Classifying Political Regimes in Latin America, 1945-1999"

This article is about how political regimes should generally be classified, and how Latin American regimes should be classified for the 1945-99 period. We make five general claims about regime classification. First, regime classification should rest on sound concepts and definitions. Second, it should be based on explicit and sensible coding and aggregation rules. Third, it necessarily involves some subjective judgments. Fourth, the debate about dichotomous versus continuous measures of democracy creates a false dilemma. Neither democratic theory, nor coding requirements, nor the reality underlying democratic practice compel either a dichotomous or a continuous approach in all cases. Fifth, dichotomous measures of democracy fail to capture intermediate regime types, obscuring variation that is essential for studying political regimes. This general discussion provides the grounding for our trichotomous ordinal scale, which codes regimes as democratic, semi-democratic or authoritarian in nineteen Latin American countries from 1945 to 1999. Our trichotomous classification achieves greater differentiation than dichotomous classifications and yet avoids the need for massive information that a very fine-grained measure would require.

If a regime classification is intended primarily to measure democracy, as is the case with ours and many others, it should be hinged on a definition of democracy. We define a democracy as a regime (1) that sponsors free and fair competitive elections for the legislature and executive; (2) that allows for inclusive adult citizenship; (3) that protects civil liberties and political rights; and (4) in which the elected governments really govern and the military is under civilian control. This minimalist procedural definition contrasts with nonprocedural definitions such as Bollen's (1980Bollen's ( , 1991 and with subminimal procedural definitions such as Schumpeter's (1947) and Przeworski et al.'s (2000).
Our second building block is explicit coding and aggregation rules for classifying regimes. Without such rules, other researchers cannot understand the procedures used to classify the regimes and the classification will be vulnerable to serious problems of reliability. Our regime classification is based on first disaggregating the concept of democracy into the four defining criteria discussed in the previous paragraph and then reaggregating to form an overall regime assessment.
Following this general theoretical and methodological discussion, we classify the political regimes in nineteen Latin American countries from 1945 to 1999. Our trichotomous classification places regimes into democratic, semidemocratic, and authoritarian categories by building on all four dimensions of our definition of democracy. This trichotomy achieves greater differentiation than dichotomous classifications and yet avoids the need for massive information that a fine-grained continuous measure would require.
Our endeavor is predicated on two additional beliefs about regime classification. First, this enterprise demands some subjective judgments about the nature of political regimes. By "subjective" we do not mean arbitrary, but rather informed judgment based on knowledge of the cases and guided by explicit coding rules. An assessment limited to elections leaves out some elements that are essential to a democracy, producing a subminimal definition. Our viewpoint challenges recent arguments advanced by Adam Przeworski and his collaborators (2000) in their seminal project, as well as Vanhanen (1990), who argues for purely objective measures of democracy.
Second, we argue that dichotomous classifications are insufficiently sensitive to regime variations because many regimes fall into an intermediate semidemocratic zone. An ordinal, trichotomous classification--democracy, semi-democracy, and nondemocracy or authoritarian--better captures the significant variations in regimes.
The final section of the article compares our trichotomous measure with the three most widely used measures that provide annual ratings of democracy over a long period of time: Freedom House, Polity III, and Przeworski et al. (2000). We point out some deficiencies of the existing measures and argue that our trichotomous classification is a useful alternative.
Classifying political regimes for an intermediate number of countries and an extended period of time is a labor-and time-intensive enterprise. Both because of the time demands and because regime classification is usually a means toward an end, there would be no reason to take on that enterprise unless exist-ing classifications had some shortcomings. We undertake this classification because of a conviction that the existing ones that provide annual democracy measures over a long period of time have flaws that require more than piecemeal reform. Compared to the existing measures, our classification yields different substantive results on several questions: how pervasively authoritarian Latin America was before 1978, how profound the change between earlier decades and the post-1978 wave of democratization has been, and whether the region suffered a minor democratic erosion in the 1990s.
Without careful regime classification, it is impossible to adequately study important substantive issues related to political regimes. For example, any attempt to assess whether democracy in Latin America suffered a minor decline in the 1990s (Diamond 1996(Diamond , 1999) must rest on careful regime measurement. Classifying regimes is a necessary step to asking important questions about the causes and consequences of different regimes and of transitions from one kind of regime to another. For decades, these have been leading questions in comparative politics and political sociology, and they are likely to remain at the center of intellectual debates for decades to come. If we cannot measure democracy, these enterprises are impossible. Moreover, we believe that it will serve the scholarly community to bring together and critically reflect on different approaches to regime classification in Latin America.

Defining Democracy
Our challenge is to build on the positive contributions of earlier classifications while attempting to address their shortcomings. Following , Coppedge and Reinecke (1990), and Hadenius (1992), we believe that the first step in classifying political regimes should be defining them; it is impossible to classify whether a regime is a democracy, or how democratic it is, until we know what a democracy is. Much has been written on this subject, but definitions of democracy still vary widely. The first part of this section proposes a definition of democracy that is minimal; in the second part we argue that this definition is complete. Following Sartori (1976: 60-64), we advocate minimal definitions. "A definition is minimal when all the properties or characteristics of an entity that are not indispensable for its identification are set forth as variable, hypothetical properties--not as definitional properties. This is the same as saying that whatever falls beyond a minimal characterization is left to verification--not declared true by definition" (p. 61).
Modern representative democracy has four defining properties. The first two involve the classic dimensions analyzed in Dahl's renowned work (1971) and in many other discussions. First, the head of government and the legislature must be chosen in open and fair competitive elections. 1 Such elections are a core ingredient of modern representative democracy. Fraud and coercion may not determine the outcomes for democratic elections. Elections must offer the possibility of alternation in power even if, as occurred for decades in Japan, no actual alternation occurs for an extended time.
Second, today the franchise must include the great majority of the adult population. This means something approximating universal adult suffrage for citizens, but many countries have minor exclusions (the insane, convicts) that do not detract from their democratic credentials. If large parts of the population are excluded, the regime may be a competitive oligarchy, but in the past few decades, it could not be considered a democracy.
Although this criterion seems obvious for contemporary times, less clear is how rigidly one should apply it to the past. Especially with respect to the dimension of participation, a critical and yet under-theorized issue in classifying democracies is whether scholars should use international standards for a given period (we call this a retrospective standard) or today's international standards. Most scholars (e.g., Gurr et al. 1990;Huntington 1991;Przeworski et al. 2000) implicitly use a retrospective standard; they view regimes based on nearly adult male suffrage as democratic. Recently, however, Paxton (2000) has persuasively argued that there is a contradiction between most definitions of democracy, which call for universal suffrage, and most operationalizations, which are based on adult male suffrage for earlier periods. She notes that regimes based on nearly universal adult suffrage are more democratic than those based on nearly universal adult male suffrage, and calls for continuous measures of democracy to capture this qualitative difference.
With an ordinal classification such as that proposed here, scholars can legitimately use either the retrospective or contemporary standards. But they should be clear about which they are using. Each mode of classification has an advantage and a disadvantage, and the question of both whether a regime was democratic by standards of the time and how democratic it is by contemporary standards is legitimate. The retrospective standard fails to capture some changes over time; it is predicated on the idea that democracy is an ever-changing type of political regime. Using today's standards to judge earlier regimes makes it easier to capture changes in how democracy is perceived and practiced, but it imposes an anachronism. As democratic rights expand, each new generation would have doubts about classifying all earlier regimes as democratic because earlier generations of democrats had not institutionalized or even conceived of some rights and practices.
On the question of inclusion, we use a retrospective standard for earlier regimes to avoid the problems of anachronism. Until shortly after World War II, we consider some countries democratic even if women had not yet gained the right to vote. In a similar vein, the non-enfranchisement of the illiterate did not automatically prevent us from coding a regime as democratic. Our criteria for judging inclusiveness are to some degree historically contingent because democracy itself is ever changing (Markoff 1996)--although democracy also embodies a few core unchanging principles such as free and fair elections and respect for basic civil liberties. The breadth of the franchise does not filter out many regimes in the contemporary world, as standards for inclusion have become quite universal in modern democracies.
Third, democracies must protect political rights and civil liberties such as freedom of the press, freedom of speech, freedom to organize, the right to habeas corpus, etc. Even if the government is chosen in free and fair elections with a broad suffrage, in the absence of an effective guarantee of civil liberties it is not democratic, as that word is understood in the modern world. E1 Salvador and Guatemala in the 1980s, among other cases in contemporary Latin America, illustrate the point. A liberal component--the protection of individual liberties--is a necessary element of contemporary democracy. Because the liberal dimension is a defining characteristic of contemporary democracy, Diamond's (1999: 42-51), Merkel's (1999), and Zakaria's (1997) concept of "illiberal democracy" is problematic; it suggests that regimes that do not protect civil liberties and political rights might still be called democracies (Plattner 1998). 2 Illiberal regimes with competitive elections are semi-democratic at best and in some cases downright authoritarian.
Fourth, the elected authorities must have the real governing power, as opposed to a situation in which elected officials are overshadowed by the military or by a non-elected shadow figure (Valenzuela 1992). If elections are free and fair but produce a government that cannot control major policy arenas because the military or some other force does, then the government is not a democracy. By our stringent definition, some of the "defective democracies" of which Merkel (1999) speaks are not merely defective; they are not democracies.
All four elements of our definition are necessary and must be included even if doing so requires making subjective judgments about regimes. This definition meets the test of being minimal but complete if (a) all four criteria are necessary components of democracy, without which a regime should be not considered democratic, and (b) no other discrete features are necessary to characterize a democracy. We believe that this definition meets both conditions, while some recent definitions fail the second by neglecting some essential characteristics of a democracy.
Our definition is focused on procedure but adds a concern for civil liberties and effective governing power. It is close to that proposed by many scholars (Collier and Levitsky 1997;Diamond, Linz, and Lipset 1989: xvi-xviii;O'Donnell 2001). It diverges from nonprocedural definitions that do not explicitly refer to elections (e.g., Bollen 1991: 5) and from Schumpeter's classic definition and Przeworski et al.'s (2000) more recent one, both of which are subminimal.
Finally, following Przeworski et al. (2000), it is useful to make explicit two points that our definition does not include. First, it leaves out substantive resuits such as social equality. We limit the definition of democracy to procedural issues and leave as an empirical question the relationship between democracy and equality. It muddles the picture to include social equality in the definition of democracy, even though high levels of inequality might well work against democracy. Second, again following Przeworski et al., our definition says nothing about accountability, 3 defects in the rule of law except those that impinge on civil liberties and political rights, and instances of "decretismo." Adding such criteria that are not inherent in the nature of democracy leads to a nonminimal definition.

The Perils of a Subminimal Definition
Our definition bears one important similarity to Schumpeter's (1947) and to Przeworski et al.'s (2000): all three focus on procedures. In insisting on the second, third, and fourth dimensions of democracy, however, our definition differs from that used by Schumpeter and Przeworski et al. Both equate democracy with holding free and fair elections that allow for an alternation in power regardless of the lack of civil liberties or the presence of "reserved domains" in public policy that are under the control of unelected figures (Valenzuela 1992). 4 Przeworski et al. (2000) carefully articulate their criteria for coding and hinge their regime classifications on a definition of democracy and non-democracy. On the other hand, they use a subminimal definition of democracy, which results in counting some authoritarian regimes as democracies. Their definition of democracy (p. 15) revolves exclusively around competition for office. "Democracy ... is a regime in which those who govern are selected through contested elections." More specifically, in a democracy, the head of government and the legislature must be elected, and there must be more than one party.
Przeworski et al. are not alone in defining democracy strictly in terms of competition for office. In his classic work, Schumpeter (1947: 269) also focused on electoral competition among political elites and parties: "the democratic method is that institutional arrangement for arriving at political decisions in which individuals acquire the power to decide by means of a competitive struggle for the people's vote." In the past two decades, however, most political scientists have used a more expansive and less parsimonious definition, more akin to the one we presented above.
Przeworski et al. represent an outlier in defining democracy so parsimoniously, but the prominence of the scholars and the project makes them an outlier to take seriously. They argue that classifications of political regimes should follow "an exclusive reliance on observables rather than on subjective judgments" (Alvarez et al. 1996: 3). Yet this distinction between "observables" and subjectivity is drawn too sharply; the authors understate the subjectivity involved in their own assessments of whether elections are free and fair. Regimes should be classified according to "observables," but social scientists must make judgments about whether an infringement is sufficiently serious as to regard a regime as less than democratic. Moreover, relying on observables need not restrict a definition of democracy to the electoral sphere. The state of human rights and civil liberties, the breadth of participation, and the degree to which nondemocratic actors have veto power over government policy are all observables. It is often more difficult to get reliable data on the human rights situation and on whether the elected government really governs than on the fairness of elections. It is a mistake, however, to classify political regimes without making judgments about respect for civil liberties and whether the elected officials really govern.
An exclusive focus on political competition leads to a subminimal definition and to errors in regime classification. As Karl (1986) has argued, "electoralism"--the equating of competitive elections with democracy--misses fundamental dimensions of democracy. Competitive elections without broad adult suffrage can exist in an oligarchic pre-democratic regime or in a racially or ethnically exclusive regime that excludes the majority of the population (e.g., South Africa before the end of apartheid). But if a contemporary government is elected in contests that exclude most of the adult population, this exclusion violates the root meaning of democracy--rule by the people. However far the definition of representative democracy has strayed from this original root meaning, it is essential to preserve the idea that democratic governments are elected by the people. In practical terms, this second criterion is not very discriminating in the post-1974 wave of democratization because almost every country that has sponsored free and fair competitive elections has had a broadly inclusive franchise.
Przeworski et al. (Alvarez et al. 1996) explicitly rejected using subjective judgments about civil liberties in classifying regimes. 5 Yet without respect for the core civil liberties traditionally associated with democracy, a regime is not democratic as we understand that word today. Without protection of civil liberties, the electoral process itself is vitiated. Elections are not free and fair if the opposition risks reprisals for criticizing the government, opposing points of view are not permitted any outlet or dissemination, political parties cannot form or meet, journalists cannot publish freely, candidates are not permitted to travel, and so on. Przeworski et al. (2000) attempt to correct for this shortcoming in their definition by counting as democratic only those regimes in which there has been at least one alternation in power, thus evaluating the effectiveness of the electoral process in retrospect. "Whenever a ruling party eventually suffered an electoral defeat and allowed the opposition to assume office, the regime is classified as democratic for the entire period this party was in power under the same rules" (p. 24). But this criterion is both over-and under-inclusive. The violations of civil liberties or political rights may be directed at one political viewpoint--even a dominant one--but still leave the electorate some choices, thus producing the required alternation in office without ever permitting free and fair elections. This occurred in Argentina between 1958 and 1966, when the Peronist party was proscribed, denying people the opportunity to vote for the most popular party.
In addition, even if alternation in power showed that elections are free and fair at time t, it would hardly be an indicator of their fairness at t-1. Consider, for instance, the case of Jamaica. Given the alternation in power in 1989, Przeworski et al. coded Jamaica retroactively as a democracy for the whole 1962-90 period even though the ruling Jamaica Labor Party (JLP) manipulated the electoral calendar and ran virtually unopposed in the 1983 election. The JLP controlled all seats in parliament between 1984 and 1989. This problem is important in some Latin American countries as well. Under these coding rules, given the alternation in power in 2000, Mexico's elections would presumably be considered free and fair during the late twentieth century and perhaps even earlier. Yet the fact that the 2000 elections permitted an alternation in power says nothing about the fairness of previous elections.
The rule sometimes generates the opposite problem by excluding from the democratic category countries in which the liberties that underlie free and fair elections are present, but the electorate is satisfied with the party in power. Japan was a democracy for decades before there was an alternation in power.
But under the alternation rule, it is unclear whether Japan would have qualified as a democracy had Przeworski et al. coded this case a few years earlier. The Liberal Democratic Party (LDP) never left power between 1955 (when the party was born) and 1990 (the last year in the Przeworski et al. dataset). The presence of a dominant party that consistently wins elections for a few decades is not inherently incompatible with democracy.
Przeworski et al. claim that we should empirically investigate whether freedom from arbitrary violence is associated with democracy rather than include such a freedom in the definition (Alvarez et al. 1996: 20). But one of the defining characteristics of modern representative democracy is its liberal dimension: limited government and adherence to some core values about the sanctity of the individual. Some competitively elected governments have seriously infringed upon such rights. In the early Hitler years, the competitively elected government was already beginning its systematic campaign to deny basic rights to certain groups. Similarly, in parts of Colombia (1980s to the present) and Peru (1980s and early 1990s), governmental or paramilitary campaigns against guerrillas and drug trafficking have meant a less than democratic experience for peasants caught in the middle. When these conditions are generalized, affecting a large portion of the population, the country should not be labeled a democracy. In short, it is important to distinguish between illiberal elected governments and liberal democracies.
In addition, a government is not democratic unless the elected officials actually govern. Przeworski et al. explicitly reject this criterion and maintain that regime classifications should not be based on judgments about the actual exercise of power: "In some democracies (Honduras and Thailand are prototypes) civilian rule is but a thin veneer over military power, exercised by defrocked generals. Yet as long as office holders are elected in elections that someone else has some chance to win and as long as they do not use the incumbency to eliminate the opposition, the fact that the chief executive is a general or a lackey of general does not add any relevant information" (2000: 35).
We are skeptical about this argument when "civilian rule is but a thin veneer over military power." If the government elected by the people does not actually govern, it is not democratic. The contrary argument appears to be premised on the assumption that there is an option, that people can choose between the lackey of a general and someone who is not a lackey. But in some cases, all the candidates are not so much lackeys as hostages. In these cases, decision making in fundamental areas is constrained by the threat of military intervention or by a lack of control over the military.
In Latin America, examples abound of freely elected governments constrained by a military "guardianship." In Argentina from 1955 to 1966, certain electoral outcomes were ruled out a priori because the military proscribed the party that enjoyed most popular support. Guatemala's military played a de facto guardian role in the 1980s and early 1990s. About two years after winning in the largely free and fair 1985 elections, President Cerezo admitted that when he took office, the military permitted him to exercise only an estimated 30 percent of his constitutional powers. He claimed that the situation improved thereafter, while another local observer estimated that the percentage of power he could exercise actually decreased to 10 percent or 15 percent by 1988. 6 A similar situation prevailed in El Salvador from 1982 until shortly before the 1994 elections. The military and the paramilitary were beyond the control of the civilian government and ruthlessly killed tens of thousands of leftists and purported leftist sympathizers. Electoral outcomes unacceptable to the military were ruled out.
Governments in these countries were chosen in elections that were reasonably though not completely free and fair. But the military and paramilitary controlled a wide range of policy choices, including the range of permissible political opinion (the military violently repressed the left), human rights policy, the means employed in fighting the civil war, important aspects of labor policy (labor unions were brutally repressed), agrarian policy, and many other policies. The governments chosen by the people did not effectively govern in important policy areas. In many policy areas, the ruler was the military and/or paramilitary, and none of the options on the ballot provided an alternative to it. To call such a government "democratic" does not do justice to the word. If an elected government cannot govern because the military or some other actor dominates the political system, it is not democratic. The use of a subminimal definition of democracy leads Przeworski et al. to count many regimes as democratic despite practices that would lead most observers to a contrary judgment.

Coding and Aggregation Rules
In this section we present our coding and aggregation rules for classifying political regimes. Explicit and sound coding and aggregation rules form an important building block of regime classifications (Munck and Verkuilen forthcoming). Such rules promote evenness of assessments across cases and times. They reduce the subjectivity and measurement errors that affect regime classification (Bollen 1993;Bollen and Paxton 2000) unless one relies exclusively on strictly objective criteria, as Vanhanen (1990) does, but with considerable loss of conceptual soundness. They also make it easier for other scholars to assess criteria and actual classifications. 7 Our coding rules assess to what extent the four defining criteria for democracy are violated. They typify possible violations of these democratic principles and rank them as major failures or partial ones. The coding scheme follows:

Elections for Legislature and Executive
Has the head of government (in Latin America, the president) been freely and fairly elected or is he/she a constitutionally designated replacement for a head of government w h o died, resigned, or was impeached? By "elected" we mean, as did Przeworski et al. (2000), that he/she was chosen in fair direct elections or elected by a body that was itself mostly chosen in direct elections. In a similar vein, is there a legislative body in which the vast majority of the members have been fairly elected? With very few ex-ceptions, in the post-1945 period elections for the national congress have been direct, though on rare occasions, as in Argentina's upper chamber until the 1994 constitutional reform, some members were elected indirectly in free and fair elections.
In a democracy, the head of government and the legislature are chosen in free and fair elections.
A major violation of this democratic principle occurs if: a) the head of government or the legislature is not elected. b) the government uses its resources (patronage, repression, or a combination of both) to ensure electoral victory--i.e., there are systematic complaints about fraud or repression, and there is virtual certainty about the outcome of presidential elections (e.g., Mexico 1945-88, Argentina 1952-55, E1 Salvador 1952-63, Paraguay 1960. c) through fraud, manipulation, or outright repression, the government makes it impossible for a wide gamut of parties to compete (or if they do compete, to take office).
A partial violation occurs if: there are systematic complaints of rigged elections and/or harassment of the opposition but there is still uncertainty about electoral outcomes and the government fails to capture large majorities in the legislature; or the military vetoed a few "unacceptable" but important presidential candidates (e.g., Argentina 1958-66); 8 fraud affected but did not thoroughly skew electoral results; or the elections were conducted under substantially unequal playing rules (e.g., Nicaragua in 1984 because the Sandinistas dominated the media and pressured opposition groups, E1 Salvador in the 1980s because the left faced massive repression).

Is the Franchise Inclusive?
In a democracy, the franchise is broad compared to other countries in the same historical period, and disenfranchised social categories (e.g., children) are not seen as politically excluded groups with distinctive electoral preferences.
A major violation of this democratic principle occurs if a large part of the adult population is disenfranchised on ethnic, class, gender, or educational grounds in ways that: a) likely prevent very different electoral outcomes (or so is widely believed); or b) are unusually exclusionary for that historical period; or c) trigger mass social protests.
A partial violation occurs if disenfranchisement of some social groups occurs in ways that are not likely to significantly shape electoral outcomes. These criteria for inclusion involve complex judgments, but the obvious and simple criterion--universal adult suffrage--is misleading and unrealistic if we use a retrospective standard for classifying regimes. Even today, few if any countries observe universal adult suffrage. Some countries that are widely seen as democracies exclude the insane, convicts, permanent residents, non-resident citizens, or members of the armed forces. In addition, we overlooked the disenfranchisement of women and the illiterate for the early part of the time period under consideration. These earlier exclusions were cultural artifacts of a time past; this criterion of democracy has changed over time.

Are Civil Liberties Respected?
In a democracy, violations of human rights are uncommon, parties are free to organize, and the government respects constitutional guarantees.
A major violation of democratic principles occurs if: a) gross human rights violations or censorship against opposition media occur systematically; or b) political parties are not free to organize--i.e., most major parties are banned, just a single party is allowed to exist, or a few parties are tightly controlled by the government (e.g., Panama 1968, Paraguay 1947-59, Brazil 1965. A partial violation occurs if: a) violations of human rights are less widespread but still affect the opposition's capacity to organize in some geographic areas or some social sectors; or b) there is intermittent censorship of the media or regular prohibition of one major party or candidate.

Do the Elected Rulers Enjoy Real Governing Capacity?
In a democracy, military leaders and the military as an institution have negligible or minor influence in policies other than military policy, and their preferences do not substantively affect the chances of presidential candidates.
A major violation of this democratic principle occurs if: a) military leaders or the military as an institution openly dominate major policy areas not strictly related to the armed forces; or b) if the elected head of government is a puppet, such that the electoral process does not really determine who governs.

A partial violation occurs if military leaders or the military as an institution
are able to veto important policies in a few areas not related to the armed forces (e.g., Ecuador 1961-62). Munck and Verkuilen (forthcoming) have pointed out the importance of explicit and sensible rules for moving from disaggregated dimensions of regimes to an aggregated regime classification. Our aggregation rule is simple. We used these four dimensions to construct a trichotomous measure of democracy in Latin America. We classify governments as democratic, semi-democratic, or authoritarian. When governments commit no violation of any of the four criteria, they are coded as democratic. They rank as authoritarian if they present one or more major violations or as semi-democratic if they display only partial failures in one or more categories.
The criteria in our definition of democracy involve some discretionary coding--i.e., subjective assessments. But we prefer a complete definition even if it requires such assessments to one that is subminimal. It is small comfort that a classification is based on objective data if it does not include enough criteria to distinguish between democracies and nondemocracies. Social scientists should not ignore major components of democracy simply because they are hard to measure. Informed judgment oriented by well-specified coding rules is better than no measurement at all.
We recognize the value of continuous scales of democracy. We could have constructed a more continuous scale through an additive aggregation principle by adding up points (from zero to two for each dimension) on the four dimensions-thus producing a nine point scale from zero to eight. At this stage of our endeavor, we have conceptual and practical reasons for preferring a trichotomy to a more differentiated scale constructed in this fashion. Conceptually, adding points along the four dimensions could produce distortions because it assumes that the four dimensions (1) can be measured at the interval level; (2) can be measured in the same units; and (3) have the same conceptual weight, such that a strong score on some dimensions can offset a weak score on others.
Our coding and aggregation rules and hence our regime classification do not rest on such assumptions. Our four dimensions are coded along ordinal scales (major violation, partial violation, no violation), and mathematical aggregation might be misleading because the distances between ordinal categories are not necessarily uniform. The assumption of equal conceptual weight that inheres in an additive aggregation process is equally problematic because all four dimensions are individually necessary and jointly sufficient for democracy. Egregious violations along one dimension cannot be compensated by adherence to democratic principles on the others. For example, if elections are fraudulent and the government engages in widespread intimidation that makes it impossible for opposition parties to compete, broadening the franchise will not increase the democratic character of the regime.

A Classification of Regimes in Latin America
In this section and the next one, we discuss in greater detail our choice of the measurement level. Why have we chosen a categorical scale instead of a continuous index? And why is this scale trichotomous rather than simply di chotomous?
In addition to the problems of aggregation discussed in the previous section, the reason for choosing a categorical measure is twofold. First, notwithstanding the virtues of continuous measures for some research purposes (Bollen 1980;Bollen and Jackman 1989: 612;Coppedge and Reinecke 1990;Dahi 1971;Diamond 1999;Elkins 2000;Hadenius 1992: 36-71;Vanhanen 1990), it remains useful for other purposes to label political regimes. Both continuous and ordinal measures of regimes serve important research g~als (Collier and Adcock 1999). Regime labels are essential for analyzing comparative historical processes, for describing regimes, and for studying regime breakdowns and transitions.
Second, given our cost and time constraints, it would have been difficult to construct a more fine-grained measure for each country and each year since 1945. We may not know whether a country should be scored as a 6 or a 7 on Freedom House's interval scale, but we can be confident it is a semi-democ racy. By constructing a trichotomous scale with a modest information demand, we can significantly reduce the number of coding errors and thus achieve greater reliability than would be possible under a more demanding classification scheme. Our scheme has enough categories to avoid forcing cases into classes that violate our common sense understanding, yet has few enough that we do not need to draw fine distinctions among regimes.
Because this coding is qualitative and historical, it involves some degree of subjective judgment. But with explicit coding rules, the four dimensions that are being judged are uniform and the general parameters for making those judgments clear. The fact that the secondary data generated by country experts are available to many, in combination with explicit and relatively simple coding standards, makes it easier for other scholars to check our coding of cases.
Of course, even with explicit coding rules, some cases present difficult borderline judgments. Should Brazil 1946-63 be coded as a semi-democracy be cause of the restrictions on participation and competition, or as a democracy because competition at the national level was vigorous and participation fairly broad? A reasonable case can be made either way.
In sum, our ordinal trichotomous classification summarizes a lot of information, is descriptively and conceptually richer than quantitative codings, and permits us to map actual regimes onto the continuous scale of political practices in a way that matches an intuitive understanding of the nature of regimes and regime change. Our categories are readily comprehensible in ordinary social science parlance. This trichotomy allows for a meaningful range of variance without losing parsimony in the construction of regime types. Finally, our ordinal, trichotomous approach is consistent with the continuous nature of democratic practice. It imposes theoretically driven cut-points on a more continuous range of practices) Przeworski et al. (2000: 57-59) and Sartori (1987: 182-185; have argued that democracy should be conceived as a dichotomous phenomenon: a government is either democratic or it is not. We opt for a trichotomous measure and hence come close to Przeworski et al. and Sartori, but we nevertheless have reservations about a dichotomy.

Dichotomous Versus Trichotomous Measures of Democracy
Przeworski et al. code as democracies countries in which the president and the legislature are elected, more than one party exists, and alternation in power proves (in retrospect) to be possible. There is no middle ground; all other countries are dictatorships. Less authoritarian may be a good thing but it is not, in this conception, more democratic. Sartori (1987) also argues that regimes must first be classified as democracies before it makes sense to explore the degree to which they are democratic. Sartori and Przeworski et al. acknowledge that there are gradations among democracies: some are more democratic than others. Their argument is not that there are no degrees of democracy, but that there is a "natural zero point" for democracies. Regimes above a certain point are democratic and can be so to differing degrees, while all others are not and are qualitatively different.
Part of this argument is compelling: it is sometimes useful to go beyond a continuous measure of regimes and assign them a qualitative label. Many regimes are unabashedly non-democratic, and it makes sense to label them as such. Moreover, regime labels are useful for analyzing abrupt changes in regimes. Interval measures can capture such events, but it is useful to establish some cut-off points that indicate that a regime change has taken place. Also, a regime that is clearly democratic should be labeled as such. Przeworski et al. and Sartori thus keep intact a necessary distinction between democracy and nondemocracy; we should not abandon all efforts to categorize regimes and simply give them a quantitative score.
But Przeworski et al. and Sartori are excessively parsimonious in dichotomizing between nondemocracy and democracy. Even if one recognizes the value of labeling regimes in a qualitative manner, in contemporary Latin America a simple dichotomy does not suffice. A trichotomous classification is more useful: democracy, semi-democracy, and nondemocracy or authoritarianism. Przeworski et al. (2000: 57-58) explicitly reject the idea that there are borderline regimes between democracy and dictatorship, but this argument ignores the realities of many regimes in the contemporary world, especially outside the Western industrialized countries. Many competitively elected re-gimes fall well short of democracy, yet to call them authoritarian is also misleading. This problem is acute with respect to many post-1978 regimes in Latin America. As Diamond (1996Diamond ( , 1999, Hartlyn (1998), Karl (1995), O'Donnell (1994O'Donnell ( , 2001, Valenzuela (1992), and others have argued, many Latin American regimes have satisfied the requirements of fair competitive elections but on other important dimensions have fallen short of being democratic. A dichotomy is too parsimonious; it loses too much information about regimes.
Many regimes do not clearly approximate the ideal types of either democracy or authoritarianism and thus do not fit well in either camp. Attributes of both authoritarianism and democracy coexist in many regimes that fall between the two poles. This intermixing of attributes places these regimes in a "gray area," an intermediate category in the space of properties between democracy and authoritarianism.
A dichotomy requires very sharp distinctions among regimes when the reality may not justify them. Dichotomous classifications force the large number of intermediate cases into one of two categories, both of which may be misfits. A dichotomous classification is neither conceptually necessary or, for most research purposes, an optimal choice.
We prefer a trichotomous coding because of these problems with dichotomies. The presence of authoritarian elements in competitively elected regimes is well established (Diamond 1996;Karl 1995;O'Donnell 1993O'Donnell , 1994O'Donnell , 2001Valenzuela 1992); the existence of electoral institutions does not preclude the presence of authoritarian restraints on the use of those institutions. These regimes are what Collier and Levitsky (1997) call "diminished subtypes" of democracy. In Latin America, many competitively elected governments have failed to respect the civil and political liberties that enable free and fair elections to take place. Moreover, the holding of free and fair elections is no guarantee of the other three defining criteria of democracy; neither does the absence or malfunction of these institutions make all cases equally authoritarian. Elements of authoritarianism are present to varying degrees in many regimes based on reasonably free and fair elections, without completely destroying their effectiveness in expressing popular choice. This fact justifies an intermediate category of "semi-democracies." The concept of semi-democracy allows us to identify the many regimes in which imperfections in democratic practice impair but do not completely destroy the effectiveness of electoral institutions. By incorporating the category of semi-democracies, our scale gains in discrimination (presumably reducing measurement error), but still allows us to think of regimes in conceptually rich, categorical terms.

Comparing Measures of Democracy in Latin America
In this section we compare the three existing measures of democracy that provide annual democracy scores over a wide time, with ours, and assess their validity and reliability for Latin America: Przeworski et al. (2000: 13-77), Freedom House (Gastil 1991), and Polity (Gurr, Jaggers, and Moore 1990;Gurr 1995, 1996). We focus mainly on these three classifications because coding every country for every year enables scholars to track when regime transitions occur and hence ultimately to explain such transitions. Yearly coding also facilitates tracing region-wide trends; we know when waves of democracy and reverse waves began. Finally, this type of coding allows us to examine the causes and consequences of different kinds of regimes, as Przeworski et al. (2000) illustrated in their path-breaking study.
As Table 2 shows, the four measures of democracy are highly correlated. 1~ The strong correlations, however, do not mean that the choice of one measure over the others has no substantive implications. In fact, the different measures produce different substantive conclusions about key issues--for example, about how dramatic a change the post-1978 wave of democratization represents relative to the past, and whether the 1990s marked a slight democratic erosion, as Diamond (1996Diamond ( , 1999 asserts on the basis of Freedom House scores. A strong correlation would exist even if one measure systematically overrated the degree of democratization in every country. And many of the differences among these four measures are not random but are the result of the choice of a subminimal definition , a change in coding standards (Freedom House), or a political bias that worked against leftist governments (Freedom House). Freedom House. Beginning in 1972, every year Freedom House has ranked all independent countries from 1 (the best score) to 7 on both civil liberties and political rights (Gastil 1991).11 Freedom House evaluations have been used as measures of political regimes by combining the two scores to provide an assessment of how democratic a regime is (Diamond 1996(Diamond , 1999. Its scores provide a reasonably differentiated measure of democracy and offer comprehensive scope over nearly three decades. They implicitly incorporate at least three of our four dimensions of democracy: free and fair competition, broad participation, and civil liberties and human rights.
On the other hand, Freedom House's evaluations have two shortcomings for measuring democracy. First, they lack explicit coding rules. This makes it impossible to know what criteria are used in assessing regimes, leading to potentially serious problems of reliability and validity.
Second, its measurements contain two systematic biases: scores for leftist governments were tainted by political considerations, and changes in scores are sometimes driven by changes in their criteria rather than changes in real conditions. The first of these shortcomings is manifest in the harsh treatment of Nicaragua under Sandinista rule ) as compared to El Salvador for the same period. Freedom House scores suggest a more democratic government in E1 Salvador (a combined inverted score of 6) than in Nicaragua (a combined inverted score of 4) in 1984. Yet in 1984, the military in El Salvador was carrying out widespread political and labor repression and was violently suppressing the leftist opposition. In Nicaragua, three parties to the right and three to the left of the ruling Sandinista regime participated in elections in the same year, and these elections were certified by most European observers as fair and free of outright fraud and manipulation. Political violence outside the area of Sandinista-Contra conflict was limited. Despite occasional harassment of political opponents, the Sandinista regime did not murder, imprison, or torture large numbers of opposition leaders. Most observers agree that these elections and their surrounding circumstances were more democratic than those in E1 Salvador. This misclassification is not an isolated incident. Bollen and Paxton (2000: 77) show that Freedom House has a systematic bias against leftist governments.
Second, many scores of the 1970s and early 1980s are too lenient compared to scores in the 1990s. For example, Mexico's scores ranged from 6 to 8 throughout the authoritarian 1970s and 1980s. Yet during this time, political competition was very restricted. The PRI won every single gubernatorial and senate seat from the 1930s until the late 1980s; there was absolutely no chance of an alternation in power at the national or even the state level; and the opposition was harassed. Colombia also received a 10 in the early seventies when competition was still quite restricted . The National Front agreement of 1958 established that regardless of election results, congressional seats would be equally divided between Liberals and Conservatives, and the traditional parties colluded to alternate in power with every presidential election. The Dominican Republic (1972-73) andE1 Salvador (1972-75) were coded 9 during semi-democratic and authoritarian periods, respectively. And the aggregate scores for Guatemala were 9 in 1972 and 10 in 1973 during an authoritarian regime.
Freedom House scoring became more stringent in the 1990s and does not reflect the improvements that took place. For example, Mexico's political system was more democratic after 1988 than it had been previously. The 1988 presidential election, though vitiated by fraud, was easily the most competitive Mexico had experienced since the foundation of the PRI/state regime in 1929. By 1990, the opposition had become a serious political contender in many states. Yet Freedom House's 1980 (inverted) combined score (7) is slightly more democratic than the 1990 score (6). Political rights improved substantially in Brazil between 1984, when the military was still in power, and the early 1990s, but Freedom House scores indicate the opposite. In 1984, the last of the military presidents was still in office; citizens in state capitals and scores of other cities were not able to elect their own mayor; one-third of the federal senate had been elected indirectly in rules designed to guarantee majorities for the military government; communist parties were outlawed; and the left still faced sporadic repression. By 1990, these vestiges of authoritarian rule had been eliminated.
In E1 Salvador, the human rights situation improved substantially between the grizzly mid-1980s and the mid-1990s, but Freedom House scores reflect no change. A large UN-sponsored mission monitored a peace process and guaranteed human rights, the military scaled back its repressive activity, and the paramilitaries were brought more or less under control. The left began to speak out without violent reprisals, and new political parties started to come out into the opel-... By 1994, the formerly insurrectional FMLN, the object of brutal repression throughout the 1980s, felt secure enough to participate in the electoral process, and it won a substantial portion of the vote. None of this would have been possible ten years earlier, and a scoring of democratic practice should reflect this improvement. These biases in Freedom House scores are systematic rather than random. If the flaws were simply the result of random disagreements on particular cases, the differences would have less substantive impact. Freedom House might fare better in some of its judgments than we do; this would offset cases where our judgments are better. Random errors might create some noise in the analysis but would not necessarily skew the conclusions. But a systematic bias in measurement can lead to mistaken conclusions that are immune from correction through statistical means. Consequently, one must exercise caution in using Freedom House scores, especially to compare over time. Some conclusions based on Freedom House scores might be misleading because of its systematic biases, and the reliability and validity of its scores are subject to question because of the lack of explicit coding rules.
There are some significant differences between our measure and the Freedom House scores. For the 1972-96 period, of 475 cases (19 countries times 25 years), 89 Freedom House scores (18.7%) diverged from our assessments.~2 Most of the divergences (56) resulted from cases we coded as authoritarian but that had inverted Freedom House scores of 6 or more (e.g., Brazil 1979-84, Dominican Republic 1974-77, E1 Salvador 1972-77, Guatemala 1972-76, Honduras 1980-81, Mexico 1973-75, 1979. Polity IlL A second continuous measure of democracy is provided by the Polity III dataset (Gurr, Jaggers, and Moore 1990;Jaggers and Gurr 1995), which covers 177 countries for the period 1800-1994. The Polity project covers a very long period, and for Latin America, Polity III appears to have fewer systematic problems than Freedom House. Polity III provides explicit coding and aggregation rules, although they are not as clear as those provided by Przeworski et al. (2000).
Following a procedural conception, Jaggers and Gurr argue that democracy has three defining features: (1) "the presence of institutions and procedures through which citizens can express effective preferences about alternative political policies and leaders. This is accomplished through the establishment of regular and meaningful competition among individuals and organized groups, an inclusive degree of political participation in the selection of leaders and policies, and a level of political liberties sufficient to ensure the integrity of democratic participation, procedures and institutions... (2) the existence of institutionalized constraints on the exercise of executive power... (3) ... the guarantee of civil liberties to all citizens" (1995:471). Their first dimension in fact consists of three discrete issues (elections to determine who governs, in-clusiveness, and the protection of political liberties); the third of these (political liberties) overlaps substantially with their third dimension (civil liberties).
Gurr and his collaborators created eight ordinal scales with 35 categories in order to typify patterns of participation, the constraints on executive power, the recruitment of the chief executive, and the complexity of power structures in different societies (Gurr et al. 1990;Jaggers and Gurr 1996). Noting that some categories reflected traits of a democratic polity while others reflected autocracy, they selected 21 categories that correspond to five dimensions, weighed them, and integrated them in two scales (institutionalized democracy and institutionalized autocracy) ranging from 0 to 10.
The democracy scale assumes a "zero point" (it is a ratio scale) and is continuous. It reflects the degree of competitiveness in political participation and in the selection of the chief executive, the openness of the executive recruitment process, and the political and constitutional constraints on the executive. The autocracy index reflects the extent to which participation is suppressed or regulated, the degree of competitiveness or restrictions in the executive recruitment process, and the absence of checks and balances to executive powers. Following Jaggers and Gurr (1996), we subtracted the autocracy score from the democracy score, building an interval scale of democracy that ranges between -10 and 10.
Despite its merits, the Polity scale has some disadvantages. First, the relationship between their definition of democracy and their operationalization is muddled. Jaggers and Gurr (1995: 471) initially discuss the three components or dimensions of democracy noted earlier, but when they construct their indicators (p. 472), they have five broad categories that do not correspond to the three dimensions: competitiveness of political participation, regulation of political participation, competitiveness of executive recruitment, openness of executive recruitment, and constraints on the chief executive. They do not present a detailed justification for selecting these five categories, which omit some of the dimensions included in their discussion of democracy. In particular, their operationalization omits the protection of civil and political liberties and the inclusiveness of political participation, both of which are fundamental to their own definition and most others (ours included) of democracy.
Second, compared with several measures of democracy, especially to Przeworski et al. (2000), their coding and operationalization rules are cumbersome and are not sufficiently clear. Some of their dimensions operate at a high level of abstraction. It is hard to infer violations of human rights or the presence of reserved domains from broader concepts such as "regulation of participation" or "constraints on the chief executive." Third, some categories shaping the Polity coding are of little relevance for studying modern politics. For example, one dimension reflects whether the chief executive is "determined by hereditary succession" (Gurr et al. 1990: 81-82). Fourth, although the initial coding is based on rich, ordinal categories, the final measure is a continuous index for which values have no substantive meaning.
Finally, the authors do not provide a rationale for their aggregation rules, which weight some of their dimensions inordinately while neglecting others completely. Gleditsch and Ward (1997) noted that just two dimensions (con-straints on the chief executive and, to a lesser extent, competitiveness of executive recruitment) account for most variance in the democracy and the autocracy scales. The dimensions related to participation have little leverage on the Polity measure: "The extent and character of popular participation in selection of leaders is either totally absent or relatively unimportant in determining the degree of democracy" (Gleditsch and Ward 1997: 376). This finding holds for the subsample of Latin American countries. 13 This creates a problem of validity since democracy includes some elements (protection of human rights and civil liberties and the breadth of enfranchisement) that Polity does not measure.
As is the case with Freedom House scores, disagreements between our trichotomous classification and the Polity measure emerge mainly from cases that Polity codes as democratic but we believe should be coded as semi-democratic or authoritarian. Of the 344 Polity III cases ranging between -10 and -4, 99 percent (340) were classified as authoritarian regimes in our scale. In contrast to this near unanimity about the authoritarian cases, there is substantially more disagreement about the democracies. Polity scores between 5 and 10 (336 cases) were mostly coded as democracies (76%) or semi-democracies (23%). Przeworski et al. Previously we discussed two problems of the Przeworski et al. (2000) classification: the use of a subminimal definition of democracy and the decision to dichotomize democracy. Those choices lead to many coding differences with our classification. Spearman's rho between Przeworski et al.'s (2000) and our categorical measures is .83, but this association is stronger for the set of democracies than for the authoritarian regimes. All but one of the cases coded as democratic in our classification (228/229) are also democracies in the Przeworski et al. dataset. But only 90 percent (386/428) of our authoritarian cases are coded as dictatorships by Przeworski et al. This divergence underscores again that their operational definition of democracy is more lenient than ours. They classify 79 percent (96/122) of our semi-democracies as democratic regimes. They state that when in doubt they prefer to err on the side of counting a regime as nondemocratic, but--demonstrating that their definition is subminimal--they include scores of cases as democratic that we regard as authoritarian or semi-democratic.
For example, they consider Brazil during the waning years of military rule (1979-84) a democracy even though the head of government was chosen by the military and ratified by an electoral college designed to ensure subservience to the military's choice, governors were not democratically elected until 1982, and the leftist opposition and rural social movements were still subjected to frequent repression. They label Guatemala after the 1954 military intervention and from 1966 to 1981 a democracy, even though gross violations of civil and political liberties and the proscription of the left made these elections unfree at best, if not a total sham. The army and paramilitary carried out widespread killing of possible leftists, labor leaders, and Indians suspected of harboring or sympathizing with leftists. Most elections were attended by waves of state-sponsored terrorism and tainted by massive fraud. With the sole exception of civilian Julio Crsar Mrndez Montenegro (1966-t970), who ruled in   as democratic though by then Per6n had restricted the dissemination of opposing viewpoints and was silencing dissent and persecuting the opposition. None of these cases should be called democracies, but the presence of elections and the prior or eventual alternation in office lead Przeworski et al. to misclassify them as such. Figure 1 graphs the annual mean scores for the region for the four measures of democracy under discussion. For comparability, we normalized the values so all the scores run from 0 to 1, and inverted where necessary so that higher scores denote greater democracy. All four measures show a dramatic process of democratization in the region between 1978 and 1989. Despite this convergence, the four measures produce different perceptions about the process of democratization in Latin America, with Freedom House being the outlier. Our measure registers a lower level of democracy than the others for the pre-1978 period, and it suggests a sharper contrast between the more democratic 1990s and the authoritarian past than the other measures. Both our measure and Freedom House scores show the region's worst years to be 1976-77. From then until 1990, both measures show a marked improvement in levels of democracy. But Freedom House begins with a higher estimation of the level of democracy in the region than the other measures and ends with a lower one, so the slope of the line is flatter than it should be. As a result of tightened coding standards over time, the Freedom House mean line gradually approaches our evaluation (and the other two), crossing our line in 1989 and ending below all the other estimates by the mid-nineties. In short, Freedom House scores suggest a less dramatic improvement in democracy than the others.
According to Freedom House, levels of democracy peaked in 1990. Its data provide the empirical basis for Diamond's (1996Diamond's ( , 1999 argument that democracy suffered a decline in Latin America in the 1990s. In contrast, our measure shows levels of democracy as improving slightly in the 1990s.
By Freedom House's measure, region-wide levels of democracy were slightly worse in 1991-96 than in 1985-89. But a look at the region indicates otherwise. In the second half of the 1980s, Central America was still extricating itself from civil wars. Guatemala in 1985 was not a democracy by any measure, though things improved in 1986. El Salvador was still bogged down in a horrific civil war with massive human rights violations, and Panama was ruled by Noriega. South America also showed pockets of authoritarianism in the 1980s that were gone after 1990. Chile was governed by Pinochet, Brazil was under military rule until 1985, and Paraguay had a dictatorship until 1989. Finally, Mexico was more firmly in the grip of one-party rule in the 1980s than in the 1990s. The only countries where the outlook for democracy was worse in the 1990s were Peru after Fujimori's 1992 autogolpe; Colombia, where paramilitary and guerrilla violence increasingly constrained democratic practice beginning in the 1980s; and Venezuela. Freedom House scores from the 1980s and 1990s lead to a doubtful conclusion.
In a similar vein, note the brief period of democratization in the late 1950s and early 1960s. Przeworski et al.'s dichotomy and our trichotomy strongly register this trend, but the Polity index barely records it. An analysis that depended on Polity's indicator would miss this earlier "mini-wave" of democratization. In sum, despite the high correlations among the four measures of democracy, the choices of regime classification have important implications for the substantive understanding of politics.
Other Continuous Measures. The Polity and Freedom House indicators are not the only interval measures of democracy available, but they have advantages in terms of historical coverage. Some thoughtful continuous measures of democracy exist, but they are available for only a few years. Bollen (1980), Bollen and Paxton (2000), Coppedge and Reineke (1990) and Hadenius (1992: 36-71) constructed valuable multidimensional indicators of democracy. But their indicators require substantial qualitative information that is costly to collect on an annual basis for a long timespan and large number of countries. Not coincidentally, they restricted their measure to a single year (1985 for Coppedge and Reineke, 1988 for Hadenius) or three time-points for Bollen (1960Bollen ( , 1965Bollen ( , and 1980. In theory, these interval measures could be extended to cover a longer time period, but the cost of doing so would be enormous, especially for earlier decades. Some continuous measures that use less burdensome information present insurmountable problems of validity. Vanhanen (1990) measured democracy using two dimensions of elections: competition and participation. His measures involve objective quantitative information that could readily be extended for a long historical period, but they are conceptually flawed. He measured competition by subtracting the largest parties' share of the vote from 100 and participation by taking the percentage of the total population that voted. He then multiplied these two indicators to derive an index of democracy. But the measure of competition is flawed because it is too highly correlated with party system fragmentation. Contrary to what the index suggests, a system in which the largest party wins 50 percent is not necessarily less democratic than one in which the largest wins 35 percent, and a system in which the largest party wins 25 percent is not necessarily twice as democratic as one in which it wins 50 percent. The measure of participation--voter turnout--is also flawed. Voter turnout depends too much on the age structure of the society; it discriminates against countries with youthful populations in which a large share of the popu lation have not yet reached voting age. Higher rates of electoral participation may reflect compulsory voting laws rather than a more participatory environment. For democracy, the crucial point is that legal barriers and civil rights and political conditions be such that the adult population c a n participate; greater turnout does not necessarily imply more democracy. Most importantly, Vanhanen's measure fails to incorporate any assessment of civil liberties and political rights.

Conclusions
In this article, we present an alternative categorization of political regimes in Latin America since 1945. Our attempt to construct a new classification pushed us to rethink five broader issues regarding regime classification.
First, regime classification should rest on sound concepts and definitions. A classification of the extent to which regimes are democratic should be based on a procedural, minimalist but not subminimal definition of democracy. Our definition fails squarely within the contemporary debate, yet it is more stringent than many, leading to different perceptions about how democratic Latin America was before 1978. And the measurement of democracy should rest on the same dimensions as those included in the definition, contrary to what Jaggers and Gurr (1995) do. Among the three previous measures of democracy that we discussed at length, only Przeworski et al. hinge their measure on their definition.
Second, regime classification should be based on explicit and sensible coding and aggregation rules (Munck and Verkuilen forthcoming). Explicit rules enable other scholars to more easily evaluate classifications and also promote evenness of judgments. Przeworski et al. do a particularly good job of laying out their coding rules; conversely, Freedom House fails to indicate how it as sesses cases.
Third, although social science should when possible rely on objective and clearly measurable concepts, a fast distinction between "observables" and sub jective judgments is not useful for classifying political regimes. Regime classification must rest on empirically observable phenomena, but judgments about whether a violation of a particular dimension of democracy warrants classifying a certain regime as less than democratic are inevitably partly subjective. All four dimensions of democracy discussed in this article require an evalua-tion of observable phenomena, though the civil rights component, for example, is often harder to assess than the presence of competitive elections. Moreover, subjective judgments are unavoidable if we are to retain a conceptually valid definition of democracy. We rely on informed judgment and knowledge of the cases to make the coding decisions reliable while retaining essential aspects of the definition of democracy to make them valid.
Fourth, although continuous measures of democracy offer advantages, we agree with Przeworski et al. (2000) and Sartori (1987Sartori ( , 1991 that categorical classifications also serve useful purposes. The traditional discourse on political regimes is categorical. Our trichotomous measure efficiently captures conceptual distinctions that are important to comparative social scientists. In addition, continuous measures usually fail to convey the rich theoretical implications that more conceptually grounded categories do. They also demand a level of information that may not be available or may be very costly to develop. Fifth, we advocate a trichotomy rather than a dichotomy for classifying regimes. Dichotomous measures fail to capture intermediate regime types, obscuring variation that is essential for studying many political regimes in what Huntington (1991) called the "third wave of democratization." Our trichotomous ordinal scale acknowledges the trade-off between using meaningful regime labels and fine measurement. The idea of semi-democracy allows us to conceptualize historical regimes that do not fit neatly in a dichotomous classification, such as the first Per6n administration in Argentina (1946-51), the MNR government in Bolivia after the revolution (1952-64), the Frente Nacional period in Colombia (1958-74), and the Ar6valo and Arbenz administrations in Guatemala (1945-54). It also describes the many incomplete processes of democratization during the third wave--e.g., Mexico until the July 2000 presidential election, Nicaragua, and Paraguay--and cases of democratic erosion in the 1990s (e.g., Colombia).
Our trichotomous measure is based on a more stringent definition of democracy than Schumpeter's and Przeworski et al.'s, and yet is designed to minimize information costs and ensure reliability. It attempts to strike a balance between the inadequate differentiation of dichotomous measures and the huge information demands of continuous measures. It is based on enough knowledge of the nineteen countries to make reasoned judgments about the less easily observable dimensions of the regimes in question. Its combination of a thick conceptual grounding and a parsimonious coding demand is well suited for a medium-sized N study in which a research team can make informed judgments about cases.
We hope that our regime classification also contributes to comparative scholarship on democracy and on Latin America. Much of the comparative research into the conditions for, or the causes and consequences of democracy, or the differences among countries on regime characteristics ultimately rests on regime classifications. If, as we argue, the main existing classifications are flawed by political considerations, subminimal definitions, invalid measures, or other sources of systematic bias, conclusions about political regimes may be affected. 10. In this section, for the sake of simplicity in presenting results, we assume interval properties for our trichotomous measure (e.g., when we run Pearson correlations with other measures, estimate means, and so on). For ease of comparison with other measures, we combined the two Freedom House scores and inverted the measure (by subtracting it from 14) so that it runs from 0 (least democratic) to 12 (most democratic) 9 11. Freedom House scores are labeled with two consecutive years--1972-1973, 1973-1974, and so on. We use a single year as the label; we identified the Freedom House scores with the first year of the pair, which most closely reflects the year for which the conditions are reported. 12. A divergence occurred if a regime we coded as a democracy had a combined inverted Freedom House score of 7 or less; if a regime we coded as a semi-democracy had a combined Freedom House of more than 8 or less than 6; or if a regime we coded as authoritarian had a Freedom House score of 6 or more. The intentional overlap of the democratic and semi-democratic categories makes this a lenient test for divergence. 13. An OLS model predicting the Polity score (democracy minus autocracy) based on all five dimensions yielded greater standardized coefficients for the Executive Constraints dimension (.47) and the Executive Recruitment Competition dimension (.40) 9 All other components, Competitiveness of Participation (.29), Recruitment Openness (-.20), and Regulation of Participation (-. 11), showed lower beta coefficients 9 All t-tests significant at .00 l level, R-square = .97 (n = 918).