Party Systems, Governability, and the Quality of Democracy in Latin America

Michael Coppedge Kellogg Institute, Hesburgh Center University of Notre Dame Notre Dame, IN 46556 coppedge.1@nd.edu tel 219-631-7036 fax 219-631-6717

Paper prepared for presentation at the conference on Representation and Democratic Politics in Latin America, organized by the Department of Humanities of the Universidad de San Andrs and the Department of Political Science of the University of Pittsburgh, Buenos Aires, Argentina, June 7-8, 2001.

Party Systems, Governability, and the Quality of Democracy in Latin America

The purpose of this paper is to define and describe aspects of Latin American governability and the quality of democracy that are related to party systems. As democracy has become more common around the world, scholars have shifted their focus to understanding differences in kinds and degrees of democracy, which have come to be called issues of the quality of democracy. At the same time, other scholars and policymakers have focused attention on a different set of qualities that are called governance or governability. Quality of democracy and governability are both such complex, multifaceted concepts that it would be impossible to define and describe either one fully in a single paper. However, it is possible to treat adequately particular aspects of the two concepts that depend on the nature of party systemspartisan powers and the quality of representation. I will do this by defining both concepts, elaborating some theory to guide measurement of them, and presenting actual measurements for 123 elections in 11 Latin American countries. This analysis makes it possible to support three general observations about the impact of party systems on partisan powers and the quality of representation. First, these two concepts represent distinct dimensions: neither one can be reliably derived from the other; they vary rather independently. Second, there is, nevertheless, a rough inverse relation between them: a tradeoff between partisan powers and the quality of representation. High-quality representation tends to be associated with weak partisan powers, and low-quality representation tends to be associated with strong partisan powers. Finally, despite this tradeoff, there are ways to combine the best of both dimensionshigh-quality representation with strong partisan powers. Only about one quarter of the Latin American elections in the sample analyzed here achieved this happy state. The great majority were prone to either weak partisan powers, tending toward ungovernability; or a low quality of representation, tending toward a low quality of democracy. Party systems combining the worst of both dimensions are much rarer, but all were soon followed by an interruption in the democratic regime. Party Systems and the Quality of Democracy Party systems affect the quality of democracy at only one of several stages of the democratic process. That process could be said to begin with the formation of preferences by citizens and continue on through the representation of preferences, decision-making, policy implementation, and policy adjudication (which affects the formation of preferences, bringing the process full circle).1 At each stage of the process, different criteria for democratic quality apply.
1

This is not the only possible scheme for breaking up the quality of democracy into manageable components, but it is a useful one for purpose at hand, as the relevance of party systems is well confined to the stage of representation. One limitation of this scheme is that it assumes a procedural model of democracy. For broader views of other models of democracy, see Held (1996) and Katz (1997). For a defense of including the state's role in policy implementation and adjudication in a definition of democracy, see ODonnell (1993). 1

2 For preference formation, the relevant criteria include transparency and access to information, freedom of speech and organization, campaign finance arrangements, and so on. For decisionmaking, the relevant criteria are careful and reasoned deliberation, equal votes by representatives, inclusion of all representatives at all stages of decision-making, and so on. For the stage of representation of preferences, I would argue that quality hinges on institutions and procedures that faithfully translate the diverse preferences of voters into a representative microcosm of society called a legislature. Ideally, every preference and combination of preferences found in society should also be found in the legislature, in proportions that closely mirror the distribution of preferences in society. This democratic ideal is not feasible, or necessarily even desirable.2 Real-world legislatures do not represent their societies anywhere near this faithfully. However, this unrealizable ideal can serve as a useful standard for comparison: the more closely a party system adheres to it, the higher the quality of representation. Party systems affect the quality of representation by defining the number and quality of choices available to voters for the expression of their preferences. The more parties there are, the more likely it is that every voter or group of voters will be faithfully represented by one of them. But at the same time, not just any set of parties will do. They must be parties that are programmatically distinct, parties that take clearly different positions on issues that are relevant for giving the voters some control over what the government does. This is a requirement for any semblance of a mandate and accountability in democratic politics; without it, elections would be meaningless and irrelevant. Therefore, the clearer and more elaborate party programs are, and the more distinct they are from those of other parties in the system, the better the quality of representation. Some readers may object that I am advocating extremely fragmented and polarized party systems (contra Sartori 1976, 131-173). I am not. What I am doing is defining an analytical standard for the quality of representation by party systems, without (at this point) taking a position on the separate normative question of the desirability of some quality of representation.
2

This model of representation comes from the democratic tradition of popular sovereignty and is usually called the "delegate" model of representation. It is often contrasted with the "trustee" model of representation, in which representatives are not obliged to respect the wishes of their constituents; rather, representatives must use their best judgment to favor policies that are in their constituents' best interests, whether the constituents agree with that judgment or not. See Edmund Burke (1774) and Pitkin (1967). I view the delegate model as the more democratic one. The trustee model strikes me as an attempt to justify limitations on democracy in order to reconcile democracy with other values, such as political order, informational disparity, or even hegemony of a ruling class. I do not mean to argue that higher-quality representation in the delegate mold is always preferable; rather, I believe that it is merely truer to the core values of democracy, with the understanding that some extreme versions of democracy would not be desirable. Indeed, the conclusion of this paper is that some Latin American countries might be better off if the quality of representation were dampened a bit in order to achieve a more governable regime.

3 This standard is designed to make it easier to say what makes a party system more representative or less representative; it says nothing about whether a more or a less representative party system would be a desirable goal. Confusing democracy with "the good" in politics is a bad habit that leads us to believe that more democracy is always better. Although I am a fervent supporter of democracy, I believe that there is (in principle) such a thing as too much democracy. Our indicators should allow us to measure democracy even at the extremes so that such ideas can be tested. Measuring the quality of representation with the number of programmatically distinct parties is a good example of this principle. Fragmented and polarized party systems may be undesirable, but they are undesirable from the standpoint of other norms, such as political order or decisive policy-making, not because of how representative they are. In fact, measuring the quality of representation even at an undesirable extreme is the only way it will be possible to discover whether there really are tradeoffs between the quality of representation and other core values. Of course, there is more to the quality of representation than the number and distinctiveness of parties. A truly comprehensive examination of the quality of representation would have to consider the breadth of the suffrage, the frequency of elections, the ratio of voters to representatives, assembly size, opportunities for voting on issues directly, the possibility of a recall, the ease of registering parties, opportunities for participation in candidate selection, and many other conditions. However, the topic of this section is not the quality of representation overall, but the contributions of party systems to the quality of representation. Strictly speaking, the concept of interest would be called quality of representation by party systems, but for ease of expression I will call it simply the quality of representation. With that clarification in mind, we may proceed. The simplest way to proceed would be to create an indicator of the number of parties and an indicator of programmatic distinctiveness. Several ways to measure each already exist. For measuring number, one may simply count the number of parties running candidates or winning seats or winning a certain percentage of the vote, calculate the vote shares of the largest n parties, calculate Rae's index of fractionalization or the Laakso-Taagepera Effective Number of Parties index, or calculate various other indices (Rae 1970, Laakso and Taagepera 1979). Distinctiveness could be measured by the vote shares of non-personalist parties, some indicator of the reliability of judgments about the parties' left-right positions, the number of ideological blocs into which parties may be grouped, or various indicators of polarization. But the number of available indicators for number and distinctiveness actually complicates measurement of the quality of representation because one must decide which one(s) to use and how (or whether) they may be properly combined. It is desirable to use these indicators to create a single measure of the quality of representation, if possible (Coppedge forthcoming 2002). But creating a single measure is worth doing only if it can be done well. It would not be sound practice simply to select one best indicator as a proxy for the quality of representation because this would be reductionist: it would fail to reflect all the relevant aspects of the quality of representation. Neither would it be correct to add or average several indicators,

4 because addition and averaging assume that the components are unidimensional (i.e., strongly associated) and equally weighted. I do not believe that the number and distinctiveness of parties are sufficiently unidimensional to justify averaging. Certainly they are associated at the lower extreme. When there are few parties, of necessity few differences are represented. But the same is not true of the upper extreme. There can be many parties without meaningful differences, for example when the left bloc contains many different leftist parties, or when there are many nonideological, personalist parties. Two dimensions are therefore required to represent all the possible combinations. Fortunately, a single indicator can be constructed from multidimensional components such as these. However, success in doing so requires careful theorizing to define how to combine them properly. For example, in geometry, area is a two-dimensional concept that can be measured meaningfully. But measuring area requires using the correct theoretically-derived formula: side2 for squares, length*height for rectangles, 0.5*length*height for triangles, **radius2 for circles, etc. To derive my indicator of the quality of representation, I start with two variants of the Laakso-Taagepera index: ENPV for the effective number of parties based on votes for the lower or sole legislative chamber;3 and the effective number of blocs (ENB), which uses the same formula but substitutes vote shares of ideological blocs for vote shares of individual parties.4 (I chose an indicator of the number of blocs rather than an indicator of ideological distance or polarization because what matters for the quality of representation is that each ideological

The Laakso-Taagepera index is the inverse of the sum of squared party shares (Laakso and Taagepera 1979). The effective number of parties based on seats could be substituted for ENPV, but I have chosen ENPV because an indicator of the quality of representation should reflect the diversity of choices available to voters more than it reflects how those choices are translated into the distribution of representative offices. Also, my indicator of governability or partisan powers necessarily makes use of information about some seat shares, and I want to ensure that these two indicators are constructed from different variables, even if they are empirically associated. Bloc shares for this variable come from an updated version of my classification of Latin American parties. For this, I drafted a classification of all the parties that contests lower-chamber elections, identifying parties with a position on a left-right spectrum (left, center-left, center, center-right, or right) and a religious/secular dichotomy; or, if that was not possible, it classified parties as personalist, other (e.g. regional, ethnic, environmental), or simply unknown. I then sent this draft and my explicit coding criteria to 80 country specialists, asking for their advice in correcting any misclassifications. Fifty-three of the experts provided feedback, which I then used to make corrections. Only 3 percent of the vote in this sample of elections went to parties that fell into the unknown category. Complete documentation for this variable is in Coppedge 1997, and the percentages of the vote won by each bloc are available at <http://www.nd.edu/~mcoppedg/crd>.
4

5 position be represented, not how far apart those positions are or how skewed the electorate is toward the extremes.) Both variables have a minimum value of 1, so I subtract 1 from each to make the minimum value of the quality of representation index (QR) zero. The index has no maximum: in effect, the more parties, the better.5 The party and bloc numbers should be combined in such a way that when either one approaches zero, QR approaches zero, because the quality of representation is nil if there is either no party competition or nothing of substance to their competition. An additive index would not have this property, but a multiplicative index does (especially when each component has a lower limit of zero rather than one). A disadvantage of simply multiplying ENPV-1 by ENB-1 is that the product becomes hard to interpret. However, by taking the square root of the product, we obtain the geometric mean of ENPV-1 and ENB-1, which restores the same meaningful scale of the effective number of parties (minus one). QR can therefore be interpreted as the effective number of ideologically distinct opposition parties. The final formula for QR, however, incorporates three additional features. The formula is: (1)

QR =

( ENPVid 1)( ENBid 1) [(1 .01P) reliability ]

This formula treats personalist parties differently than the more ideological parties. ENPVid and ENBid above are simply ENPV and ENB recalculated using only non-personalist party or bloc shares of the non-personalist vote.6 The next term, 1-.01P, stands for the proportion of the vote going to non-personalist parties (P being the percentage of the vote going to personalist parties or a personalist bloc). This first modification therefore takes the geometric mean effective number of ideologically distinct parties and then discounts that number by the share of the vote won by personalist parties. In this way, QR attains its maximum values only when there is no personalism. It could conceivably be discounted to zero if all parties were personalist, but in this sample the lowest weight is .32, corresponding to a personalist vote of 68 percent in Peru in 1995.

As a practical matter, the maximum value is 10 because there are only 11 possible nonpersonalist blocs in my classification scheme (5 left-right X 2 Christian-secular + Other). With an infinitely divisible classification scheme, the theoretical maximum would be the assembly size because there cannot be more parties than there are seats in the legislature. But because no actual party system approaches that level of fragmentation, the number of parties may as well be treated as unlimited. ENPVid and ENBid could be calculated by first inflating the vote shares of all nonpersonalist parties or blocs by dividing them by 1-P/100. (P is the percentage of the vote won by personalist parties.) However, I used a shortcut that yields identical results: ENPVid = [1sum(P/100)]2/[(1/ENPV)-sum(P/100)2].
6

6 Equation 1.1 also weights the geometric mean by a reliability index for the classification of parties in each election. It matters whether a party is very clearly identified with a certain bloc or only uncertainly identified with that bloc. Voters are better represented when they are presented with clear options, and poorly represented when they cannot rely on their party of choice to remain faithful to the ideas that it seemed to advocate during the campaign. My indicator of reliability also comes mostly from my classification of parties. It is a weighted average, over all the parties contesting a given election, of the percentage of country experts agreeing with the final classification of each party.7 The third modification, not shown in equation 1.1, is a correction for the cases in which a party or parties boycotted or was banned from an election. Obviously, in such cases the party system is less representative of the voters, so QR is adjusted downward. The adjustment was accomplished by substituting for ENPVid an estimate of what ENPVid would have been if the absent party had participated, but making no adjustment to ENBid.8 The effect is usually to reduce QR, mainly due to the reduction in the effective number of blocs caused by the withdrawal of a party.9 With these additional features, QR can be interpreted as the effective number of reliably ideological, non-personalist opposition parties. It ranges from zero to a theoretical maximum of 10, although in this sample the maximum is 4.83. A significant benchmark is a value of one, which corresponds to meaningful competition between two parties. Party systems with a QR of

This indicator rests on the assumption that if a partys position is not clear to the experts, it was not clear to the voters. The average reliability rate for this sample is .86. For details, see Coppedge 1998, 553-56. Too few experts responded to the draft classification of parties in Bolivia and Uruguay to calculate reliability scores for these countries, so I have assumed reliability of .8 for Bolivia and .85 for Uruguay. Also, the original classification ended with elections in 1994 or 1995; my classifications of parties in later elections have not been reviewed by experts, so in these cases I have repeated the most recent reliability scores for the relevant country. Adjusted ENPVid = [(1+B)2]/[1/ENPVid +B2], where B is the share of the vote won by the boycotting or proscribed party in the last election in which it participated. The use of the estimated ENPVid is a compromise solution designed to avoid a measurement paradox. When a large party is proscribed, such as the Peronists in Argentina or the APRA in Peru, ENPVid tends to become larger because the remaining parties are smaller and more equal in size. This then tends to make QR larger as well, implying that banning a large party improves the quality of representation! This is clearly false. On the other hand, the withdrawal of a party of intermediate size can inflate the dominance of the largest party in the system, reducing the quality of representation by too much. The compromise solution used here holds ENPVid more constant, which transfers more of the measurement burden to changes in ENBid in such situations. This seems to work reasonably well.
9 8

7 less than one fall below a reasonable minimum standard for representation. Such low values correspond to dominant-party systems, systems of multiple but indistinct parties, a high degree of personalism, parties with unreliable positions, or systems from which a large party has been excluded. Where QR is greater than one, no large parties are banned and personalism is low, and there are at least two competitive parties with ideologically meaningful and reliable differences. The larger QR is, the greater is the variety of distinct positions represented by the parties. Party Systems and Governability Governability has been defined in a great variety of ways. For the World Bank and other international economic actors, it is closely related to the notion of transparency. But in a broader sense, governability could embrace every political phenomenon related to stability, order, and legitimacy: the rule of law, law abidingness, efficient bureaucracy, a strong merit system, low crime rates, constitutional succession, low strike rates, long-lasting cabinets, strong corporatist institutions, and many other aspects of institutionalization (Huntington 1968, Coppedge 1995). Focusing on the contribution of party systems to governability narrows the list of topics considerably, but it is still long and could include electoral volatility, party penetration of associations, and even levels of party identification. In this paper I somewhat arbitrarily narrow the focus even further to the impact of the party system on the ability of governments to make policy decisively. This is still a large topic, as evidenced by the list of symptoms of ungovernability in this sensestalemate, impeachment, vetoes, cabinet instability, and any other manifestation of executive-legislative conflict, which in extreme forms can escalate to constitutional crises and regime breakdown. Furthermore, this choice of focus is not completely arbitrary, because this is the most thoroughly studied aspect of the connection between party systems and governability (Lijphart 1984, Shugart and Carey 1992, Mainwaring 1993, Linz and Valenzuela 1994, Jones 1995, Mainwaring and Shugart 1997, Lijphart 1999). The concept treated in this paper is very close to a concept developed in this literature: partisan powers. The term refers to the partisan powers of the president, as opposed to the presidents constitutional powers, such as the veto, appointment powers, and dissolution powers. Mainwaring and Shugart (1997), for example, argued that presidents with weak constitutional powers can compensate with strong partisan powers (as in Mexico before 1997), and presidents with weak partisan powers can compensate with strong constitutional powers (as in Brazil after 1986). Although this literature defined the concept as an aspect of presidential power, it works equally well, if not better, as a characteristic of the party system: the capacity of the party system to ensure passage of the presidents legislative agenda. There is general agreement on two components of partisan powers: the size of the presidents party or coalition, on the one hand, and party discipline, on the other. These two characteristics combine to define what I call a reliable majoritythe percentage of congressional seats that the president can count on to vote in favor of his typical bills. This figure should also be in line with the presidents legislative success rate: the proportion his or her bills that win approval in congress (unless the president anticipates defeat and does not submit the more controversial ones).

8 Mainwaring and Shugart (1997, 421-434) measure partisan powers as (2) SIZE * (1 - (.11*NOS))

where SIZE is the percentage of congressional seats controlled by the presidents party and NOS is the number of Nos listed for a country in their table describing electoral and party rules that they believe to affect party discipline. The table permits up to three Nos, so party discipline is estimated to range from .67 to 1.00. My indicator of partisan powers shares this assumption of an interaction between party discipline and the size of the presidents party or coalition, but differs in four ways. First, because I am not persuaded that party discipline is powerfully associated with electoral or party rules, I base my estimates of party discipline on actual data from a few countries and extrapolations to others informed by conventional wisdom and anecdotal evidence. My basic source is a paper by Scott Morgenstern (1999) that provides Rice unity scores for major parties in Argentina, Brazil, Chile, and Uruguay.10 Fortunately, this small sample covers a wide range, from Chile, with highly disciplined parties, to Brazil, which has parties with very poor discipline (Mainwaring 1999). These Rice scores range from 100 percent (PRSD in Chile, 198991) down to 45.1 percent (National Party in Uruguay, 1990-94which was unusually low for Uruguay). My estimates of party discipline are 1 for Mexico, Venezuela, and authoritarian Brazil; .9 for Argentina, Peru, and Uruguay; .85 for Costa Rica; .816 for Chile (the average across parties); .8 for Bolivia; .75 for Colombia; and .6 (which is little better than chance) for Ecuador and democratic Brazil. Second, I take into account a baseline success rate. A great deal of legislation typically passes by consensus in most countries most of the time. However, this "floor" of success can be assumed to vary from country to country and year to year. I assume that the floor rises and falls along with the degree of polarization in the party system. Polarized party systems have a low floor, allowing few bills to pass easily; party systems with low polarization have high floors, enabling a high rate of legislative success in spite of institutional impediments. I therefore estimate the floor as (1 - IP/100)2. IP is my index of polarization, once again based on my classification of parties. IP measures the dispersion of the vote away from the center and toward the extremes; it ranges from 0 (no polarization) to 100 (complete polarization) (Coppedge 1998, 556-558). In this sample, the lowest floor was .026 (only 2.6 percent of legislation assumed to pass easily) in Chile in 1953, while the highest floor was .920 in Mexico in 1964. Third, I model the possibility that the president will lose the support of his or her party, for a large and disciplined party is of no help if it is at odds with the president. I make sparing use of this possibility, preferring to factor in routine defections as discipline problems whenever
10

The Rice index is |#yes - #no|/(#yes + #no), yes and no referring to legislators votes. This index is not ideal, as it ignores abstentions and absences, but it probably provides a good rank ordering, which in the narrow .6 to 1.0 range used here cannot be far off.

9 possible. For this reason, this variable (LOSS) is set to zero for most observations. However, it equals .1 for Bolivia, Costa Rica in 1958-1994, Peru in 1980, and Venezuela in 1968 and 1993; .15 for Venezuela in 1958; .2 for Brazil in 1945-62, Colombia, Ecuador, and Peru in 1985; .25 for Uruguay (due to fractionalism); and .4 for Venezuela in 1963 and 1988. Fourth, I combine all of these variables into a nonlinear function that models a sharp disjunction between controlling a little more than 50 percent of the seats in the lower chamber and controlling a little less than 50 percent of the seats. This makes good theoretical and intuitive sense, as the difference (for the chance of passing legislation) between 45 and 55 percent of the seats is much greater than the difference between 25 and 35 percent, or 65 and 75 percent. If this were not so, no one would care about party discipline. The tighter party discipline is, the more difference it makes whether the president's party is above or below that 50 percent threshold. Therefore, the partisan powers variable is modeled with an S-curve that is very steep around the crucial 50 percent threshold when discipline is tight, but nearly flat when discipline is loose. The formula and an intuitive derivation can be found in the appendix. This rather complex indicator agrees well with the Mainwaring and Shugart index at the extremes. In fact, the overall correlation between the two is .83.11 However, in the MainwaringShugart 25 to 50 range, where the bulk of the cases lie, there is little relation between their index and mine. I believe that an indicator that makes appropriate use of a greater variety of relevant information should measure partisan powers better, but only actual use of these indicators will tell whether one is superior to the other. Eleven Case Histories The remainder of this paper presents empirical data from 123 Latin American elections. One purpose of this quick survey of the cases is to allow readers to compare the quantitative record with their own historical knowledge so they can judge for themselves whether these two indicators actually measure well what they purport to measure. The other purpose is make a diagnosis of the challenges that party systems have posed for these countries over the last few decades. In some cases, party systems have been quite functional; in others, party systems have promoted governability at the expense of representation; in still others, party systems have permitted higher-quality representation at the expense of governability. I will present the data graphically. All of the country graphs are formatted identically to aid comparisons across countries. Figure 1 below serves as a key for interpreting these graphs. All of them plot partisan powers on a 0-1 horizontal axis and QR on a 0-6 vertical axis. The partisan powers axis is bisected at the .5 threshold so that it is easy to see that cases to the left of this vertical threshold have weak partisan powers and those to its right have strong partisan
11

Mainwaring and Shugart provide only averages for periods containing several elections in most cases, but they also provide sufficient information to reproduce their index for individual elections. The correlation reported in the text is based on individual elections rather than averages.

10 powers. Similarly, the vertical axis is bisected at QR=1. Cases below 1 are those with inadequate quality of representation, and those above 1 present voters with a choice of meaningfully distinct parties that is adequate or better. Together, these two thresholds divide the graphs into four quadrants. The upper right quadrant contains cases of high-quality representation and strong partisan powers: the best of both dimensions. The lower left quadrant contains cases of low-quality representation and weak partisan powers: the worst of both dimensions. On the lower right are cases of strong partisan powers but low-quality representation; in the upper left quadrant are cases of high-quality representation but weak partisan powers. Figures 2-12 plot each election as a point in this space. Each data point is labeled with the last two digits of the corresponding year. For example, a 69" would refer to 1969. Elections held in the same political regime are connected with straight lines; elections held in different regimes are not connected. A good way to read the figures is to start with the earliest year and then trace the evolution of QR and partisan powers from each election to the next, taking note of the quadrants in which the cases fall. Ecuador Ecuador (Figure 2) falls squarely in the upper-left high-weak quadrant. Its consistently weak partisan powers should be no surprise. They were a bit stronger when Rolds Concetracin de Fuerzas Populares (CFP) (1979) and Borjas Izquierda Democrtica (ID) (1988) won large pluralities, but never enough to push partisan powers over the .5 threshold. In fact, this figure overstates partisan powers, because midterm elections and party-switching have almost always shrunk the presidents party after the term begins. The rather high quality of representation may be more surprising because the unusually high level of personalism in Ecuador usually receives attention. But beyond the colorful politicians, Ecuador has one of the party systems that most equally represents the whole range of left-right blocs, with some presence of separate Christian blocs as well. The high degree of fragmentation, particularly bloc fragmentation, more than compensates for the personalism in the calculations of QR. Mexico Mexicos party system is hard to see in Figure 3 because it is pushed deep into the lowerright, low-strong corner. Before 1988, this was a party system that clearly promoted partisan powers at the expense of representation. However, there has been some noticeable movement toward higher-quality representation, moving into the optimal upper-right high-strong quadrant. (I expect much more dramatic movement toward the upper left quadrant for the 1997 and 2000 elections, with the Partido Revolucionario Institucional (PRI) first losing its majority in the Congress and then losing the presidency.) Bolivia Figure 4 portrays the Bolivian party system since 1979 as very similar to Ecuadors, with the major exception of 1993-1997. The most common tendency has been weak partisan powers combined with impressive representation of a great variety of blocsnow including some indigenous partiesdespite high and growing personalism. The 1993 election was a major exception. It increased partisan powers dramatically (at least initially), but at the expense of the

11 quality of representation. This was the result of a pact between the right-wing Accin Democrtica Nacionalista of Hugo Bnzer and the center-left Movimiento de la Izquierda Revolucionaria of Jaime Paz Zamora. This pact presented the voters with an extremely confusing alliance that actually won fewer votes than either party had won on its own in the previous election. Also, the new alliance was situated in the same center-right bloc occupied by the victorious Movimiento Nacionalista Revolucionario (MNR), dramatically reducing the number of both parties and blocs. However, the pact inadvertently boosted support for the MNRs Gonzalo Snchez de Lozada and lowered polarization enough to improve his chances of legislative success substantially. Brazil Brazil illustrates the full range of variation found in Latin America. From 1950-1964, Figure 5 reveals it to have had a party system little better than that of Ecuador or Bolivia: a good variety of blocs, but with weak partisan powers. The authoritarian party system of 1966-1978 falls in the low-strong quadrant like the Mexican case of electoral authoritarianism. However, an electoral reform at the end of the military regime brought briefly into being (1982) a configuration of parties that combined adequate representation with sufficient governability. A similar combination existed in 1945, at the conclusion of the Estado Nvo; these were the only two Brazilian cases in the high-strong quadrant. With full liberalization of electoral politics, the Brazilian party system quickly moved back into the high-weak quadrant and kept moving at least through the 1994 election: ever-more representative and ever-more ungovernable. Colombia Before 1990, Colombias party system provided poor or borderline representation due to the small number of parties (very close to two most of the time) and the scant and unclear ideological differences dividing them (Fig. 6). Nevertheless, it was a highly governable system (as far as parties were concerned) due to low polarization between Liberals and Conservatives and the relatively large governing parties, despite the low discipline and tenuous loyalty of some factions to their own presidents. The first major exception (a mirror image of Bolivia) was the constituent assembly election of 1990 which, because of low turnout, produced a much more fragmented and diverse legislaturebut at the expense of partisan powers. Since the constituent assembly, the Colombian party system has moved toward its old self, but not all the way, and in 1991 and 1994 at least it approximated the best of both dimensions. Obviously, these dimensions say nothing about ungovernability caused by the guerrilla war, the drug war, and the collapse of the state, but there is no harm in admitting that some outcomes are beyond the influence of parties and party systems. Argentina The Argentine party system has experienced two modes in the postwar period: one with Peronism and one without. As long as Peronism was banned (elections of 1958-1965 here), the quality of representation in Argentina was substandard (and partisan powers weak as well in 1965, just before the coup of 1966). But whenever the Justicialist party was allowed to contest elections (1946, 1973, and the whole period beginning in 1983), the quality of representation

12 became at least adequate (Fig. 7). All of these elections also produced a party system that afforded the president strong partisan powers. It is also interesting to note that partisan powers tended to weaken under Alfonsn (1983-1987) and strengthen under Menem (1987-1995). Costa Rica Costa Rica presents a somewhat similar secular improvement in the quality of representation with consistently strong partisan powers (Fig. 8). The inadequate quality of representation in the early years may contradict the conventional wisdom that Costa Rica has long had a healthy party system. The truth is, however, that some of the virtues of the Costa Rican party system are relatively recent developments. One of the two main parties, the Partido Unidad Social Cristiana (PUSC), did not form as a very united alliance until 1966 and did not take its present name until 1986. In addition, the personalist vote often exceeded 20 percent before 1978 as a result of the incomplete unification of the center-right and a legacy of predominant personalism before the 1948-49 revolution. When the PUSC became consolidated as a party and personalist vehicles declined in importance, the quality of representation rose to the level of respect it enjoys today. Uruguay The other four cases show a pattern of oscillation between strong and weak partisan powers, usually with adequate- or high-quality representation. Uruguay has experienced movement basically in one direction: from moderately strong partisan powers from 1942 to 1966 to partisan powers below the .5 threshold and falling further in successive elections from 1971 to 1999. This movement has coincided with some improvement in the quality of representation (Fig. 9). These are the consequences of a transition from the party system dominated by Blanco and Colorado competition to a three-party system, caused by the growth of the Frente Amplio beginning in 1971. These tendencies are actually probably understated in the figure because for some purposes fractions, not parties, are the relevant political actors. These fractions are clearly ideological, so the true quality of representation would be much higher if they were taken into account. Fractions also are the key players in coalition politics, which produces dramatic declines in the presidents partisan powers over the course of each administration as fraction after fraction deserts the president to position itself for the next election (Altman 2001). If Figure 9 were redrawn using fraction-level data, there would be much more frequent and dramatic oscillation between strong and weak partisan powers and a much higher quality of representation. Chile The partisan powers data for Chile (Fig. 10) are based on coalition support for the president rather than just support from his own party, as long-lasting coalitions are fairly typical of Chile. If they were based on party support alone, Chile would be confined to the high-weak quadrant. This illustrates, then, the crucial importance of coalition-building in fragmented party systems. Chile shows dramatic swings from strong to weak partisan powers and back as a result. Such swings are also more likely in a disciplined party system, in which small changes in coalition membership can have a large impact on partisan powers. The swings are also driven in part by the Chilean electoral calendar before 1973, which confronted presidents with midterm

13 elections and usually forced them to spend their first years working with a congress inherited from the previous administration. The quality of representation, in conformity with the stereotype of the Chilean party system, has been uniformly high. It was noticeably lower from 1965 to 1973 due to the polarization between left and right, which actually reduced the diversity of bloc representation. Nevertheless, even during those polarized years the quality of representation remained higher that it is in most countries. Venezuela Venezuela has experienced quite significant swings from systems of clearly weak partisan powers (1963 and 1968), to extremely strong partisan powers (1973 and 1983), to minority governments barely above the threshold (1978 and 1988) (Figure 11). Presidents with weak partisan powers attempted to form coalitions, which surely improved their partisan powers while they lasted, but these coalitions were not very durable. The quality of representation from 1963 to 1988 was adequate, although not impressive; but unlike Uruguay and Chile, QR dipped below the minimal threshold in the 1958 and 1993 elections that bracketed this period. The low quality of representation was due to strong personalism, which increased further in 1998-2000. This figure highlights the fact that the renowned Venezuelan party system deserved its praise only during a fortunate period from 1973 to 1988, when it possessed the best of both dimensions. Before and after that period it has had significant problems. Peru Peru experienced a slow swing from adequate representation and weak partisan powers during its transitional constituent assembly (1978), to the best of both dimensions under Belaunde (1980) and Garca (1985), and then to the worst of both dimensions in the more fragmented and personalistic election of 1990. Following his 1992 autogolpe, Fujimori held congressional elections that produced a partial and short-lived improvement in 1992 (despite the APRA and AP boycotts, as these parties were much diminished by then anyway). But by 1995, the system had moved to the low-quality representation and excessive partisan powers that seem to be typical of authoritarian party systems. General Tendencies These data allow us to propose several generalizations about the relationship between partisan powers and the quality of representation. A first generalization, which is so obvious in the preceding survey that it requires no test, is that partisan powers and the quality of representation constitute separate dimensions. The characteristics in figures 2-12 vary both vertically and horizontally so independently that they could not be reduced to a single dimension (i.e., a straight line in these figures) without a significant loss of information. The independence of these two dimensions is even more obvious when the data for all these countries is pooled, as in Figure 13. Knowing how strong partisan powers are does not enable one to predict the quality of representation very accurately, and vice versa. These two characteristics of party systems should be measured separately. There is, however, a significant empirical association between these dimensions that

14 suggests a tradeoff between partisan powers and the quality of representation. Figure 13 includes a regression line to illustrate the general tendency captured by the correlation of -.535: an inverse relationship between these two dimensions. Most of the cases cluster in the upper-left highweak quadrant or the lower-right low-strong quadrant. The good news is that the worst of both dimensions is very rare; the bad news is that most cases are either governable at the expense of the quality of representation, or representative at the expense of governability. However, there are some cases that satisfy the best of both dimensions. Those nearest the center of this upperright quadrant are Chilean coalitions; if they were recoded on the basis of party size alone, they would shift to the upper left quadrant, making the tradeoff more severe. Most of the remaining best of both dimensions party systems would then be found at a much lower quality of representation. The regression line suggests that the most functional party systems are most likely to be found in a smaller zone within the upper-right quadrant: when partisan powers are between .5 and about .9, and when QR is between 1 and approximately 2 (which corresponds to two or three meaningfully distinct parties). If partisan powers are stronger than this, the quality of representation tends to suffer; if the quality of representation is higher than this, partisan powers tend to be too weak for governability. What happens when the party system tends toward the worst of both dimensions? There are only three instances in this sample: Argentina in 1965, Peru in 1990, and Venezuela in 1993. All three were soon followed by an interruption of the democratic regime. Argentina sank into seven years of military rule in 1966. In Peru, Alberto Fujimori carried out a presidential coup less than two years after winning the presidency in the midst of a party system of low-quality representation that gave him weak partisan powers. In Venezuela, Rafael Caldera was elected in 1993 and managed to finish his term of office. But his successor was Hugo Chvez Fras, who disbanded the Congress, dismantled the supreme court, and had a new constitution written that allowed him to rule with near-dictatorial powers. The common elements in all three situations were parties lacking in legitimacy and unable to govern, which prompted voters and elites to search for an alternative formula for governing. As Table 1 shows, presidents have been less likely to finish their terms when partisan powers are weaker than .5, and far less likely to finish their terms when partisan powers are weaker than .5 and QR is less than one. The worst of both dimensions seems to be a situation to avoid. There are ways to achieve the best of both dimensions, fortunately. The theory developed here suggests that this optimal situation requires a low level of personalism, a moderate degree of party discipline, and one of three combinations of party-system characteristics that aid the construction of working majorities. The first combination is a moderate number of parties: at least two but no more than three ideologically distinct parties. If the number of parties is greater than three, then working majorities can still be formed by the well-known strategy of building coalitions. And even if the number of parties is large and coalitions cannot be formed, shifting majorities can be negotiated if the party system possesses a low degree of ideological polarization. These propositions are little more than hypotheses derived from the theory, however. Careful testing is needed before they can be trusted.

15 Table 1: Regime Breakdowns by Party-System Characteristics Quality of Representation High (>1) Strong (>.5) Low (<1) High (>1) Weak (<.5) Low (<1) 2 3 .67 2 5 42 34 .05 .15 .19 Partisan Powers Number that broke down 2 Number of cases in category 43 Rate of breakdown

.05 .05

16 Appendix The partisan powers index (PP) is based on a cumulative probability distribution function interacting with party discipline within the bounds of a floor (determined by the level of polarization) and a ceiling (set by the chance of a fallout between the governing party and the president). The formula is (3)

PP = floor + disc(1 floor loss)[1 e( e

10disc 2 )( size10disc )

FLOOR establishes the lower limit on the president's legislative success rate. It is operationalized as (1 - .01*polarization)2. This term is squared because passage of the presidents bills requires agreement between two sides. It can be thought of as the probability that the governing parties will support a bill times the probability that the opposition parties will support that bill. Each probability alone is assumed to be inversely proportional to polarization. The second term, DISC(1 - FLOOR - LOSS), establishes the range of possible success above that floor, and therefore sets the upper limit on legislative success. DISC is the party discipline estimate. LOSS is an estimated probability of a falling out between the governing party and the president. The term in brackets defines the parameters of the S-curve that links the floor to the ceiling. It is based on a cumulative probability distribution function that models increasing legislative success as the size of the governing party (SIZE) increases. How steeply the rate of success increases for each increase in party size depends on party discipline. If discipline is very loose, increasing the size of the party has little impact. If discipline is very tight, increases in party size make little difference before the party approaches 50 percent, where legislative success suddenly increases sharply, approaching 1; and then further increases in party size have little additional impact. The SIZE variable in this sample refers to the president's party alone except for Chile, where I have used Arturo Valenzuela's data on coalition composition instead (Valenzuela 1994, 197-99). As long-lasting coalitions have been more typical of Chile than its neighbors, this exception probably improves the validity of the partisan powers index for predicting consequences.

17 References Altman, David. 2001. The Politics of Coalition Formation and Survival under Multiparty Presidential Democracies. Ph.D. dissertation, University of Notre Dame. Burke, Edmund. 1774. Speech to the Electors of Bristol. In Edmund Burke, Works, Vol. II. Coppedge, Michael. 1995. "Instituciones y gobernabilidad democrtica en Amrica Latina." Sntesis (Madrid) 22 (July-December). -----. 1997. A Classification of Latin American Political Parties. Kellogg Institute Working Paper No. 244 (November). -----. 1998. The Dynamic Diversity of Latin American Party Systems. Party Politics 4:4 (October ): 547-68. Also published in Spanish as La diversidad dinmica de los sistemas de partidos latinoamericanos, POSTData (U. Quilmes, Argentina) 6 (July 2000): 109-134. -----. Forthcoming 2002. Combining Large- and Small-N Approaches in Democratization Research. in David Collier and Gerardo Munck, eds., Regimes in Latin America: Concepts, Methods, and Findings. Baltimore: Johns Hopkins. An earlier condensed version was published as Thickening Thin Concepts and Theories: Combining Large N and Small in Comparative Politics. Comparative Politics 31:4 (July): 465-76. Held, David. 1996. Models of Democracy. 2nd ed. Stanford: Stanford. Huntington, Samuel. 1968. Political Order in Changing Societies. New Haven: Yale. Jones, Mark P. 1995. Electoral Laws and the Survival of Presidential Democracies. Notre Dame: Notre Dame. Katz, Richard. 1997. Democracy and Elections. New York: Oxford. Laakso, Murkuu and Rein Taagepera. 1979. Effective Number of Parties: A Measure with Application to Western Europe. Comparative Political Studies 12: 3-27. Lijphart, Arend. 1984. Democracies: Patterns of Majoritarian and Consensus Government in Twenty-One Countries. New Haven: Yale. Lijphart, Arend. 1999. Patterns of Democracy: Government Forms and Performance in ThirtySix Countries. New Haven: Yale. Linz, Juan J. and Arturo Valenzuela, eds. 1994. The Failure of Presidential Democracy. Baltimore: Johns Hopkins.

18 Mainwaring, Scott. 1993. Presidentialism, Multipartism, and Democracy: The Difficult Combination. Comparative Political Studies 26: 198-228. Mainwaring, Scott and Matthew Soberg Shugart. 1997. "Conclusion: Presidentialism and the Party System." In Mainwaring and Shugart, eds., Presidentialism and Democracy in Latin America, 394-439. Cambridge: Cambridge. Morgenstern, Scott. 1999. Explaining the Unity of Legislative Actors in the Southern Cone. Draft prepared for the conference on Los Partidos Polticos ante los Retos del Siglo. O'Donnell, Guillermo. 1993. "On the State, Democratization, and Some Conceptual Problems: A Latin American View with Glances at Some Post-Communist Countries. World Development 21: 1355-69. Pitkin, Hanna. 1967. The Concept of Representation. Berkeley: California. Rae, Douglas. 1970. The Political Consequences of Electoral Laws. 2nd ed. New Haven: Yale. Sartori, Giovanni. 1976. Parties and Party Systems: A Framework for Analysis. Cambridge: Cambridge. Shugart, Matthew S. and John M. Carey. 1992. Presidents and Assemblies: Constitutional Design and Electoral Dynamics. Cambridge: Cambridge. Valenzuela, Arturo. 1994. Party Politics and the Crisis of Presidentialism in Chile: A Proposal for a Parliamentary Form of Government. In Juan J. Linz and Arturo Valenzuela, eds., The Failure of Presidential Democracy, pp. 165-224. Baltimore: Johns Hopkins.