and stable institutional system is more important than having a perfect one. We owe this observation to Hyug Baeg Im.
28. The 1853 Constitution sets the period between the election and the inauguration at nine months because that is how long it took electors to travel from the interior to Buenos Aires. The transfer of office from President Raúl Alfonsín to the President-elect Carlos Menem was shortened as a result of a mutual agreement under the pressure of an inflationary crisis.
29. Robert A. Dahl, “Transitions to Democracy” (address delivered to the symposium on “Voices of Democracy,” University of Dayton, Center for International Studies, 16–17 March 1990).
30. Guillermo O’Donnell, “Partial Institutionalization: Latin American and Elsewhere” (paper presented at the conference on “Consolidating Third Wave Democracies: Trends and Challenges,” Taipei, Taiwan, 27–30 August 1995), implies that “institutionalization” can be understood in two ways: either as a process of gradual stabilization of expectations that a particular institutional system will orient political actions or as an increasing fit between formal institutions and real practices. If “institutionalization” is taken in the first sense, it is tautologically related to “consolidation.” But whether democracy can survive when the formal institutions do not describe real practices is an empirical question.