Conditional Consociationalism: Electoral Systems and Grand Coalitions

Transcription

1 Conditional Consociationalism: Electoral Systems and Grand Coalitions Nils-Christian Bormann March 25, 2011 Abstract Consociationalism is a complex set of rules and norms that is supposed to enable democratic governance and peaceful coexistence of different social segments in plural societies. Statistical studies of conflict often reduce it to either a PR or federalism dummy in a regression. I extract the core definition of consociationalism from Lijphart s writing and explicitly link its institutional and behavioral dimensions. I also address the possible endogeneity of electoral systems and show that once endogeneity is accounted for PR has a positive effect on ethnic elite cooperation although historical, socio-structural and international factors exert a more robust influence. A history of violence in a country seems to antagonize elites and hinder cooperation. Paper to be presented at the ECPR Joint Sessions workshop on Political Violence and Institutions from April in St. Gallen, Switzerland. I thank Manuel Vogt and Julian Wucherpfennig for helpful discussion and comments. Center of Comparative and International Studies, ETH Zurich, Switzerland. bormann@icr.gess.ethz.ch 1

2 1 Introduction Was Lijphart (1977, 238) correct in pronouncing that [f]or many of the plural societies of the non-western world (... ) the realistic choice is not between the British normative model of democracy and the consociational model, but between consociational democracy and no democracy at all? The appraisal of the alleged blessings of consociationalism has been incomplete and/or hotly disputed. Most studies focus on the application to single cases, and large-n studies have only gained systematic insight at the expense of conceptual clarity. Too often consociationalism is equated with proportional representation and/or federalism. However, consociationalism is more than institutions. It consists of at least two dimensions: institutions and elite behavior. The behavioral dimension has been completely ignored by quantitative scholars, and it is unclear whether the institutional side, often PR electoral rules, actually increases the likelihood of elite accommodation in a grand coalition, the primary characteristic of consociationalist arrangements (Lijphart, 2002b, 39). Anecdotal evidence seems to suggest that there may be a connection. Up to 1977, politics in Brazil were not structured along ethnic lines. When the military regime partially retreated in 1977 the sizable Afro-Brazilian minority began voicing political demands. Ethnicity became a relevant cleavage in the political system governed by proportional electoral rules. Afro-Brazilians remained powerless for 26 years, but in the early 21st century the Lula government implemented anti-discriminatory reforms, appointed multiple Afro- Brazilians to the cabinet and made some executive power-sharing a reality. 2

3 In the United States, a far more economically advanced but similarly unequal presidential and federal democracy, it took until 2009 before minority ethnic groups gained access to the highest state offices. 1 Halfway around the globe, in a completely different political environment Zimbabwe and South Africa emerged out of years of the worst racial discrimination in the early 1980s and 1990s respectively. While South Africa adopted a completely new constitutional framework, Zimbabwe stuck to its British majoritarian institutions although it reserved some parliamentary seats for minorities. While under Mugabe s rule a majority of Blacks was represented by the government, whites basically left the country and the political exclusion of other linguistically different groups continued and even increased in recent years. In South Africa, under a system of proportional representation the ANC has managed to give representation to all ethnic groups within the country. But did the electoral system really play a decisive role? After all, India has managed to maintain ethnic cooperation among the majority of ethnic groups in the similar catch-all Congress party under a majoritarian electoral system since independence in As much as these examples refer to the interaction between institutions and elite behavior within a country they also address a second fundamental problem of the quantitative literature on institutional effects. To my knowledge, all studies that investigate the consequences of institutions assume them as given. However, institutions understood as the rules of the game are often the outcome of historical legacies, social structure and dis- 1 Although some minority members did serve in the Reagan, Bush, and Clinton administration, none of them really made any claims on representing an ethnic group or tried to push minority issues explicitly. 3

4 ruptive events (North, Wallis and Weingast, 2009). In early writings, Lijphart (1977, 53-99) himself argued that the success of consociational institutions depends partly on exogenous conditions among which he listed colonial history, cleavage structure, or the presence of prior conflicts. Unfortunately, he later cast aside these considerations and nowadays recommends consensus institutions to all societies irrespective of their social make-up (Lijphart, 1999, 1999). Yet, external constraints need to be taken seriously because PR systems surely will work differently in the Netherlands than in South Africa. Put differently, proposing that electoral rules will have the same effects in all countries irrespective of conditions historical conditions is a daring statement or as Cheibub (2007, 25) puts it: The language of institutional choice must be used carefully, since this choice is usually constrained by historical circumstances. In order to take into account the conditions under which electoral systems operate in a country I employ conditional recursive mixed-process models, a generalization of the more widely known bivariate probit regressions. The models allow me to account for the endogeneity of electoral rules and estimating the endogenous effect on ethnic elite cooperation in all states where ethnicity is or was politicized from 1946 until My empirical analysis still falls short of capturing the complexity of consociational systems. Yet, it steps beyond equating consociational democracy with the presence or absence of PR, and makes an explicit link between the institutional and behavioral components of consociationalism. The paper is organized in four parts. First, I will superficially review the long debate on consociationalism and try to dissect from it the core definition 4

5 of consociationalism. Second, I will argue that attempts to measure consociationalism have been insufficient so far because they have solely focussed on one institutional aspect of consociational theory but ignored its behavioral components. Third, I will develop hypotheses on the conditional nature of consociational institutions and how they influence consociational practices. Fourth, I will describe my data and method in greater detail, and finally analyze the effect of electoral systems on ethnic elite cooperation. 2 Confusing Consociationalism One possible reason for the simplistic equation of consociationalism with PR electoral rules and the complete disregard for the behavioral pillars may be Lijphart s frequent reformulations of consociationalist theory. In this section, I will give a concise summary of the evolution of consociational theory. Since its first formulation in the late 1960s (Lijphart, 1968), the concept of consociational democracy has been a moving target. Lijphart and his critics altered and redefined the concept multiple times and extended it far beyond its originally conceived scope (Lijphart, 1977, 1985; van Cranenburgh and Kopecky, 2004; Van Cranenburgh, 2006). Lijphart first described consociational democracy in the Netherlands as seven implicit political rules governing elite behavior (Lijphart, 1968) a characterization lacking any mentioning of institutions and one he never returned to again (Bogaards, 2000, 400). Later consociational democracy was integrated with the prevailing types of democracy: majoritarian, centripetal and centrifugal democracies (Almond, 1956; Lijphart, 1969). It was understood as the type of 5

6 democracy present in the low countries in Europe and juxtaposed with the Anglo-American majoritarian version as an alternative stable democratic system. In these early formulations, the institutional setup was inextricably linked to the social structure of society. The more homogeneous societies of the West could sustain the conflictual atmosphere inherent in majoritarian democracy, while the plural societies of the third world could not. Accordingly, Lijphart moved from case studies of smaller European countries to a comparison of states around the globe. Here for the first time, consociational democracy was clearly defined by four pillars: (1) a grand coalition of elites from different groups, (2) a veto for each group in important policy areas, (3) proportional representation in key institutions, and (4) group autonomy (Lijphart, 1977, 25). The first pillar, the grand coalition, is the primary characteristic of consociational democracy, while the three remaining pillars are secondary instruments. Behavioral components, i.e. elite cooperation, and institutional rules, e.g. federalist arrangements as an expression of group autonomy, are both equally important features of consociationalism. Later, consociational democracy was re-baptized power-sharing democracy (Lijphart, 1985). Now, it was recommended not for plural societies but for post-conflict environments like Northern Ireland (Nagle and Clancy, 2011), Bosnia (Monteux, 2006), or South Africa (Lijphart, 1998). Over time, power-sharing took on various meanings and lost its erstwhile consociational definition. While it could be conceived that power-sharing is a more general and more loosely defined collection of conflict-resolution strategies than consociationalism with its more rigid definition, the two have been too often conflated in practice. Starting with Power-sharing in South Africa, Li- 6

7 jphart has used power-sharing as a synonym for consociational democracy (Bogaards, 2000, 415) and this has become common practice in academic writing. 2 This had unfortunate consequences, as power-sharing is sometimes used as a description of any post-conflict solution. Sisk (1996), for example, subsumes Lijphart s consociational and Horowitz vote pooling suggestions under the heading of power-sharing, although the two are usually regarded as rival, or even polar opposite, concepts (Bogaards, 2000, 416). Returning full-circle to the older juxtaposition of majoritarian and consociational democracy Rothchild and Roeder (2005) have advanced power-dividing institutions based on the US constitutional model as the best method to resolve conflict in plural societies. Finally, in a series of statistical studies, Hoddie and Hartzell have lumped together power-sharing and power-dividing institutions into one compound measure to assess their effect on post-conflict peace (Hoddie and Hartzell, 2005; Hartzell and Hoddie, 2007). Partly due to the complexity of the consociational package, partly due to a simple lack of data, Lijphart turned to the more tractable concept of consensus democracy (Lijphart, 1984, 1999) that focuses on purely institutional characteristics of established democracies 3. Surprisingly, Lijphart found two consensus dimensions instead of a simple one dimensional majoritarian-consensus continuum (Lijphart, 1999, 248). In short, the executive-parties dimensions assesses the broadness of the government coalition, while the federal-unitary 2 In From Power Sharing to Democracy the first chapter is entitled Debating Consociational Politics (O Leary, 2005) and in Consociational Theory the first chapter s heading is Power shared after the deaths of thousands (McGarry and O Leary, 2009). Subsequent chapters in both books alternate between power-sharing and consociationalism. 3 Although note the later addition of corporatism, clearly a more informal, potentially even a behavioral arrangement (Lijphart and Crepaz, 1991). 7

8 dimension captures a separation of powers outside the parliament. Interestingly, the only first-dimension component of consensus democracy that can be directly engineered seems to be a proportional electoral system which, ironically, has the lowest correlation to the index measuring consensus democracy (Taagepera, 2003, 7). Given the confusion of concepts and types in Lijhart s own writings as well as in the expansive literature on power sharing, it is little wonder then that statistical studies trying to measure consociationalism have frequently resorted to a simple dummy variable indicating the presence of a system of proportional representation (Reynal-Querol, 2002; Schneider and Wiesehomeier, 2008; Brancati and Snyder, 2010; Selway and Templeman, forthcoming). However, using a dummy for proportional electoral system is not faithful to the concept of consociationalism. It is not even a good proxy for the pillar of proportionality which Lijphart described as a method of allocating civil service appointments and scarce financial resources in the form of government subsidies among the different segments (Lijphart, 1977, 38). If a credible link could be established between PR electoral rules and the proportionality principle in key state institutions, especially the civil service or the army, and the proportional distribution of resources, then a PR dummy would be a partial proxy for a sub-component of consociational democracy. Yet, civil service representation of ethnic groups may have very different sources. Two of the countries that are regularly credited with effective accommodation of their diverse population are Botswana (Rothchild, 1997, 78/9) and India (Lijphart, 1996). Neither of these two countries make use of proportional electoral rules but both have managed to achieve proportional 8

9 outcomes in resource distribution and/or civil service appointments (Jalal, 1995, 20-1). Moreover, countries like Spain or Italy that do have PR electoral rules are not or at least not any longer considered consociational prototypes (Bogaards, 2005). This is not to suggest that Arend Lijphart has not regularly argued for the adoption of PR rules in the electoral arena. Quite on the contrary, he has repeatedly recommended PR in his work (Lijphart, 1985, 1999, 2002b, 2004). However, PR is only a small part of consociationalism, it has little to do with the proportionality concept in consociational theory, and does not really differentiate between consociational and non-consociational cases. Some authors also assess the difference between parliamentary and presidential systems (Gerring and Thacker, 2008) or the presence and absence of federal arrangements (Selway and Templeman, forthcoming) as operationalizations for consociationalism. However, as noted above, presidentialism is not a central feature of consociationalism although Lijphart does seem to favor parliamentary systems (Lijphart, 2004, 101-3). Additionally, introducing federalism as a consociational indicator confuses more than it explains. Federalism is only a good measure of group autonomy if it overlaps with group boundaries, not if it cuts across them. US- or German-style federalism is clearly different from the arrangements in former Yugoslavia, in India or in South Africa; the convoluted federal arrangement in Belgium might even be so different that it escapes any attempt of comparative analysis. Specific cases aside, a dummy indicating the presence or absence of federalism does not adequately measure the presence of consociational systems. Moreover, group autonomy can also be assessed by non-federal solutions like ethnically 9

10 separated legal codes, self-managed educational institutions or the recognition of other linguistic or religious prerogatives. In light of the lack of adequate empirical measures of consociational concepts, I will argue for an ideal measurement close to the original definition, and then introduce a sub-optimal but still much improved proxy for consociatioal arrangements. 3 Getting Consociationalism Before describing an ideal consociational measurement it is important to ask what cases should be included in the measurement? A good starting point is Lijphart s conjecture that consociational democracy is appropriate for plural societies (Lijphart, 1977) which are characterized by politically and socially charged as well as differentiable segments. Of the classical consociational cases, the Netherlands are not a good example any longer, as Lijphart admitted himself (Lijphart, 2002a). I suggest to focus instead on societies where ethnicity is a politically charged category. While other lines of division exist within a society, ethnically charged nationalism is among the most divisive and most consequential of cleavages. Ethnic fragmentation is linked to a number of vices within societies like a lower provision of public goods (Alesina, Baqir and Easterly, 1999), lower economic growth (Alesina and La- Ferrara, 2005), and a higher probability of ethnic civil wars (Buhaug, 2006). If ethnicity becomes politically mobilized and the legitimacy of rulers is derived from their ethnic basis, ethno-nationalism can become the grounds for violent struggles for power (Wimmer, 2002; Wimmer, Cederman and Min, 10

11 2009). Indeed, the pervasive logic of the nation-state makes ethnic cleavages more often a salient political identity than other types of identities (Wimmer, 1997, 632), and thus more relevant for explaining institutional engineering. 4 In order to gauge consociationalism accurately in ethnically diverse societies, what should be measured? At best, this measure should include an indicator of a grand coalition in the executive of all relevant ethnic groups, an indicator of territorial and cultural autonomy regulations, an indicator assessing proportionality in the bureaucracy, the military, the legislature, the judiciary and in the distribution of resources, and finally, an indicator of de facto and de jure veto rights. These individual measures would measure proportions and deviations from an ideal distribution, and be aggregated into one scale to assess the degree of consociational practices in a country. Unfortunately, data for the majority of consociational pillars does not exist. Existing measures of consensus democracies (Lijphart, 1999) or veto players (Tsebelis, 2002) are not of much help. Their temporal scope is extremely limited, they are restricted to economically advanced states, and include cases independent of the plurality of their societies. Data on proportional representation in key institutions is almost non-existent outside of Europe, and data about cultural autonomy as coded in the Minorities at Risk dataset is troubled by selection bias. Given the absence of empirical indicators of consociationalism s key dimensions is there any way to improve upon a dummy for proportional representation? I believe that at the very least, an empirical indicator should 4 A focus on countries where ethnicity is politically relevant does only exclude a minority of all countries from the analysis. Examples include Burkina Faso, Germany or Ireland. In more than 150 states with more than 500,000 inhabitants, ethnicity is politically relevant. 11

12 represent the institutional as well as the behavioral side of consociationalism. As argued above, consociationalism is mostly about elite accommodation, and the rules to ensure it. At the same time, Ljphart has argued that proportional representation will lead to this elite accommodation (Lijphart, 2002b, 53). While PR is an imperfect proxy for the proportionality principle of consociationalism it is hypothesized to be the institution that enables the grand coalition. This is exactly the causal mechanism which Lijphart s main critic Donald Horowitz has challenged by arguing that a simple presence of ethnic groups in parliament does not incentivize elite cooperation (Horowitz, 2002). Acknowledging both the behavioral and the institutional dimension of consociationalism, I suggest to measure the grand coalition as the proportion of the ethnically relevant population included in the central government. Data on ethnic inclusion was recently assembled by Cederman, Wimmer and Min (2010). The data is especially pertinent because it measures de facto cooperation at the elite level. Not only has Lijphart repeatedly alluded to the centrality of the grand coalition among consociational features (Lijphart, 2002b, 38-9), he has very much emphasized the elitist nature of the grand coalition, going so far as to call it an elite cartell. On the institutional side, several measures of proportionality exist but there are important differences between them. Golder s classification of electoral systems around the world is probably the best as it records electoral rules from , and has the decisive advantage of only counting electoral rules that are actually used but not those that only exist on paper (Golder, 2005). Looking at the interplay of institutions and behavior now comes much closer to Lijphart s original 12

13 formulation of consociationalism and also, for the first time, actually assesses the hypothesized relationship. Previous research has already demonstrated that ethnic inclusion or Lijhpart s grand coalition decreases the risk of ethno-nationalist civil war substantially (Cederman, Wimmer and Min, 2010). However, it is not clear what drives inclusion of different ethnic groups at the center. In the next section I will address the possible links between electoral rules and the inclusiveness of government. 4 Conditional Consociationalism How to arrive at elite cooperation in a grand coalition is probably the most rigorously debated question in the long debate on consociationalism (Lijphart, 1977, 1985; Horowitz, 1991, 2000, 2002; Lijphart, 2002b). Pre-1975 Lebanon and post-1995 Bosnia where elite representation of all ethnic groups in the executive was/is prescribed by the constitution are grist to the mill of consociational critics. Both cases vividly exemplify that ethnic elites mere presence in government institutions does not entail cooperation between the different segments. 5 This is why Lijphart and his supporters have later promoted proportional representation as the liberal consociational solution (McGarry and OLeary, 2007, 675/6) in which groups are not pre-determined but those political forces that are most appealing to the electorate find expression in parliament. Nonetheless the question remains whether elites opt 5 Another disheartening example of the negative effects of the prescribed segmentation of an electorate is colonial India, where Hindus and Muslims voted in different constitutencies despite being neighbors. The bloody history of partition in 1947 is interpreted by some scholars as a result of exactly that segmentation. 13

14 to cooperate only because they are proportionally represented in parliament. The critics of consociationalism contend that ethnic elites, especially in polarized environments, have few incentives to cooperate and prefer deadlock or coercive control (Lustick, 1979) to accommodation. Research on civil wars has reinforced the perception of ethnic group representatives as strategic and rational actors (Stedman, 1997; Cunningham, 2006; Metternich and Wucherpfennig, 2010). From the theoretical vantage point of self-interested ethnic elites Horowitz and other critics of consociationalism have argued for vote pooling mechanisms that force ethnic elites to reach across ethnic boundaries in order to gain parliamentary majorities. They assume a strongly constructivist view of ethnicity when arguing that appropriate electoral rules are able to alter ethnic allegiances (Posner, 2005; Chandra, 2005). There are several problems with the institutional proposal and the underlying assumption. First, Bogaards (2003) shows that the suggested mechanism does not work in many African states and suggests an alternative electoral system that he coins constituency pooling. Second, both Bogaards and Horowitz suggestions lack any comparable empirical record. Bogaards draws his example from a now defunct electoral mechanism in Uganda and Horowitz only cases are Papua New Guinea from 1964 until 1975 and the short-lived Fiji constitution of Third, I believe that the absence of AV electoral mechanisms as well as their inappropriateness for several of the most ethnically divided countries is due to a lack of appreciation for their endogenous adoption. In other words, in those societies where ethnicity is a politically charged category, it will be very difficult to convince leaders of ethnic segments to agree to letting go of their powerful positions. AV systems are so rarely found be- 14

15 cause they are not appreciated by ethnic entrepreneurs. On the contrary, PR systems increase the likelihood that ethnic leaders will keep their powerful positions even if an ethnic group is not geographically concentrated. Which factors then influence their adoption? A few years ago the adoption of PR systems in the established democracies was already investigated by Boix (1999) and Blais, Dobrzynska and Indridason (2005). 6 While developed Western countries are not free of tensions between ethnic segments, most problems occur in the developing countries. What are then the driving factors behind electoral systems in a expanded sample? Colonial heritage exerts probably the strongest influence on the adoption of any electoral system. Former Spanish as well as Belgian and Dutch colonies usually employ PR electoral systems, while British colonies most often run majoritarian systems. Similarly, the majority of former French colonies inherited the French two round run-off system which is majoritarian in nature. At the same time several continental European countries operate PR systems. In Lijphart s 1999 sample of advanced democracies three out of four consensus democracies are economically highly developed while only one in two majoritarian democracies show similar levels of development (Muller-Rommel, 2008). More generally, democratic systems are much more likely to be present in economically highly developed countries. PR systems are also more likely to be found in smaller countries. The operation of list-pr systems that only feature one country-wide district is hardly feasible in countries beyond a certain threshold. In larger countries, local electoral districts are much more appealing. Finally, the violent history of a 6 Also see Brambor, Clark and Golder (2006, 79) for a critique of Boix s findings. 15

16 country should play a role. In the wake of civil wars new electoral systems are adopted, and more recently these systems have been proportional systems. Examples include Northern Ireland, Bosnia, Iraq, South Africa or Indonesia. As about two thirds of all civil wars recur, not taking into account the possible endogeneity of electoral systems to violence is a possible omission in estimating their causal effects. Regarding elite cooperation, Lijphart himself formulated a number of favorable conditions for consociationalism to work. Conducive to elite cooperation are the balance of power among ethnic segments, smaller states, a developed sense of nationhood, cross-cutting cleavages, and traditions of elite accommodation (Lijphart, 1977, 53-99). Bogaards (1998, 478) who peruses all of Lijphart s publications on the conditions for consociationalism adds the geographical concentration of segments, the absence of a majority segment, socioeconomic equality between segments and the moderation of party systems. Several of this conditions like national loyalties or socio-economic equality are rather effects or by-products of consociational practices than causal priors. Other factors, like country size, colonial history, cross-cutting cleavages or the absence of a majority segment can more safely be assumed exogenous. Returning to electoral rules it should be noted that a grand coalition can also exist within authoritarian regimes. Indeed some states in Africa like Côte d Ivoire under Houphouet-Boigny built up sophisticated clientelist systems of elite accommodation. On average however, dictatorships like Al- Assad s Syria, Indonesia under Suharto, or Guatemala under military rule should have lower levels of ethnic inclusion than their democratic competitors. Combining the expectations of endogeneity and the inclusiveness of 16

17 democratic regimes I hypothesize: H 1 Considering historical and sociostructural constraints, PR electoral rules increase the degree of ethnic inclusion in comparison with nondemocracies. When evaluating Lijphart s main argument, it is necessary to note that he always juxtaposed the British-style majoritarian system with the proportional consociational system. Majoritarian electoral systems would inevitably lead to tyranny of the majority, especially under plurality regulations: For many of the plural societies of the non-western world, therefore, the realistic choice is not between the British normative model of democracy and the consociational model, but between consociational democracy and no democracy at all. This is not difficult to fathom. As soon as an ethnic group makes up a plurality in the population it is attractive for group leaders to directly appeal to their own group for votes. Although PR is assumed to lead to similar dynamics, the distortive nature of first-past-the-post rules severely aggravates the outcome. Ceteris paribus, in majoritarian systems groups smaller than 51% can already dominate the center while in PR systems that should not be possible. Additionally, in PR systems very small groups will be more likely to gain representation and can become crucial players in coalition governments a path closed to them in majoritarian systems. H 2a Considering historical and sociostructural constraints, PR electoral rules increase the degree of ethnic inclusion in comparison with majoritarian systems. 17

18 5 Analysis 5.1 Data My dataset consists of over 5000 country-year observations from 1946 until The dependent variable is the proportion of the included population of the politically relevant ethnic population as coded in the Ethnic Power Relations (EPR) dataset (Wimmer, Cederman and Min, 2009). The included population is defined as the sum of the relative sizes of those ethnic groups in a country that have de facto access to executive positions. In non-democratic states executive power can be vested in a variety of institutions like a royal family, the top army brass or revolutionary councils. In multiple countries the ethnically relevant population is not tantamount to the overall population. Prior to the passing of the voting rights act in 1965, for example, Asians or Hispanics are not coded as a relevant ethnic categories in the United States. Taking the proportion of the included population would therefore paint a too optimistic picture of the actually included population. 7 As an alternative dependent variable I also code the relative number of included groups from all politically relevant ethnic groups in a given country and year. There are vast differences between the most ethnically diverse states like Russia with up to 60 groups and less plural societies like those in South America with only two or three groups. As my principal independent variable I take Golder s classification of electoral systems from (Golder, 2005). Each major family of electoral 7 A group is defined as ethnically relevant if representatives of the group make political claims on behalf of the entire group or if the group is discriminated against by the state. 18

19 systems, i.e. PR, mixed and majoritarian rules, takes a value of 1 if it was in effect in a given year. The base category are non-democratic regimes which take a value of zero. In alternative specifications I also test the impact of a less precise classification of electoral systems (Gerring and Thacker, 2008) which is coded from 1946 until The majority of my control variables comes from the Ethnic Power Relations dataset. In order to test the historical influence of elite cooperation, I include the degree of inclusion in the first year of independence of a state. Additionally, I control for the size of the largest group as a proxy for the balance of power between ethnic segments. I expect countries with numerically dominant groups, to have a lower degree of inclusion as those groups have less incentives to build alliances with other ethnic groups to attain executive power. 9 To gauge the effect of cleavages, I make use of an ethno-linguistic fractionalization index calculated from the groups in the EPR dataset. I also use the linguistic distance measure by Fearon (2003). Higher degrees of ethno-linguistic fractionalization or cultural distance should decrease the included population as there are less symbolic ties that bind elites and the groups they represent together. Additionally, I include a count of previous ethnic civil wars in a country to assess the hostility of inter-ethnic relations Gerring & Thacker often backdate their measure of electoral systems even if it was not used in practise due to authoritarian rule. 9 In alternative specifications I considered the log of the largest group and a quadratic specification to control for the fact that with increasing size the dependent variable of inclusion may be growing in direct proportion. It could be that the effect of the largest group reaches a maximum at mid-range before inclusion grows because the largest group is growing. This effect should not play a role in specifications using the proportion of included groups as a dependent variable. 10 The conflict data is adopted from the Expanded Armed Conflict Data assembled by Cunningham, Gleditsch and Salehyan (2009). 19

20 More civil wars should translate into a lower degree of ethnic inclusion. To gauge the effect of external threats, I include the international rivalry data from Klein, Goertz and Diehl (2006). According to Lijphart a greater feeling of external threats should lead to a higher degree of internal cooperation. Furthermore, I incorporate data on country size and economic development from Gleditsch (2002). A larger population should increase the complexity of ethnic relations and therefore decrease the size of the included population, while GDP/capita, as an indicator of modernity, should exert a positive effect on the included population. Finally, I include dummy indicators for world regions and colonial history from the Quality of Government project (Teorell et al., 2010). Summary statistics and correlations between the main variables can be found in tables 3 and 4 in the appendix. 5.2 Method: Conditional Mixed Process Regression In order to correctly estimate proportions which are continuous between zero and one, I utilize the Tobit regression model with a lower and upper boundary. 11 As argued above the influence of electoral systems on inclusion may be endogenous to a variety of factors such as colonial heritage or economic development. Therefore, I estimate the endogeneity of electoral systems with conditional mixed-process (CMP) models (Roodman, 2008). CMP regression is a generalization of Seemingly Unrelated Regression (SUR) models that can be used with response variables that are not continuous but bounded at an 11 I also tried OLS regressions. The results were not substantially or statistically significant different. Optimally, I would have employed beta regressions but they are not available for conditional models. I ran a few one-stage beta and logistic regressions and they provided similar results as simple tobit or OLS models. 20

21 arbitrary threshold value and are assumed to have a normally distributed error. Normally distributed errors are not unique to probit models but are a characteristic of several Maximum-Likelihood (ML) estimators like multinomial probit, ordered probit or tobit regression models. A more widely-known variant of a CMP model is a bivariate probit regression model which allows bivariate normal error distributions from the SUR framework and the inclusion of an endogenous indepdendent variable in the second-stage equation. Essentially, two equations are estimated together and the endogenous variable y1i appears as a predictor in the second equation: y1i = Xβ 1j + ε 1i y2i = Xβ 2j + δ 2 y 1i + ε 2i where Xβ is a matrix of j independent variables and their respective coefficients. y 1i is dummy variable indicating the choice of an electoral system. It enters the second equation as an endogenous factor estimated by δ. In a probit regression framework the latent variable y1i would theoretically indicate the utility of choosing one electoral system over another. It is further assumed that the assumed outcome is observed whenever the utility of choosing the alternative is positive: y 1i = g(y1i) = (1{y1i > 0}) y 2i represents the degree of ethnic exclusion at the center. As a proportion it cannot take values below zero or above one. A tobit regression model with an upper and lower threshold captures these natural boundaries. c if y2i c y 2i = g(y2i) = y2i if c < y2i < c c if y 2i c 21

22 The linked error distributions consist of a shared part η and the unique part u: ε 1i = η i + u 1i ε 2i = η i + u 2i Moreover, the bivariate error variance ε is assumed to be normally distributed with mean zero and variance Σ: ε = (ε 1, ε 2 ) N (0, Σ) Σ captures the correlation ρ between the omitted factors explaining the choice of an electoral system y 1i and the degree of ethnic cooperation y 2i. 1 ρ Σ = ρ 1 Essentially, this means that unidentified factors like elite choices are connected through both equations. As argued above I expect ρ to be negative since PR should more often be present in those situations when there is conflict over inclusion, i.e. when inclusion is smaller. The model can be identified in one of two ways. In the classical simultaneous equation model an exclusion restriction needs to be introduced. Put differently, an appropriate instrument needs to be entered into the first stage regression. I utilize colonial heritage as a clearly exogenous factor driving the adoption of electoral rules from a colonial power. However, I admit that colonial history is not the best instrument since a large number of colonies were and are authoritarian regimes that do not hold elections. A second possibility is to identify the model based on the assumption that the error distributions are truly bivariate normal distributed. Admittedly, this is a somewhat heroic assumption which is hard to defend. However, an indirect 22

23 way to assess the viability of the assumption is to try out different error distributions. Recently, Winkelmann (2009) has developed bivariate ML models that rely on a number of Coppola distributed errors. If the results hold under different error distributions, a higher confidence in their reliability would be achieved. 5.3 Results In this section I present two regression tables, one cross-sectional analysis from the 1990s and one time-series cross-section analysis from 1946 until Table 1 presents the conditional estimates of majoritarian and PR electoral systems in 1996, and their effect on ethnic inclusion at the center in Ethnic inclusion in 2004 is still strongly shaped by the extent of ethnic cooperation at independence which forcefully underlines a sticky view of (informal) institutions. Another historical variable, the age of a state, has virtually no effect on inclusion. It is either a bad proxy of nationalist consolidation or its effect is washed out by the history of elite cooperation. Other influential indicators are the degree of ethno-linguistic fractionalization and the size of the largest group. Number of wars fought in a given country also increases the degree of inclusion, showing the endogeneity of ethnic inclusion to actual violence. Surprisingly, neither population size nor economic development seem to have any effect on the degree of inclusion. 12 The year 2004 was chosen as the last year in my dataset for which all independent variables other than the electoral systems had no missing values. The year 1996 was chosen because it is the first year after the democratic transitions in the early nineties in which the state system contains the same members as in I also ran analyses from 1995 and 1994 to 2004 which showed no significant differences. 13 The upper part of the table displays the second stage equation and the lower part shows the equation explaining the distribution of electoral systems in

24 Table 1: Variables Cross-section CMP model Incl. Pop. Inclusion Year (0.108)** PR (0.053) Mixed (0.047) Major (0.116)** ELF (0.338)** Stage Age (0.000) Max. Group (0.375)* War Hist (0.008). Log(Pop.) (0.017) Log(GDP) (0.022) Incl. Pop (0.114)** (0.118)** (0.063) (0.054) (0.336)* (0.000) (0.198)* (0.008) (0.018) (0.026) AIC Electoral System Maj. PR Civil War History (0.092) Log(Population) (0.114)* Log(GDP/capita) (0.177) Hispanic Colony (0.463)* British Colony (0.398) French Colony (0.540) ρ (0.236). N (0.099) (0.107) (0.155)** (0.311) (0.417) (0.572)** (0.146)** Signif. codes: 0 *** ** 0.01 * ; Standard Errors in Parentheses 24

25 The effect of electoral systems is not straight-forward. If not modeled as endogenous indicators their effect on grand coalition formation in the executive of a country is non-existent. 14 However, if the endogenous adoption including of electoral systems is accounted for in terms of colonial history, economic development, population size and history of violence, both majoritarian and PR electoral systems increase the degree of ethnic inclusion. Additionally, the correlation ρ between the two equations is negative and significant as expected, meaning that PR systems are more often found where inclusion is difficult to establish. This also seems to be true for majoritarian systems in the cross-section. It could be that the effect is caused by those African countries that became democratic in the early 1990s. That view is supported by the fact that majoritarian systems seem to be found in poorer countries despite the fact that the base category are authoritarian systems. Civil war history actually decreases the likelihood of any electoral system although it is not significant. 15 Table 2 includes four conditional time-series cross-section models from 1946 until The first and second column show the effects of endogenous PR institutions and the third and fourth column display the consequences of endogenous mixed and majoritarian systems respectively. 16 Again the strongest predictor of ethnic elite cooperation is the amount of cooperation 14 Cf. 5 for time-series results. I do not represent simple cross-section estimations. 15 It may be that a simple count of small-scale civil wars is not equivalent to disruptive events and that the number of more intensive or a different operationalization of intensity does a better job of accounting for different electoral systems. It may also help to only look at a sample of post-war societies only. 16 All four equations include controls for time-trends that are not shown in the table. Time, operationalized as the cubic polynomial of state age and age of electoral systems does not exert a sizable influence. 25

26 in the first year. Mixed and majoritarian electoral systems (unconditional) have small effects while the conditional estimate of PR systems exerts a pronounced and highly significant influence on inclusion. While none of the control variables from the cross-sectional comparison undergoes any substantive changes in signs, the confidence in their statistical accuracy declines despite an increase in observations. The drop in statistical significance levels of the ELF index and the war history variable was only to be expected. The ELF measure is not time varying and cannot account for changes in inclusion. The war history variable is a count indicator that cannot be high in early years of a countries existence. Increasing the time frame of the analysis allows the introduction of international rivalries into the analysis. Against Lijphart s expectation external threats do not increase internal cooperation. It is rather the other way around. One explanation could be that external threats divert attention from the plight of excluded minorities. An alternative mechanism could be that international rivalries exist exactly because of internal exclusion as in irredentist conflicts. The size of the largest group does not seem to influence the level of inclusion over time. This may be due to the fact that in EPR, ethnicity is coded from a constructivist point of view, i.e., groups can join or dissolve into subgroups. Indeed the mean of the size of the largest group is slightly bigger than the sample average in More importantly, the long-standing exclusion of large swaths of the population in Liberia and South Africa had ended in Regarding the endogeneity of the electoral system, PR rules are more often to be found in smaller and more economially advanced countries. Colonial history now also proves to be a reliable predictor of the presence or absence of PR systems. Against Lijphart s prediction 26

27 that Belgian or Dutch colonies would inherit consociational institutions these countries show a strong strain of authoritarianism. The second hypothesis stated that conditional on the environments in which electoral systems operate proportional systems should lead to higher levels of inclusion than majoritarian systems. Running a naive regression analysis without taking into account the endogeneity of electoral systems as in table 5 on page 40 in the appendix would lead us to believe otherwise. PR systems do not seem to have any effect on inclusion while majoritarian and mixed systems are at least partially successful in increasing the level of inclusion. Even in the cross-sectional analysis presented above, majoritarian systems seem to at least be on par with proportional systems although a direct comparison of effects is difficult if they come from two different equations. 17 In table 2, conditional estimates of majoritarian and mixed models are presented for the period from 1946 until While mixed electoral rules show a similar positive effect on the degree of inclusion as do PR rules, majoritarian electoral rules lose statistical significance. Although their substantial effect is still positive, the confidence intervals are too large to be overlooked. Before drawing too sweeping conclusions one should be wary of these estimates as the selection equations do not do a good job of predicting mixed or majoritarian systems I tried to run a multivariate regression model in which both, majoritarian and PR systems would be introduced as endogenous variables, however, the likelihood was intractable with the amount of data I have at my disposal and the model did not converge. 18 The addition of other control variables in table 6 on page 41 seems to support the reached conclusions but changing the dependent variable to the number of included groups or using a different operationalization of electoral systems as in table 7 weakens the confidence in the results. The different operationalization of electoral systems by Gerring and Thacker (2008) also counts the de jure existence of electoral rules or put differently, authoritarian spells. Therefore, it is not too surprising that both PR and majoritarian 27

28 Variables Table 2: TSCS CMP models of electoral systems Incl. Pop. Inclusion at Independence (0.066)*** PR (0.046)*** Mixed (0.024). Majoritarian (0.021). Incl. Pop (0.067)*** (0.043)*** (0.026) (0.022). ELF (0.171). Stage Age (0.000) International Rivalry (0.019). Largest Group (0.202) Civil War Historty (0.011) Log(Population) (0.010) Log(GDP/capita) (0.015) Incl. Pop (0.064)*** (0.021) (0.071)** (0.025) (0.178)* (0.000) (0.021)* (0.202) (0.011) (0.009) (0.012) Incl. Pop (0.067)*** (0.029) (0.036)* (0.284) (0.176)* (0.000) (0.023)* (0.198) (0.011) (0.018) (0.025) AIC Electoral System PR PR Mix. Maj. Civil War History (0.119) Log(Population) (0.071)* Log(GDP/capita) (0.109)*** Hispanic Colony (0.201)* British Col (0.462)** French Col (0.610)** Belgium or Dutch Col (0.658)** ρ (0.112)*** (0.104) (0.081)* (0.115)*** (0.190)* (0.433)** (0.563)** (0.591)*** (0.091)*** (0.092) (0.062) (0.104) (0.226) (0.634) (0.607). N/A (0.245) (0.080) (0.125)* (0.177)* (0.402) (0.639) (0.586) N/A (0.784) N Signif. codes: 0 *** ** 0.01 * ; Standard Errors in Parentheses; Errors clustered by country 28

29 Why do majoritarian systems have such varying effects in a cross-sectional analysis, in a naive regression and in a conditional TSCS regression? As mentioned above several African countries that do have majoritarian traditions democratized in the early 1990s (Bratton and Van de Walle, 1997). At the same time international activism by the international community led to peace missions that introduced power sharing agreements in the wake of civil wars. Both democratization and these power-sharing effects increased the level of elite inclusion in Africa as can also be seen in figure 1 in the appendix. During the Cold War period majoritarian regimes were rather sparse and often they would exclude at least a small part of their population from de facto access to executive power. The UK excluded the Northern Irish for longer periods of time, Australia even discriminated against the Aborigines, India while more inclusive toward a vast variety of linguistically diverse groups, nevertheless excluded non-hindu minorities and the US, the oldest democracy in the world has struggled to incorporate much bigger segments of Blacks and Hispanics. 6 Conclusion Going back to the examples at the beginning of this paper, do electoral systems really make a difference? Is the PR system in Brazil responsible for the faster inclusion of minorities? Is PR also responsible for the greater stability of ethnic elite cooperation in South Africa? The preceding analysis electoral systems have a negative influence on inclusion since they are now conflated with authoritarian regimes. However, this does not explain the catastrophic performance of PR systems when tested in relation to the proportion of included groups. 29