Nationalization in New and Old Party Systems: Electoral Incentives and Cross-District Coordination

Transcription

1 Nationalization in New and Old Party Systems: Electoral Incentives and Cross-District Coordination Carolina de Miguel, University of Toronto August 2013 Earlier versions of this paper were presented at MPSA (April 2013) and the Council for European Studies (CES) Conference (June 2013). Comments are welcome. Please write to 1

2 Abstract Recent literature suggests that parties in new democracies are different from parties in old democracies in terms of party identification, electoral volatility and clientelism Surprisingly, little work has been done exploring differences in the territorial structure of parties between old and new democracies. This paper uses an original dataset measuring the degree of party system nationalization for 67 democracies between 1970 and 2008 to explore the relationship between the age of parties in a democracy and the degree of party system nationalization. I show that countries with old and consolidated parties tend to be nationalized, whereas countries with young parties show significant variation in the territorial nature of their party systems. I argue that the variation among countries with young parties is explained by differences in electoral institutions. I show that majoritarian electoral institutions increase the costs of cross-district coordination making parties less likely to become nationalized. On the other hand, majoritarian systems tend to encourage candidate autonomy within parties, and I find that allowing for greater candidate autonomy allows parties to spread nationally. The paper tests these two countervailing effects of electoral institutions on party system nationalization. Keywords: democratization; nationalization; parties and party systems; electoral systems; candidate autonomy; mediation analysis

3 Introduction There is a large body of literature arguing that parties and party systems in young democracies are substantially different than parties and party systems in older and more consolidated democracies (Brader and Tucker 2001; Mainwaring 1999; Roberts and Wibbels 1999; Webb and White 2007; Ziegfeld 2009). Parties and party systems in new democracies are associated with greater volatility, weaker voter attachments, more clientelistic practices, and less party system institutionalization. One issue that this literature does not explore, however, is party nationalization. Less is known about whether party systems in younger democracies are also more localized (i.e. less nationalized) than in older democracies. The lack of work on the development of nationalized party systems among newly formed democracies is surprising given the implications that the territorial nature of party systems can have on political phenomena ranging from the nature of policies produced in a country (Jones and Mainwaring 2003; Hicken, Kollman and Simmons 2008; Rodden 2009) to levels of violent conflict and demands for secession (Bakke and Wibbels 2006; Brancati 2006). By way of example, Hicken, Kollman and Simmons (2008) argue that when political competition at the national level occurs between parties that represent specific sub-national [i.e. regional or local] constituencies, then the outcomes of policy debates and conflicts can lead to an oversupply of pork-barrel policies and an undersupply of nationally-focused public goods (Hicken 2009, 3). In contrast, in nationalized party systems, political cleavages are more likely to fall along functional, ideological, or class lines, rather than along lines that correlate with geography (Hicken 2009, 3), which leads to nationalized, comprehensive policy programs 1. These implications for the effective provision of public goods are just one way in which the level of nationalization of a countrys party system can be especially important for new democracies. This paper presents an exploration of the development of nationalized party systems among newly formed democracies by focusing on the role of electoral incentives in hindering or promoting nationalization. I make two claims in this paper. First, I argue that electoral incentives matter for the development of nationalized party systems, yet their effect is not as straightforward as one might imagine. On the one hand, majoritarian electoral systems tend to increase the costs of coordination and thus decrease the incentives to form nation-wide party platforms across many diverse electoral districts. This makes countries with majoritarian electoral systems more likely to have territorialized party systems. On the other hand, parties under majoritarian electoral systems tend to have greater candidate autonomy (or less party control over their candidates) which can in turn help overcome some of the costs of coordination and facilitate nationalization. In other words, the territorializing effect of majoritarian electoral systems is often mitigated by the presence of more flexible party organizations in these types of systems. Second, I argue that the effect of electoral institutions on cross-district coordination only matters among relatively young democracies. As party systems become older and more mature candidates and parties develop effective strategies for building cross-district alliances to form national parties, and are thus less influenced by the electoral system. The paper proceeds as follows. The next section presents a brief literature review on the relationship between democracy and the territorial nature of party systems. This section also introduces the empirical puzzle that motivates this paper: what explains variation in party system nationalization among young democracies? The following section develops a theory on how electoral 1 In a similar vein, Rodden (2009) argues that countries with parties that have national programmatic appeals are more likely to enact progressive redistributive policies, whereas countries dominated by parties that de-emphasize national appeals tend to develop a system of pork-barrel politics that usually results in regressive inter-regional transfers. 3

4 institutions affect cross-district coordination and derives some testable hypotheses. Section four introduces the data and empirical strategy of the paper, and section five presents and interprets the results of the multivariate analysis. Finally, section six summarizes the main findings and their implications. Democratization, Young Party Systems and Party System Nationalization Caramani s work on the nationalization of politics in Western Europe links party system nationalization (or its converse, party system territorialization) to macro-historical processes of modernization and democratization. In his book The Nationalization of Politics (2004) Caramani argues that a variety of macro-forces (such as the consolidation of the nation-state, industrialization, and the advent of mass politics) at the end of the First World War led to a rapid and inexorable development towards national electorates and national party systems in Western Europe. The extension of the franchise and the rise of political and party competition led to a push from parties to capture new electorates and thus to expand geographically. Although Caramani s book focuses on Western Europe his argument suggests that democracies in other regions of the world might follow a similar path towards nationalization. If this is the case then we should expect young and less consolidated democracies to have relatively territorialized or localized party systems (as in Western Europe prior to World War I) whereas older democracies would be more nationalized as their parties progressively acquire the resources and experience to build cross-district alliances. This logic would be in line with existing work that suggests that time matters for the development of stable and well-institutionalized party systems. Several scholars, grounded on a behavioralist tradition, have suggested that voters attachments to parties are built over time as a consequences of repeated interactions. For example, Brader and Tucker (2001) show that so long as parties endure over time, mass partisanship will emerge as citizens acquire an attachment to, or identification with, specific parties (Brader and Tucker 2001, 70). Similarly, Lupu and Stokes (2010) argue that [t]he repeated experience of voting builds partisanship (Lupu and Stokes 2010, 92), and in turn partisanship contributes to stabilize voting behaviour and decrease electoral volatility 2. Similar to these accounts, we might expect time to matter in the development of nation-wide party alliances and nation-wide voter attachments. A first look at the data suggests that there is some relationship between party system nationalization (or its converse, party system territorialization) and the age of a democratic party system. Figure 1 presents a scatterplot of the relationship between the age of party system (on the x-axis) and party system nationalization (on the y-axis). Party system nationalization (on the y-axis) captures the extent to which parties in a country obtain votes across all electoral districts. High values indicate that the party system is nationalized meaning that most parties obtain the same proportion of votes in all electoral districts. Lower values indicate that the party system is localized or territorialized meaning that parties obtain a very uneven distribution of votes across electoral districts 3. Age of of parties (on the x-axis) measures the average age of the two main political 2 Some authors have argued that party systems do not become more stable over time, and that electorates do not become more partisan over time. According to Mainwaring and Zoco (2007) it is not the length of democracy but rather the timing of democracy that explains differences in electoral volatility. 3 The extreme case of territorialization would be one in which each electoral district has its own different set of parties. The extreme case of nationalization is one in which all districts have the same set of parties and parties obtain the same proportion of votes in each district. I will go into more detail on the measurement of this variable later on in the paper. The data comes from an original dataset that includes about 346 election years in 67 countries spanning the period from 1970 to 2009 and includes a healthy sample of both old and new democracies. See Table 5 4

5 parties in a country 4. The scatterplot suggests that countries with relatively older political parties have (on average) more nationalized party systems than countries with a younger party system; but the relationship is not perfect and among countries with young parties there is actually substantial variation: some young party systems are territorialized and some are nationalized. The rest of the paper focuses on institutions and more specifically electoral institutions to help explain some of this variation among younger party systems. Figure 1: Party System Nationalization and Age of Party System Party System Nationalization BU EL CR CR CR DR CZCZ HU SLSLSL CRCR CYHU AT AT AT CRO HUHU HU CH CZCZ GRGR CROGR EL TTO TW MAU CR AT AT AT DK AT CO BUCH BO BW BW CR DKDK GR GR GR MAU MAU PO PT SWE DR DRJAJAJAJA SA ICE DK DK ES CA ES ES DR ICE IRIR ITITLI ICE IR IRIR ICE IT BR BU BO AU AU DK CA GH ITIT JP JP IND JP JPJP LU LU LU MA MX JP MX NZNZ NZ NZ PTPTPT SP SWE SWE SWE NZ NO NO NO NO VZVZVZVZ TWNZ PERORO SA TTO RU TU SP LALA AR AU AU BO BW EC AU AUAU FIN DEDE DE SPSPSP SPSP MX RO TU MX SR TU ITIT SP UKUK JP FINFIN BR AU CA CA IND FIN BR BR FIN FIN FIN NZ RO FIN WDE RU SR WDE BRFR WDE WDE SK LI SK SK TTO TTO USUS UK TTO SP US US US US US US UK US US US USUSUS UKR US US US USUS BO JP ECJP CA JP SK BO SW CACA SW CA SK PH BR FR PH RU IN MX SW BR CH IN RUFR IN BH IN IN AR AR AR AR AR BE BE BE BE BEBE BE CO PH BEBE TH IN IN AR IN IN PH TH AR PH AR AR AR CO CO CO CO Age of Parties Electoral Incentives and Party System Nationalization The development of national parties and nationalized party systems is a matter of coordination of candidates and voters across electoral districts (Cox 1999; Hicken 2009; Chhibber and Kollman 2004). National parties form when candidates and voters from different districts coordinate under the same party labels. When the same set of parties competes in every electoral district and voters vote for these parties evenly across all districts then the outcome is a party system with nationally-oriented political parties, or what the literature describes as a nationalized party system. In contrast, when candidates from different electoral districts fail to adopt common party labels the result is a party system with locally oriented parties, or what the literature describes as a territorialized party system. Coordinating across electoral districts is a costly activity (Hicken 2009); it requires parties to have the necessary resources and organizational skills to reach out to a vast (and often diverse) set of constituencies. Given these costs the literature has explored several conditions under which we can for the countries and years included in the dataset. 4 I provide a more detailed explanation of the measurement of this variable in the empirical section of the paper. 5

6 expect candidates and parties to overcome these costs and form nation wide party platforms. For example, there has been a substantial amount of excellent research exploring how institutions such as presidentialism, decentralization and bicameralism affect the incentives of voters and candidates to coordinate successfully across the territory of a country (Chhibber and Kollman 2004; Hicken 2009; Cox 1997) 5. Less attention has been given to electoral institutions and the role they might play in promoting or inhibiting cross-district coordination. Majoritarian Electoral Rules and Cross-District Coordination A common starting point in the literature on electoral systems is that majoritarian systems tend to localize (or personalize) politics more than proportional systems. This general assumption however points us in two very different directions regarding the development of national parties. On the one hand, the localization of politics is linked to parties that focus on narrow constituencies and that are unable (or unwilling) to form broad-based national coalitions; the presence of such parties should be correlated with territorialized party systems. On the other hand, the localization of politics can imply greater party flexibility and adaptability which might in turn allow parties to better compete across a large number of diverse districts. In this case, the localization of politics should be correlated with more nationalized parties and party systems. I posit that these two (opposing) logics operate simultaneously. Majoritarian electoral systems make coordination across districts difficult for two reasons. First, in majoritarian electoral systems plurality is required to win the seat in a particular district therefore it is likely that parties may avoid spending the resources (good candidates, costs, and effort) to compete where they have little chance of winning (Morgenstern, Swindle and Castagnola 2009, 1327). Second, majoritarian electoral systems are associated with a greater (and more diverse) number of small electoral districts. A large number of districts means greater differentiation among districts and more electoral boundaries. These factors make the coordination of campaigns more difficult (Morgenstern, Swindle and Castagnola 2009, 1327) and increase informational costs, since parties need to determine which candidates to field in a large number of diverse districts. In contrast, under proportional representation, parties should have a greater incentive to field candidates in all districts because the more proportional distribution of seats to votes provides them with an easy electoral benefit. In other words, despite the costs inherent in coordination, parties still have an incentive to field candidates in districts that are not their strongholds because they know they will at least obtain some seats. In addition, proportional systems tend to have fewer and less differentiated districts, making it easier for parties to pick strong candidates for each district 5 For example, Chhibber and Kollman (2004) have argued that decentralization is one such important factor. If policy in a country is overwhelmingly made at the national level candidates and voters have more incentives to form national coalitions that will enable them to capture the national government. Therefore as countries centralize resources and power at the national level we should expect to see a corresponding nationalization of their party system. In contrast when power and resources lie predominantly at the sub-national level (i.e. in highly decentralized countries) the control of the national government becomes less important as a means to influence policy and thus candidates and voters have fewer incentives to unite under national party labels. Building on this work, Hicken (2009) argues that the horizontal distribution of power also matters to explain the incentives for coordination. More concentrated power at the national level should increase the incentives to coordinate across districts and form nationwide parties. Finally, regime type also seems to matter in explaining cross-district coordination. The common argument is that presidential systems lead to more nationalized party systems because presidents are often elected by plurality and parties that nominate presidential candidates for presidential office need to gain votes from a large portion of the national electorate (Samuels 2002). In contrast, in parliamentary systems parties can obtain representation at the national level without being the largest party in the country and without having to appeal to a broad sector of the electorate. This general argument is echoed by Cox (1997) in his book Making Votes Count: Strategic Coordination in the World s Electoral Systems. 6

7 and to coordinate campaigns across these districts. In sum, I posit that proportional representation systems create incentives for parties to spread nationally whereas majoritarian electoral systems increase the costs of coordination across districts in turn increasing the difficulty of developing national party organizations. This logic leads to the first hypothesis about the effect of electoral rules on nationalization: Hypothesis 1: Majoritarian electoral systems have a negative effect on party system nationalization whereas non-majoritarian electoral systems (i.e. PR and mixed systems) have a positive effect on party system nationalization. The Role of Candidate Autonomy: A Countervailing Effect The effect of majoritarian electoral systems on the coordination of candidates across districts has an additional layer of complexity. As explained above, parties in majoritarian electoral systems will have less motivation to coordinate across districts given the incentives generated by plurality rule and the number (and nature) of districts in these types of electoral systems. However, in order to pursue power and effectively implement their policy programs, parties in majoritarian systems need to win a considerable number of districts because they can only win one seat per district. In other words, parties in a majoritarian system have an incentive to spread nationally even if nationalization is costly. One way to spread nationally under majoritarian rule is to build decentralized party organizations that have weak party control over their candidates. Decentralized party organizations is key to cross-district coordination. Parties that allow for significant candidate autonomy within the party allow for greater ideological heterogeneity, and this heterogeneity facilitates reaching out to a (potentially) diverse set of districts in the country. As a party can effectively run in a larger number of districts that party will be better positioned to expand nationally. Furthermore, allowing greater candidate autonomy solves the informational problems associated with choosing strong candidates for a diverse group of districts by permitting a more efficient (bottom up) selection of viable candidates. In contrast, parties that do not allow for significant candidate autonomy within the party should have a harder time catering to a large and heterogeneous set of districts. Parties in countries with majoritarian electoral systems tend to develop party organizations with greater candidate autonomy precisely as a strategy to reduce the costs of spreading nationally. Therefore I argue that majoritarian electoral systems, despite their tendency to produce higher costs of coordination, are also associated with localized party organizations that can actually help overcome these costs of coordination. In sum, there is an indirect positive effect of majoritarian system on party system nationalization: majoritarian systems tend to produce parties with strong candidate autonomy, which in turn helps parties spread nationally. In contrast, proportional electoral systems tend to encourage parties with weak candidate autonomy, which can actually inhibit their ability to spread nationally. Hypothesis 2: Majoritarian systems have an indirect positive effect on party system nationalization. In other words, majoritarian systems should be more likely to produce parties allowing strong candidate autonomy, and parties with strong candidate autonomy should facilitate the development of nationalized party systems. In sum, I claim that electoral institutions have two opposing effects on parties incentives to coordinate across electoral districts. On the one hand, plurality rule associated with majoritarian electoral systems tend to decrease parties incentives to form party platforms across many diverse 7

8 electoral districts. On the other hand, majoritarian electoral systems tend to be associated with weak party control over individual candidates, which can reduce the costs for co-ordination across electoral districts and facilitate nationalization. These two effects are summarized in Figure 2. Figure 2: Effects of Majoritarian Systems on Party System Nationalization Candidate Autonomy + + Majoritarian Electoral System - Party System Nationalization Majoritarian Electoral System - Party System Nationalization (a) Direct Effect (b) Mediated Effect Electoral Incentives and the Age of the Party System My last claim has to do with how electoral incentives operate differently in different democracies. Coordinating across districts is a costly activity (Hicken 2004), but I argue that it is especially costly for recently created parties. In new democracies parties have not had the time to developed the necessary resources, experience and voter attachments to successfully build nation-wide party platforms, and thus the task of initial coordination is daunting. It is precisely in this context that institutional incentives and in this case electoral incentives that either reduce or increase their costs of coordination should be particularly important. In contrast, old political parties are organizations that have already developed the necessary experience and attachments to build crossdistrict alliances and have been receiving the benefits of being nation-wide party organization for some time. These parties should be less susceptible to incentives that lower the costs of coordination even further or that increase them. I thus argue that the effects of electoral rules that I have described in this section should be particularly present in countries with young party systems. This leads to the following conditional hypotheses: Hypothesis 3: The direct negative effect of majoritarian electoral systems on party system nationalization should be larger in young (rather than old) party systems. Hypothesis 4: The indirect positive effect of majoritarian systems on the development of nationalized party systems should be larger in young (rather than old) party systems. Data and Empirical Strategy In order to test the propositions presented in the previous section I use an original data set of party system nationalization that uses a large and representative sample of democracies. The universe of cases for my data set is all countries that have been democratic for at least two legislative election 8

9 Candidate Autonomy i, t = α 0 + α 1 Majoritarian i, t + α 2 X i, t-5 + α 3 Z i, t + α 4 P i + ε i,t (3) periods starting in 1970 (or anytime after that if their transition to democracy happens at a later year) and for which there is available district-level electoral data. Model 1 tests the first hypothesis (H1) on the direct effect of majoritarian electoral systems on party system nationalization and Model 2 tests the third hypothesis (H3) regarding the interaction of this direct effect with the age of the party system (H3). In both models the dependent variable is party system nationalization, which varies across countries (i) and across election years within each country (t). β 0 is a constant term. The main independent and control variables vary across countries and time (and in some cases only across countries). When there is sufficiently over-time variation in the covariates I lag them 5 years with respect to election year t in order to deal with issues of endogeneity. Given the nature of the data (time-series cross-sectional data) I cannot assume that observations within each country are independent. I thus use OLS regression with robust standard errors that are clustered by country in all regression analyses (based on Harbers (2010) and Hicken (2009), which is based on Beck and Katz (1995), Franzese (2006) and Golder (2006)). Party System Nationalization i, t = β 0 + β 1 Age of Party System i, t + β 2 Majoritarian i, t + β 3 X i, t-5 + β 4 Z i, t + β 5 P i + ε i,t (1) Party System Nationalization i, t = β 0 + β 1 Age of Party System i, t + β 2 Majoritarian i, t + β 3 Majoritarian x Age of Party System i, t + β 4 X i, t-5 + β 5 Z i, t + β 6 P i + ε i,t (2) The following two models (Model 3 and Model 4) test the indirect effect of majoritarian electoral systems on party system nationalization as it is mediated by levels of candidate autonomy within parties (hypothesis 3). Following Baron and Kenny (1986) and Judd and Kenny (1981) I test this indirect effect with the following two regression models: Party System Nationalization i, t = ω 0 + ω 1 Age of Party System i, t + ω 2 Majoritarian i, t + ω 3 Strong Candidate Autonomy i, t + ω 4 X i, t-5 + ω 5 Z i, t + ω 6 P i + ε i,t (4) Model 3 tests the path α 1 in Figure 3 (page 10) where candidate autonomy is the dependent variable. Model 4 tests path ω 2 and path ω 3 in Figure 3. To assert that there is an indirect effect of majoritarian on party system nationalization through candidate autonomy we should observe the following three things. First, in Model 3 the coefficient on majoritarian (α 1 ) should show a positive and significant correlation with the mediating factor candidate autonomy. Second, in Model 4 candidate autonomy (coefficient ω 3 ) should be positively and significantly correlated with party system nationalization. And third, the negative effect of majoritarian (coefficient ω 2 ) in Model 4 should disappear or should be smaller than the original negative effect of majoritarian 9

10 Figure 3: Indirect Effects of Majoritarian Systems on Nationalization Candidate Autonomy + ω3 + α1 Majoritarian Electoral System ω2 - Party System Nationalization (coefficient β2) in Model 1, where the mediator candidate autonomy is not controlled for 6. Finally I test for hypothesis 4 (H4) by dividing the sample between old and new party systems (instead of suing an interactive model). Dependent Variable: Party System Nationalization If we think of a country as divided up into electoral districts, a party system is nationalized if all electoral districts have the same set of parties. When this is the case the district-level party systems are the same in each district and the same at the national level. In contrast, when the party system is completely localized, each electoral district will have its own set of parties. This means that each district-level party system will be different from each other and different from the national level party system. In sum, the differences between the local and national party systems give us an intuition of the extent of nationalization. Based on this logic Chhibber, Shastri and Sisson (2004) and Chhibber and Kollman (2004) use a measure of the extent of coordination across electoral districts in a country which is the difference between the effective number of parties at the national level (ENPn) and the effective number of parties at the district level averaged across districts (ENPd): 7 ENP Difference = ENPn ENP d (7) According to this measure, if the effective number of parties at the district-level is similar to 6 For a more detailed explanation of the conditions for testing indirect effects see Baron and Kenny (1986). 7 The measure of the effective number of parties is based on Laakso and Taagepera (1979) and calculated as follows: ENPn = i=1 ( v i V )2 (5) d=1 ( 1 i=1 v 2 ij ) ENP d = (6) D In equation 5, v is the number of votes obtained by party i; V is the total number of votes cast at the national level; so the effective number of parties at the national level is the sum of squared vote shares. In equation 6, D is the number of districts and v ij is the vote share of each party in each district. Note that this measure weights parties according to their size. 10

11 the effective number of parties at the national level (i.e. if the ENP difference is small or close to 0) we can assume that the same set of parties is present in all electoral districts. And this implies that the party system is nationalized. However, if the difference is large it means that coordination between districts is not good and creates an inflated party system at the national level. In this case the party system is localized. This simple difference is then transformed into a percentage measure of how much larger the national party system is than the average district-level party system (Hicken 2009, 22), which several scholars refer to as inflation. 8 The measure of inflation is presented in equation 8 on page 11. I invert the inflation measure to obtain my measure of party system nationalization which ranges from 0 (a perfectly territorialized party system) to 1 (a perfectly nationalized party system) 9. Party System Nationalization = 1 [Inflation] = 1 [ ENP n ENP d ] (8) ENPn The data collected to construct the measure of party system nationalization consist of electoral results for each party in each electoral district at each national legislative election for every democratic country. Until recently this type of data was practically inexistent. Today, two data collection projects are underway to collect district-level electoral results for a comprehensive set of countries: (Kollman, Hicken, Caramani and Backer January 19, 2012) and CLE (Brancati 2007). In addition I have used country sources and other databases on electoral results (see Table 6 for all sources on electoral data by country). Key Independent Variables There are three key independent variables of interest: majoritarian electoral system, candidate autonomy and age of party system. Majoritarian Electoral System I classify electoral systems into majoritarian (SMDP systems) and not majoritarian, which include both proportional representation and mixed systems. The variable I create is thus dichotomous and takes the value of 1 for majoritarian electoral systems and 0 otherwise. I use Golder (2005) s Democratic Electoral Systems Around the World database to code electoral systems in addition to IPU (2012) and Political Database of the Americas. Candidate Autonomy I rely on Johnson and Wallack (2006) s dataset Electoral Systems and the Personal Vote to measure levels of candidate autonomy within parties. This dataset builds on Carey and Shugart (1995) s work on the incentives that electoral systems provide to cultivate a personal vote and provides measures to distinguish between systems where politicians careers 8 According to Hicken (2009), if inflation is 0.1 it suggests that only 10 percent of the size of the national party system can be attributed to different parties garnering votes in different parts [districts] of the country (poor aggregation), with the other 90 percent due to the average number of parties at the district level. In short, a measurement of 0.1 indicates that aggregation is very good the same parties are generally the frontrunners in most districts nationwide. On the other hand, [if the measure is 0.6 for example] we know that poor aggregation deserves most of the credit for producing a large number of parties nationally while intra-district coordination can only account for 40 percent of the national party system s size (Hicken 2009, 62). 9 This is a widely used measure of the degree of party system nationalization, but not the only one. For example, some scholars measure the extent to which each party in a country is nationalized and then take an average to have a party system level measure. Jones and Mainwaring (2003) first measure the dispersion of each party s votes across different regions (their measure of dispersion is the Gini index) to create what they call the Party Nationalization Score (PNS). And then they aggregate these individual party scores to create a party system measure. [Cite Morgenstern measures as well]. See Hicken (2009) or Leiras (2006) for a good review on the different ways scholars have calculated this inflation measure and the logic behind these different measures. 11

12 depend most on party fortunes to particularistic systems, where candidates must focus on narrow geographic constituencies (Wallack, Gaviria, Panizza and Stein 2003, 136) 10. I focus on one of the variables in this dataset called ballot, which refers to the party leaders control over candidate s access to and rank on ballots 11. The measure takes the following values: 0 means that parties control both access and position on the ballot; 1 means that parties control access to the ballot but not order; and 2 means that individuals can obtain access to the ballot without the party s blessing. This measure provides a good approximation of the the extent to which parties in a country are more likely to have weak or strong control over their candidates ranging from 0 (low levels of candidate autonomy) to 2 (very high levels of candidate autonomy). It is important to note that this measure is coded based on characteristics of the electoral system, which means that my measure of the electoral system is necessarily correlated with the measure of candidate autonomy. This could be problematic since I am introducing both variables in my regression analysis. I address this potential problem in two ways. First I show that although my measure of candidate autonomy is correlated with my measure of the electoral system, the correlation is not sufficiently high to skew the results (i.e. multicollinearity is not a problem). Second, I show in the results section that the variables majoritarian electoral system and candidate autonomy are indeed positively correlated, but they actually have an opposite effect on the dependent variable party system nationalization, which indicates that they are capturing very different concepts. Age of Party System In order to measure the length of a party system s life I follow Hicken (2009), Simmons (2008) and Roberts and Wibbels (1999, 581) and use data from the World Bank s Database of Political Institutions (Beck, Clarke, Groff, Keefer and Walsh 2001). The database covers the period for well over 150 countries of all levels of political and economic development. A party s age is coded according to the number of years since the party was founded under its current name. The investigators are aware of purely cosmetic name changes in which the party s name changes, but the party leaders, platform, and constituency remain the same and they do not code an obviously cosmetic change as the date of a new party s formation (Simmons 2008, 94). Based on this data I construct a measure that averages the age of the governing party and the age of the main opposition party in each country at each election year. 12 The measure ranges from X to X. Because the measure has a fairly right- skewed distribution I log this measure. Figure 5 shows the distribution of age of party system unlogged (top) and the distribution of age of party system logged (bottom). Logging creates a more normal distribution, which is better suited for the assumptions of OLS regression This dataset is an update of Wallack et al. (2003) s Database of Particularism, and has evolved to cover more aspects of electoral systems as well, providing additional information on more complex systems with multiple tiers and runoffs. It also covers more countries over more time periods (180 countries between 1978 and 2005). Both data sets as well as supplementary material may be found at (Johnson and Wallack 2006, 1). 11 In order to account for the complexities of mixed-member systems (which combine two tiers of membership), Johnson and Wallack code ballot separately for single-member district tiers and for multi-member district tiers, and then they calculate a weighted average measure of these two tiers (Johnson and Wallack 2006, 10). 12 I do not use the age of democracy because what is important in my argument is not so much whether a democracy is old or new, but rather the extent to which its party system is old or new. 13 In logging age of party system I follow what other scholars have done. See Simmons (2008, 94) and Hicken (2009). 12

13 Controls In addition to these key independent variables, I include several control variables to account for competing hypotheses. I include controls for fiscal and political decentralization, regime type, territorial concentration of religion, language and income, and timing of democratization. The sources for the key independent variables and for these control variables are presented in Table 8 on page 28. Descriptive statistics for all variables in the models are presented in Table 7 on page 26. In what follows I review the reasoning behind their inclusion and their coding. Political Decentralization. As explained earlier both political and fiscal decentralization have been associated with party territorialization (Chhibber and Kollman 2004; Hicken 2009; Harbers 2010). Hooghe, Marks and Schakel (2010) have developed a coding scheme to measure the extent to which a region is endowed with an independent legislature and executive: 0=no regional assembly; 1=an indirectly elected regional assembly; 2=a directly elected assembly; 0=the regional executive is appointed by central government ; 1=dual executives appointed by central government and the regional assembly; 2= the regional executive is appointed by a regional assembly or directly elected. They then add up the scores for the assembly and for the executive for each country and create a categorical variable that ranges from 0 (no independent legislature or executive) to 4 (completely independent legislature and executive). Countries that score a 4 have subnational legislative and executive bodies that are independently elected and that are independent from the central government. These are purely federal countries. Countries scoring 0 have no subnational level of government or have a subnational level where officials are not independently elected. These are unitary countries. Hooghe, Marks and Schakel (2010) have coded this variable annually for 42 OECD democracies between 1950 and I use their data for OECD countries and code the remaining cases (and time periods) myself using the same coding scheme. I create a variable which I call political decentralization that takes a value of 1 if country is politically decentralized and 0 if the country is politically centralized. This variable is lagged 5 years. Fiscal Decentralization. I use a measure that is commonly employed in the literature: subnational revenues as a % of total revenues in a country. I also perform robustness tests with a related measure that is also commonly used in the literature: subnational revenues as a % of GDP. The data for both measures of fiscal decentralization comes from the World Bank Fiscal Indicators. 14 The measures are reported in percentages but I convert them to proportions. This variable is lagged 5 years. Regime Type. As explained earlier, presidential systems tend to be associated with more party system nationalization. The data for this variable is taken from Samuels and Shugart (2010) s recently published book Presidents, Parties, and Prime Ministers: How the Separation of powers Affects Party Organization and Behavior. I create a variable called presidential system where 1=presidential systems and 0=parliamentary (and semi-presidential) systems. Geographic Concentration of Heterogeneity. Scholars have often assumed that ethnic, religious and/or linguistic diversity tends to be geographically concentrated, and thus have used fractionalization measures as proxies for geographically concentrated social diversity, but recent work by Selway (2011) suggests that this assumption is inaccurate. Some countries might have many different linguistic groups (in which case they might have a high linguistic fractionalization index) but these groups might not be geographically concentrated, whereas other countries might have 14 http : //www1.worldbank.org/publicsector/decentralization/fiscalindicators.htm 13

14 just two linguistic groups (low linguistic fractionalization index) but these linguistic groups might each be concentrated in a different region of the country. In a recent article The Measurement of Cross-cutting Cleavages and Other Multidimensional Cleavage Structures Selway (2011) introduces a new data set (CIMMSS) that includes several measures of geographically concentrated social diversity. These measures build on the concept of cross-cuttingness which Selway defines as the extent to which groups on a first cleavage are identically distributed amongst groups on a second cleavage. 15 For example, one such measure is religion-geography cross-cuttingness which gets at the degree to which religion and territory overlap in a country. Low values of religion and geography cross-cuttingness mean that individuals of religion A mostly live in region C whereas individuals of religion B mostly live in region D. In other words, religion and geography reinforce each other. In contrast, high values of religion-geography cross-cuttingness mean that individuals of religion A are equally divided in terms of the region where they live, and so are individuals of religion B. This means that religion and geography cross-cut. Based on Selway s data I construct an index of geographic concentration of religion, language and income that I call GC Heterogeneity (i.e. Geographic Concentration of Heterogeneity). This variable ranges from 0.08 (low geographic concentration of heterogeneity) to 0.6 (high geographic concentration of heterogeneity). I log the variable because its distribution is highly skewed. Timing of Democratization. Several scholars of party systems (Mainwaring 1999; Mainwaring and Torcal 2005; Hicken and Kuhonta 2011) have argued that strongly institutionalized party systems, and party systems with low levels of electoral volatility are more likely in democracies that were inaugurated earlier (Mainwaring and Zoco 2007). This is called the timing or period effect (Hicken and Kuhonta 2011). The argument is that parties in post-1978 democracies (i.e. third wave democracies) are less likely to be institutionalized than parties in pre-1978 democracies regardless of the duration of democracy. In early democracies, political parties played a lead role as mobilizing institution by, for example, incorporating new citizens into the political system and pushing for an expansion of suffrage and other rights for those citizens (Colomer, 2001; Mainwaring Zoco, 2007). This forged strong links between parties and the citizens they helped to mobilize. By contrast, in later democracies, the switch to competitive elections and new party formation was preceded by, or occurred in conjunction with, the adoption of universal suffrage. As a result, the kinds of links and networks that characterized early democratizers never developed. What is more, with the advent of mass communication, specifically television, parties and candidates had a means of mobilizing large numbers of voters without the costly investment in party organization or grassroots networks (Hicken and Kuhonta 2011, 6). In order to control for this alternative explanation I include a dichotomous variable coded 1 if the country is part of the third wave of democratization and 0 otherwise. Results Table 1 on page 15 presents the base model from Model 1 on page 9 testing the first hypothesis. The results show that age of party system is positively correlated with party system nationalization: countries with older political parties are more likely to have more nationalized party systems whereas countries with younger political parties are more likely to have territorialized party systems. This is not surprising given our initial expectation that time should matter in the development of nation-wide party alliances. More importantly the effect of majoritarian electoral system is negatively related to party system nationalization (as expected) yet not statistically sig

15 Table 1: Base Model of Party System Nationalization DV: Party System Nationalization Base Model Age of Party System (log).061*** (.02) Majoritarian Electoral System.073 (.05) Presidential.127** (.06) GC of Heterogeneity (log).124 (.08) Political Decentralization (lag).124** (.06) Fiscal Decentralization (lag).000 (.00) Third Wave Democracy.036 (.07) Constant.479*** (.17) Adj. R No. of cases 176 No. of country clusters 46 Sig:* p<0.10, ** p<0.05, *** p<0.01 Robust Standard Errors Clustered by Country nificant. The lack of statistical significance is not surprising since the direct effect of majoritarian rule should be conditional on the age of the party system (hypothesis 3). Table 2 on page 16 tests this conditional effect. The first two regression models take the base model (in Table 1) and run it against a divided sample: countries with relatively young political parties (0 to 40 years) and countries with relatively old parties (more than 40 years). This division is somewhat arbitrary but gives us a good first cut at testing the conditional effect of the electoral system between party systems that have been alive for just one generation and party systems that have been around for more than one generation. The key thing to observe is that the coefficient on majoritarian electoral system is statistically significant among the subsample of countries with young party systems, whereas it is not significant among the subsample of countries with old party systems. Furthermore, the effect of majoritarian electoral system among young party systems is quite large and in the expected negative direction. This is consistent with the story that majoritarian electoral systems increase the costs of cross-district coordination and make it less likely for parties to form nation-wide party alliances. This lends credibility to the interactive relationship presented in hypothesis 3. The correct way of testing this hypothesis, however, is to create an interaction term between age of party system and majoritarian electoral system. In addition, this solves the problem of splitting the sample arbitrarily. The results of this interactive model are presented in the third regression model in Table 2. Unfortunately, interactive terms (and their significance) are difficult to interpret solely through an analysis of the coefficients in this table. What is important in interaction terms is to gauge the significance of the marginal effect of one of the interacted variables over the range of the other 15

16 Table 2: Splitting Sample and Interactive Model Young Old Interactive DV: Party System Nationalization Party Systems Party Systems Model Majoritarian Electoral System.300** ** (.11) (.03) (.17) Age of Party System (log).032 (.02) Age of Party System (log) x Majoritarian.083** (.04) Presidential.201*** ** (.07) (.03) (.05) GC of Heterogeneity (log) **.105 (.10) (.07) (.08) Political Decentralization (lag).146** * (.06) (.06) (.06) Fiscal Decentralization (lag) (.01) (.01) (.00) Third Wave Democracy (.07) (.07) (.07) Constant.968***.540***.636*** (.23) (.17) (.20) Adj. R No. of cases No. of country clusters Sig:* p<0.10, ** p<0.05, *** p<0.01 Robust Standard Errors Clustered by Country 16

17 variables and vice-versa. Following Kam and Franzese (2007) and Brambor, Clark and Golder (2006) I calculate and draw the relevant marginal effects. This allows me to better interpret the magnitude, direction and significance of the coefficient of the interaction term. Based on the interactive model in Table 2, Figure 4 shows the marginal effect of majoritarian electoral system on party system nationalization as the age of the party system changes. The marginal effect is represented by the thick black line. If the marginal effect line is above 0 then the marginal effect is positive; if it is below 0 then the marginal effect is negative. The dotted bands represent the 95% confidence intervals. The marginal effect is statistically significant if neither of the bands crosses the 0 line. This graph tells us that having a majoritarian electoral system always has a negative and significant effect on party system nationalization, but that the magnitude of this effect is large and significant among young party systems and then decreases as party systems age. Among old party systems the effect is small and non-significant. This provides support for hypothesis 3 and suggests that electoral systems matter early on in the development of party systems 16. Figure 4: Marginal Effect of Majoritarian Systems on Party System Nationalization Marginal Effect of Majoritarian Electoral System Age of Parties (log) I turn now to test the indirect effect of majoritarian electoral systems on party system nationalization (hypotheses 2 and 4). Table 3 test the indirect effect in young party systems whereas Table 4 tests the indirect effect among old party system. Because we expect results to be significant only among countries with younger party systems we proceed to analyze Table 3. As a reminder to the reader, if the hypothesized indirect effect is correct I expect to observe the following: a) majoritarian electoral systems should encourage candidate autonomy; b) candidate autonomy should have a 16 It is also worth noting (although this has not been theorized in this paper) that other institutional variables (i.e. fiscal decentralization and political decentralization) also matter more among young party systems than old party systems. This suggests that institutional incentives are particularly important in the early stages of development of party systems, more than later on. This is an avenue worth exploring further. 17