FUNDAÇÃO GETULIO VARGAS ESCOLA DE ADMINISTRAÇÃO DE EMPRESAS DE SÃO PAULO JULIA MANTOVANI GUERREIRO

FUNDAÇÃO GETULIO VARGAS ESCOLA DE ADMINISTRAÇÃO DE EMPRESAS DE SÃO PAULO JULIA MANTOVANI GUERREIRO ELECTORAL VOLATILITY AND STRATEGY OF FEDERAL CONGRESSMEN SÃO PAULO 2017

JULIA MANTOVANI GUERREIRO ELECTORAL VOLATILITY AND STRATEGY OF FEDERAL CONGRESSMEN Tese apresentada à Escola de Administração de Empresas de São Paulo da Fundação Getulio Vargas como requisito para obtenção do título de Doutor em Administração Pública e Governo Linha de Pesquisa: Política e Economia do Setor Público Orientador: Prof. Dr. George Avelino Filho SÃO PAULO 2017

Guerreiro, Julia Mantovani. Electoral volatility and strategy of Federal Congressmen / Julia Mantovani Guerreiro. - 2017. 73 f. Orientador: George Avelino Filho Tese (doutorado) - Escola de Administração de Empresas de São Paulo. 1. Brasil. Congresso. Câmara dos Deputados - Eleições. 2. Governo representativo e representação. 3. Tática política. I. Avelino Filho, George. II. Tese (doutorado) - Escola de Administração de Empresas de São Paulo. III. Título. CDU 324(81)

JULIA MANTOVANI GUERREIRO ELECTORAL VOLATILITY AND STRATEGY OF FEDERAL CONGRESSMEN Tese apresentada à Escola de Administração de Empresas de São Paulo da Fundação Getulio Vargas como requisito para obtenção do título de Doutor em Administração Pública e Governo Linha de Pesquisa: Política e Economia do Setor Público Orientador: Prof. Dr. George Avelino Filho Data de aprovação: / / Banca examinadora: Prof. Dr. George Avelino Filho (Orientador) FGV - EAESP Prof. Dr. Ciro Biderman FGV - EESP Prof. Dr. Scott William Desposato UCSD Department of Political Science Prof. Dr. Jairo Cesar Marconi Nicolau UFRJ DCP Prof. Dr. Fernando de Magalhães Papaterra Limongi USP DCP

Para Thomas e Nicolas

AGRADECIMENTOS Esta tese não teria sido construída sem o apoio de diversos colegas e familiares. Em primeiro lugar, sou extremamente grata ao meu orientador, professor George Avelino, por ter me sugerido uma linha de pesquisa na área da Ciência Política, até então nova para mim. Também agradeço a sua disponibilidade para as constantes conversas e revisões de textos. Sem isso, a tese não existiria. Em segundo lugar, agradeço ao meu marido, Bruno, pelo incentivo ao longo dos últimos doze anos, e principalmente por ter acreditado que eu conseguiria realizar um doutorado junto com a construção da nossa família. Também agradeço aos meus pais, Marilisa e Carlos, pelo apoio incondicional. Em terceiro lugar, agradeço a todos os professores, colegas e amigos que estiveram comigo nesta trajetória. Ao Arnaldo, por me acolher nas suas pesquisas e iluminar o meu caminho para o período sanduíche do doutorado, além de estar sempre disposto a discutir e comentar a minha pesquisa; ao Ben, por ter me recebido com tanto profissionalismo e cuidado no MIT; ao Arthur, Ivan e Claudia, pelas inúmeras discussões sobre o sistema partidário brasileiro e por compartilhar dados; à Lara, pelas revisões e sugestões; ao Leo Barone, por sugestões no início da pesquisa; à Karina, pelas discussões sobre metodologia; à Daniela, Fred, Joice, Lorreine, Natália e Bruno, pelo companheirismo nas longas horas dispendidas no centro de estudos. Finalmente, agradeço às duas agências de fomento que me forneceram apoio financeiro ao longo deste projeto. Inicialmente, à CAPES e posteriormente, à FAPESP.

ABSTRACT This dissertation focuses on conciliating two apparently contradictory facts: a fall in electoral volatility and a high turnover rate in the Chamber of Deputies. We use the case of Brazil and explore the hypothesis that politicians will use the advantages resulting from the fall in electoral volatility rates in different ways, depending on the type of party to which they belong. It is assumed that the decrease in volatility is greater in more institutionalized parties and for the incumbents of these parties. Given that institutionalized parties are able to provide clearer information, which helps incumbents improve their electoral strategies, the reasons for their defeats whether due to intraparty or interparty competition will vary according to their party s level of institutionalization. Keywords: electoral volatility, electoral strategy, open-list proportional representation, legislative elections.

RESUMO Esta tese busca conciliar dois fatos aparentemente contraditórios: a queda na volatilidade eleitoral e a alta taxa de renovação na Câmara dos Deputados. Utilizamos o caso do Brasil e exploramos a possibilidade de que políticos irão se apropriar das vantagens resultantes de uma queda na volatilidade eleitoral de maneiras diferentes, dependendo do tipo de partido ao qual ele pertence. Assumimos que a queda na volatilidade é maior em partidos mais institucionalizados e para incumbentes destes partidos. Dado que partidos mais institucionalizados são capazes de prover informações mais claras para seus candidatos, o que ajuda incumbentes a melhorarem as suas estratégias eleitorais, as razões para as derrotas seja por conta da competição dentro da lista partidária ou por conta da competição entre partidos irão variar de acordo com o grau de institucionalização do partido. Palavras-chave: volatilidade eleitoral, estratégia eleitoral, representação proporcional de lista aberta, eleições legislativas.

List of Figures Figure 2.1. Electoral Volatility in Brazilian States, 1998-2014...13 Figure 2.2. Mean Electoral Volatility by State, 2006-2014...14 Figure 2.3. Mean Electoral Volatility by Period..15 Figure 2.4. Mean Party Volatility by State, 2006-2014...20 Figure 2.5. Mean Party Volatility by Period....20 Figure 2.6. Party Volatility by Party Institutionalization Level...21 Figure 2.7. Party Volatility by Party Institutionalization Level and Year...22 Figure 2.8. Party Volatility by Presence of Municipal Sections..23 Figure 3.1. Type of Defeats over Time....34 Figure 4.1. Kernel Density Estimation of Electoral Vulnerability..52

List of Tables Table 1.1. Reelection of Federal Deputies in Brazil....2 Table 1.2. Federal Deputies Running for Reelection (average for 2002-2014)......4 Table 2.1. Electoral Volatility in 2014, by State....16 Table 3.1. Reelection of Federal Deputies in Brazil.. 29 Table 3.2. Incumbent Deputies Not Reelected.. 30 Table 3.3. Future Candidacy and Success Rate of Incumbent Deputies Not Reelected 31 Table 3.4. Incumbent Deputies Not Reselected.32 Table 3.5. Federal Deputies Running for Reelection (average for 2002-2014).36 Table 3.6. Deputies from the Seven Largest Parties (in percent)...37 Table 3.7. Summary of Variables.......39 Table 3.8. Regression Results: Intraparty Volatility.. 42 Table 4.1. Measure of Intraparty and Interparty Volatility......50 Table 4.2. Correlation Coefficient of Volatility Measures.... 51 Table 4.3. Regression Results: Electoral Vulnerability..... 56

List of Equations Equation 2.1. The Pedersen Index of Electoral Volatility....8 Equation 2.2. Measure of Party Volatility... 19 Equation 3.1. Regression Specification 1.... 40 Equation 3.2. Regression Specification 2....41 Equation 4.1. Electoral Vulnerability......51 Equation 4.2. Regression Specification......55

ACRONYMS Brazilian Parties DEM Democratas PAN Partido dos Aposentados da Nação PDT Partido Democrático Trabalhista PEN Partido Ecológico Nacional PFL Partido da Frente Liberal PGT Partido Geral dos Trabalhadores PL Partido Liberal PMDB Partido do Movimento Democrático Brasileiro PMN Partido da Mobilização Nacional PP Partido Progressista PPL Partido Pátria Livre PRB Partido Republicano Brasileiro PRONA Partido de Reedificação da Ordem Nacional PROS Partido Republicano da Ordem Social PSB Partido Socialista Brasileiro PSC Partido Social Cristão PSD Partido Social Democrático PSDB Partido da Social Democracia Brasileira PSOL Partido Socialismo e Liberdade PST Partido Social Trabalhista PT Partido dos Trabalhadores PTB Partido Trabalhista Brasileiro PV Partido Verde SD Solidariedade Brazilian States AC Acre AL Alagoas AM Amazonas AP Amapá BA Bahia CE Ceará ES Espírito Santo GO Goiás MA Maranhão MG Minas Gerais MS Mato Grosso do Sul MT Mato Grosso PA Pará PB Paraíba

PE Pernambuco PI Piauí PR Paraná RJ Rio de Janeiro RN Rio Grande do Norte RO Rondônia RR Roraima RS Rio Grande do Sul SC Santa Catarina SE Sergipe SP São Paulo TO Tocantins

Table of Contents 1. Introduction... 1 2. Electoral Volatility... 6 2.1 Introduction... 6 2.2 Electoral Volatility and Party Organization... 8 2.3 The Brazilian Political System: Federalism, Multiparty Systems, and Party Organization... 10 2.4 Electoral Volatility in Brazilian Legislative Elections... 12 2.5 Measuring Party Volatility... 18 2.6 Data and Results... 21 2.7 Concluding Remarks... 23 3. Electoral Strategy of Federal Congressmen... 24 3.1 Introduction... 24 3.2 Careers of Federal Deputies... 26 3.3 The Brazilian Context... 28 3.4 Type of Defeats... 33 3.5 Empirical Analysis... 38 3.6 Results and Discussion... 41 4. Why is turnover so high in the Brazilian Chamber of Deputies?... 44 4.1 Introduction... 44 4.2 Nomination Strategies, Party Elites and Federal Deputies... 46 4.3 Electoral Vulnerability: Measure... 49 4.4 Electoral Vulnerability: Application... 54 4.5 Results and Discussion... 55 4.6 Concluding Remarks... 57 5. Conclusion... 58 APPENDIX A Number of Municipal Sections and Temporary Commissions for the 27 Traditional Parties in Brazil.... 60 APPENDIX B Number of Candidates in the Coalitions for the 2014 elections... 61 APPENDIX C Mean-Comparison Tests... 62 APPENDIX D Federal Deputies Running for Reelection, 2002-2014... 63 6. References... 67

1 1. Introduction 1 How do politicians define their electoral strategies? How does the political environment affect politicians strategies? Why do incumbent politicians lose an election even when the institutional setting is predictable? In this dissertation, we use the case of congressional elections in Brazil in an attempt to answer these questions. Congressional elections in Brazil are an interesting case to study because of the open-list proportional representation (OLPR) system used. Voters have the option of selecting a party number on the ballot (a party-label vote ), or any candidate on a party s list (a candidate vote ). As a result, candidates face competition from both within their party (intraparty competition) and other parties (interparty competition). Another important element of the Brazilian system for congressional elections is how electoral districts are organized and the impacts this has on electoral dynamics. Brazil is a federation. For congressional elections, the electoral districts are the states and, due to differences in magnitude, the number of deputies per state varies from eight to seventy. Moreover, in light of the federal arrangement, parties are decentralized and have autonomy to make decisions at the municipal and state levels. Consequently, the political parties and number of candidates involved in congressional elections differs from one state to the next. Finally, as a consequence of the OLPR votepooling mechanisms, the minimum number of votes a candidate requires to be elected varies across parties and states. This creates incentives for candidates to switch parties in order to take advantage of lower electoral quotas. In the early 1990s, 62% of candidates running for reelection won seats in the Brazilian Congress. This figure reached 67% in Argentina, 78% in Chile, and 94% in the United States (Morgenstern, 2002, 416). Lower reelection rates in Brazil, Argentina, and Chile in comparison to the United States could be attributed to differences in their institutional frameworks. A vast literature on reelection shows that incumbents in the United States have a substantial advantage over their challengers (Cox and Katz, 1996; Erikson, 1971, 2017; Mayhew, 1974a). Case studies of countries with higher institutional instability, on the other hand, have found that incumbents may engage in rent-seeking activities and when they do, 1 grant #2015/02045-7, São Paulo Research Foundation (FAPESP).

2 incumbency has a negative effect on subsequent electoral outcomes (Klasnja and Titiunik, 2017; Uppal, 2009). Despite Brazil s fragmented party system, electoral volatility declined significantly in the period from 1998 to 2010 (Bohn and Paiva, 2009; Braga, 2010; Mainwaring, Power, and Bizzarro, 2017; Mainwaring and Torcal, 2006; Peres, Ricci, and Rennó, 2011). A more stable institutional and electoral setting leads competition among parties and candidates to become more predictable, which increases accountability, helps candidates to develop more effective electoral strategies, and increases reelection rates. The decline in Brazilian electoral volatility, however, has not led to more predictable outcomes for congressional elections, as reelection rates for Congress are still low. More recently, reelection rates have remained fairly constant at around 71% 2 (see Table 1.1 below). Table 1.1. Reelection of Federal Deputies in Brazil Number of Federal Deputies in # of Deputies Deputies Reselected # of Deputies Deputies Reelected Deputies Reelected Election Congress Reselected (% of Total) Reelected (% of Reselected) (% of Total) 2002 513 369 72 263 71 51 2006 513 393 77 257 65 50 2010 513 379 74 282 74 55 2014 513 354 69 264 75 51 Average 513 374 73 267 71 52 Source: Author s calculations based on data from the Tribunal Superior Eleitoral (TSE), Câmara dos Deputados, and CEPESPDATA. Note: Reselected refers to incumbent deputies who ran for reelection, and reelected, the incumbent deputies who were reelected. If electoral volatility in Brazil has decreased, then why have incumbents electoral strategies not become more efficient over time? In other words, why are reelection rates still low in Brazil? Responsibility for the low reelection rates in the country has been traditionally attributed to the candidate. Some argue that OLPR systems encourage candidates to cultivate a personal vote at the expense of a party vote (Ames, 1995; Bergman, Shugart, and Watt, 2013; Carey and Shugart, 1995; Katz, 1986; Mainwaring, 1991). According to this literature, 2 In our calculations on reelection, we took into account only congressional representatives elected during the previous election, and not runners-up who might have substituted other members of Congress who vacate their position mid-term.

3 voters do not make choices based on ideological, organizational, or sociological reasons; candidates personal influence has a considerable impact on their decisions. The strategy of seeking a personal vote has been found to be more common among losers (Avelino, Biderman, and Silva, 2011). More recently, however, some authors have shown that parties play an important role in elections, even in the most candidate-centered systems such as OLPR (Carneiro and Almeida, 2008; Cheibub and Sin, 2014). This is due to political coordination within political parties. Following the same line of thought, Nicolau (2002) argued that despite the system s incentives to personalize the vote, voters in Brazil do not often formulate personal judgements of representatives. He attributed this to the fact that in many cases, the candidate that a voter selected on the ballot is not the one that gets elected, and many voters do not remember who they voted for during the previous congressional election. Desposato (2006) suggests that frequent party switching in Brazil is also evidence that parties are indeed important during elections, as the main motives for changing one s party affiliation are greater access to distributive resources, better electoral opportunities, and policy positions. Low reelection rates in the Brazilian Congress can, therefore, be attributed to the parties inability to encourage politicians to pursue a long-term career as a federal deputy. This is a typical collective action problem that political parties are supposed to solve (Aldrich, 1995). A good party is one that provides candidates information on their electoral future. We argue that more institutionalized parties are more organized and thus, better able to provide their candidates with quality information on their chances of winning. Also, due to their greater organizational capacity, these parties should be able to build a loyal constituency, and as a result, face lower levels of electoral volatility. This dissertation explores the hypothesis that politicians will use the advantages resulting from the fall in electoral volatility rates in different ways, depending on the type of party to which they belong. It is assumed that the decrease in volatility is greater in more institutionalized parties and for the incumbents of these parties. Given that institutionalized parties are able to provide clearer information, which helps incumbents improve their electoral strategies, the reasons for their defeats whether due to intraparty or interparty competition will vary according to their party s level of institutionalization.

4 Table 1.2 presents data for each Brazilian state on the percentage of federal deputies who were and were not reelected during the 2002-2014 period and the type of defeat. While this table will be analyzed in greater detail in chapter 3, for now, it is interesting to see the variation in reelection rates and types of defeats across states. Table 1.2. Federal Deputies Running for Reelection (average for 2002-2014) % Not Reelected State % Reelected* Intrapartisan Defeats** Interpartisan Defeats*** Acre 60 16 24 Amapá 41 18 40 Amazonas 78 0 22 Pará 82 4 14 Rondônia 79 8 13 Roraima 52 18 30 Tocantins 60 25 15 North 66 12 22 Alagoas 59 4 38 Bahia 73 15 12 Ceará 77 16 8 Maranhão 72 14 14 Paraíba 77 9 14 Pernambuco 73 12 14 Piauí 77 11 11 Rio Grande do Norte 68 9 23 Sergipe 52 18 30 Northeast 73 13 14 Espírito Santo 69 12 19 Minas Gerais 78 12 10 Rio de Janeiro 64 27 9 São Paulo 65 30 5 Southeast 69 23 8 Paraná 82 13 5 Rio Grande do Sul 80 14 6 Santa Catarina 67 27 6 South 78 16 6 Federal District 56 15 29 Goiás 81 14 4 Mato Grosso 71 15 13 Mato Grosso do Sul 71 13 16 Center-west 74 15 11 Brazil 71 17 11

5 Table 1.2 (continued) Source: Author s calculations based on data from the TSE, Câmara dos Deputados, and CEPESPDATA. Notes: * Reelected: number of incumbent deputies reelected as a percentage of the total number of incumbent deputies running for reelection. ** Intraparty defeats: when at least one nonincumbent was elected. ***Interparty defeats: 1) if no nonincumbent on the list of a party or coalition was elected; or 2) if the party or coalition did not elect any candidate. This dissertation is organized as follows: in Chapter 2, we discuss electoral volatility and political parties. Traditionally, volatility has been considered a phenomenon of party systems. We include political parties level of organization in the discussion in order to show that electoral support for highly organized parties tends to be less volatile. Lower volatility should reduce electoral uncertainties and, therefore, the chances of the incumbents of these parties losing an election should be lower when compared to incumbents of parties experiencing higher volatility. In Chapter 3, we take a closer look at the type of defeats. We adopt the definition of intraparty and interparty defeats commonly used in the literature and show that incumbents of more institutionalized parties lose more often due to interparty defeats than intraparty defeats. We attribute this to the fact that institutionalized parties are more organized and better able to coordinate their candidates. In Chapter 4, we return to the fundamental question raised in this dissertation: why do incumbent politicians lose elections in predictable institutional settings? First, as we argue in Chapter 2, voter support for more organized parties is less volatile, which favors the reelection of incumbents. Secondly, as shown in Chapter 3, more organized parties also respond to personal uncertainties within the party s list. They provide incumbents with resources to reduce their volatility on the party s list. Moreover, as the information shared with candidates is of better quality, incumbents who sense defeat tend to give up running for reelection more often in this type of party, which increases the party s rate of reelection. We propose a new measure of electoral vulnerability to capture both dimensions of electoral uncertainty and demonstrate that this measure is also related to the party s organizational structure.

6 2. Electoral Volatility 2.1 Introduction The study of electoral volatility was originally developed to apply to Western democracies. Since the pioneering work of Lipset and Rokkan (1967), most studies have focused on the stability of party choices and related this to certain social characteristics of voters. Back then, electoral volatility, defined as a change in a party s electoral support from one election to the next, was due to both the failure of individual parties to establish stable ties to voters and the party system s inability to produce stable patterns of party competition. In line with this definition, the widely-used measure of electoral volatility suggested by Pedersen (1979) aggregates changes in electoral support for individual parties into a systemic measure. As most studies of electoral volatility concentrated on Western democracies, where significant changes of electoral support for an individual party were atypical, they tended to focus on the aggregate measure of party systems (Lipset and Rokkan, 1967; Bartolini and Mair, 1990). Yet, an aggregate measure can hide significant differences among its components. Studying parties individually allows one to explore volatility within parties, without forfeiting the option of applying the results to the party system at a later time. More recently, several authors have attempted to apply the Pedersen methodology to other democracies, such as India (Nooruddin and Chhibber, 2008), Latin America (Roberts and Wibbels, 1999), Eastern Europe (Powell and Tucker, 2013; Sikk, 2005; Tavits, 2008b), and new democracies in general (Mainwaring and Zoco, 2007). Not surprisingly, these authors suggested disaggregating the original measure in order to deal with the new context. It is argued that volatility is the single most important indicator of presence or absence of party system stabilization (Sikk, 2005). The frequent emergence of new parties and their electoral success hinder the development of stable links between parties and voters, which, in turn, undermines party institutionalization (Tavits, 2008b). A system in which the entrance of new parties and the exit of established ones is common will obtain a high score on the Pedersen index. The significance of this may be wrongly interpreted, however, as it would not be an indication of the volatility of the electorate s behavior, but rather the system itself (Powell and Tucker, 2013; Sikk, 2005). As argued by Andrews and Bairett Jr. (2014):

7 a voter who switches support to another party is making a different kind of choice than a voter who must switch because the party voted for in the last election no longer exists. Most studies of volatility, motivated by a framework that emphasizes social cleavages and their ideological connection to parties mass-elite linkages have sought to explain the first kind of choice and have done so in terms of ideological and social economic characteristics of the voting population (Andrews and Bairett Jr., 2014, pp. 308-309). One of the objectives of this chapter is to identify the potential gains to be made by further disaggregating the measure of electoral volatility, breaking down and analyzing its components, and proposing a new measure of party volatility. The constant changes that occur in less stable party systems generate several inequalities within the system. Parties in these systems may differ in their ability to create stable bases of support, elect representatives, and have a significant number of affiliated members. There are also systems where new parties are constantly appearing, which have not had enough time or structure to create such a broad base. We argue that more organized parties are better able to cope with the arrival of newcomers on the political scene than less organized parties are. The solid base these parties built over time should help them sail through stormy periods. We use the 2014 legislative election in Brazil as a case where several new parties entered the electoral market and caused a significant shift in votes from one party to the other - that is, an increase in electoral volatility. According to our expectations, the entrance of new parties will affect less parties that are more organized. This chapter will present data on electoral volatility across Brazilian states and municipalities. The main advantage of studying states and holding the national context constant is to keep institutional arrangements constant across space. Moreover, Brazil offers an interesting case where electoral volatility had shown a tendency to decline since the return to democracy (Bohn and Paiva, 2009; Braga, 2010; Carneiro and Almeida, 2008; Peres, Ricci, and Rennó, 2011), but has increased more recently again. At the same time, the party system seemed to be becoming more institutionalized, up until the recent surge of new parties. This chapter is organized as follows: the next section discusses the link between electoral volatility and party organization, which is followed by a description of the Brazilian political system in section 2.3. Section 2.4 provides detailed information on electoral volatility in

8 Brazilian legislative elections. Section 2.5 proposes a measure of party electoral volatility. Section 2.6 discusses the results generated by the use of this measure, and the last section presents some concluding remarks. 2.2 Electoral Volatility and Party Organization Electoral volatility captures the stability of patterns of interparty competition. Over time, it captures how much voters move around in the party system (Pedersen, 1979). It also assumes that equal changes in electoral support will cancel each other out within the party system. A widely used measure of electoral volatility is the Pedersen index (Pedersen, 1983), which is calculated by dividing the sum of each party s absolute share of votes by two, to avoid double counting (Equation 2.1). Equation 2.1. The Pedersen Index of Electoral Volatility / EV t = ( &0+ vote &,( vote &,(*+ )/2 where: EV stands for Electoral Volatility; vote &,( is the share of votes of party i at time t ; vote &,(*+ is the share of votes of party i at time t-1. The index ranges from 0 (perfect stability) to 100 (perfect instability) and has been acclaimed as a comprehensive measure of the net systemic shift in levels of party support (Roberts and Wibbels, 1999, p. 580). However, the index has also been criticized for its drawbacks. One in particular is that it does not distinguish between the changes that stem from the dynamics between previously active parties and the impact of newcomers (Sikk, 2005). Powell and Tucker (2013) attempt to tackle this issue by splitting electoral volatility up between Type A volatility caused by the entry of new political parties and the exit of old ones from the system and Type B volatility caused by the reallocation of power between

9 political actors that are already a relevant part of the political process. They show that the Pedersen index is then equal to Type A plus Type B volatility. 3 Our contribution here is related less to discussing the Pedersen measure itself than to interpreting and deconstructing it. In our view, there are gains to be made from interpreting electoral volatility as a party-related phenomenon - that is, by breaking the index down into its components. 4 Electoral volatility is determined by the parties ability to establish stable relationships with groups of voters. Treating parties individually may bring new insights on system volatility. As argued earlier, we believe that the resilience of electoral parties can be explained by their level of organization, defined by the level of party decentralization. Unlike the traditional measures used in the analysis of parties and party systems, such as volatility, fragmentation 5, and disproportionality 6, there is no consensus on how to measure the strength of a party organization. A key challenge in defining measures of party organization is to find one that is accurate and reliable across parties, varies over time, and can also be assessed at local levels. The reasoning behind the link between the level of party organization and its electoral volatility is quite straightforward. More organized mass parties should have a large number of local party organizations, which are expected to build the local basis of electoral support by developing, for instance, clientelistic relations between voters and party members, particularly those who win public office (Hagopian, 1996). Moreover, in decentralized parties, these local organizations enjoy a certain degree of autonomy, which enables them to manage political positions at the local level and attract party affiliates. In less organized parties, candidates shape parties more than parties determine the type of candidates it will have because those eventually elected are the ones who will determine much of how the party works and what it does (Hazan and Rahat, 2006, p. 109). To keep their leadership position, these politicians will build a more centralized type of party 3 For a similar approach, see the within-system electoral volatility measure proposed by Mainwaring, Gervasoni, and España-Najera (2016). 4 Using electoral volatility as an index of party system institutionalization is questionable. Electoral volatility measures the ability of parties to build loyal followings and collectively structure the electorate. These are properties of parties, individually and collectively, rather than aspects of the party system - that is, parties and the ways in which they relate to each other (Wolinetz, 2005, p. 60). 5 Party system fragmentation considers the number of parties in the system and their relative size. 6 Party system disproportionality considers the chances of small parties winning representation.

10 organization and link their individual electoral performance more closely to that of their parties. In sum, more decentralized parties are expected to have a denser network of electoral support, which contributes to a more stable electoral performance and lower electoral volatility than more centralized parties do. This approach will be particularly helpful for studying a multiparty, federal system with high district magnitudes, as is the case of Brazil, which we describe in the next section. 2.3 The Brazilian Political System: Federalism, Multiparty Systems, and Party Organization Although the Brazilian electoral system functions according to a rare combination of electoral formulas, the stability of the election calendar has increased considerably, especially in comparison to the past. Elections for national and state-level officials are held concurrently every four years and municipal elections, two years later in the interval between them. The president and state governors are elected for four-year terms using a two-round majority method, with a runoff between the two top finalists if none of the candidates obtain more than 50% of votes in the first round. Members of both the federal Chamber of Deputies and state assemblies are also elected for four-year terms using a statewide open-list proportional representation system. Finally, senators are elected to serve eight-year terms in elections, which are held concurrently with presidential and state elections according to single-round, plurality system. Only part of the Senate is renewed each time: elections are staggered so that one-third of the seats of the house are up for election and then, four years later, elections are held for the remaining two-thirds. Reelection is limited to two consecutive terms for all executive positions (president, governors, and mayors). There are, however, no restrictions on reelection for legislative positions (senators, federal and state deputies, and municipal councilors). Municipal elections take place every four years. Mayors and city councilors are elected for four-year terms. In most municipalities, a single-round plurality system is used to elect the

11 mayor, and municipal councilors positions are filled through an open-list proportional system 7. Reelections to municipal offices follow the same rule as state and national elections. Contrary to most federations, which are organized via the relationship between a central government and states or provinces, Brazilian federalism has a three-level structure, as municipalities are also recognized as autonomous sources of political power. Due to this peculiarity of Brazilian federalism, mayors have considerable authority over policies within their jurisdiction, which increases their importance. The decentralization promoted by the 1988 Constitution delegated the implementation of important public policies such as education and health to municipalities. As a result, municipal offices are coveted positions and municipal arenas have increased their importance as loci of political competition. In view of this type of federal arrangement, it is not surprising that Brazilian political parties show low levels of centralization. Typically, parties have a decentralized structure based on municipal sections, which are in charge of the party s daily affairs, such as the recruitment of new membership. Although there have been occasional top-down interventions, municipal conventions have considerable autonomy over decisions on municipal campaigns, such as selecting candidates to run for municipal executive and legislative seats, defining policy proposals, strategies on coalition building, etc. Finally, municipal conventions also select their delegates to the party s state convention, where state-level decisions on the same issues are made and state delegates to the national party convention are chosen. According to most of the literature, in federalism, state-level politics matters: they provide voters and candidates incentives to link their efforts across districts within their states, and these linkages define the party s aggregate electoral performance. In Brazilian federalism, this means that municipalities must be taken into account when making efforts to form linkages; as explained, these autonomous local powers are paralleled by strong municipal party sections. The parties municipal sections have considerable autonomy from upper-level party leadership and are able to manage party affiliations and local political positions. Thus, what may distinguish Brazil from other countries is that the geographical space that defines the primary unit of party organization corresponds to a constitutionally delimited source of political authority, selected through competitive elections. 7 In municipalities with more than 200,000 registered voters, mayors are elected through the same two-round majority system with runoff elections that is used to elect the country s president and state governors.

12 With these characteristics of the Brazilian political system in mind, the next section will discuss electoral volatility in the elections for the Brazilian Chamber of Deputies. 2.4 Electoral Volatility in Brazilian Legislative Elections Some authors have pointed to the fact that Brazil has an extremely fragmented party system 8, with high electoral volatility (Ames, 1995; Mainwaring, 1999; Nohlen, 2005). However, other authors affirm that electoral volatility has declined significantly in Brazil since 1994 (Bohn and Paiva, 2009; Braga, 2010; Mainwaring, Power, and Bizzarro, 2017; Mainwaring and Torcal, 2006; Peres, Ricci, and Rennó, 2011). Both Braga (2010) and Mainwaring, Power, and Bizzarro (2017) use aggregate data on national votes to calculate electoral volatility for the Chamber of Deputies. They agree that the Brazilian party system is becoming more viable and that its level of institutionalization is now closer to that of traditional European democracies. This decrease in volatility is important, as it makes electoral competition among parties more predictable and increases accountability. The greater predictability makes it easier for candidates to develop more effective campaign strategies. Peres, Ricci, and Rennó (2011) argue that Brazil is a very large and diverse country where congressional elections take place in districts that are very different from one another. As the use of an aggregate measure of volatility for the country can lead to spurious conclusions, it is important to compare volatility at the district level and over time. Using a disaggregated measure that provides information by state, the authors show that electoral volatility has declined over time in almost all Brazilian states, which suggests that a more stable pattern of party competition is emerging. Bohn and Paiva (2009) also adopted a disaggregated measure of electoral volatility to illustrate the general pattern of decline in volatility rates in Brazil. They warn us, though, that this decrease has not been accompanied by an increase in voters identification with parties. They point to the fact that in order to understand the relation between electoral volatility and the institutionalization of the party system, it is necessary to separate the effect of changes to 8 Nicolau (2017) analyzed 1.167 legislative elections held between 1919 and 2015 in 137 countries and found that Brazil has the highest levels of party fragmentation in the world (Nicolau, 2017, pp. 90-91).

13 the party menu from situations where these changes do not occur. This would help us understand how much of the volatility is due to changes in voter behavior and how much is due to changes in the supply of parties available to the voter 9. Figure 2.1 plots the frequency distribution of electoral volatility across 26 states (not including the Federal District) over the period from 1998 to 2014 10. The figure shows that legislative elections for the lower chamber have been quite volatile. The average electoral volatility score, calculated by using the Pedersen index, is 27.5, and the standard deviation is 10.1. The figure makes it clear that this is not the result of a few outlier elections; the median electoral volatility is 26.3. Figure 2.1. Electoral Volatility in Brazilian States, 1998-2014 Frequency 0 10 20 30 3 17 29 24 16 9 4 1 1 0 20 40 60 Volatility Source: Author s calculations based on data from the TSE. 9 Tavits (2008a) addresses this issue in Central and Eastern Europe and concludes that, in these regions, instability results from changes in party supply. 10 The Pederson index was used to calculate the electoral volatility score for all parties that took part in each election. Following the method developed by Sikk (2005), parties that merged with others during the period (1998-2014) were treated as the original party since 1998. For instance, the PST and the PGT, which were incorporated into the PL in 2003, were analyzed as PL since 1998. Integrated into the PL in 2006, PRONA was presented as PL since 1998. Finally, the PSD and the PAN, which joined the PTB (the former, in 2003 and the second, in 2006) were both treated as PTB since 1998. The Brazilian party system is now composed of 32 parties: 25 that were present throughout the 1998-2014 period, and 7 genuinely new parties (PRB and PSOL were registered in 2005; PPL and PSD, in 2011; PEN, in 2012; and SD and PROS, in 2013).

14 Are all states equally volatile? Figure 2.2 shows the 26 states arranged in order of increasing volatility. There is significant variation in volatility levels - a fact that has been demonstrated before by others, such as Bohn and Paiva (2009) and Peres, Ricci, and Rennó (2011). Why is there so much variation? Although a definitive answer is still to be found, certain elements can help understand the issue. First, Brazilian states differ from one another significantly in terms of population, economy, culture, habits, and geography. This is relevant for electoral volatility because the dynamics of electoral competition turn out to be very different in states where municipalities and the population are concentrated in one or a few regional centers from the ones where both the population and economic activity are scattered evenly all throughout the state s territory (Avelino, Biderman, and Silva, 2011, 2016). Secondly, Brazilian states are also different in terms of their magnitude, which alters the electoral incentives offered in each district. The number of deputies ranges from as low as 8 in states such as Acre, Amazonas, Mato Grosso, and Rio Grande do Norte, to as high as 53 in the state of Minas Gerais and 70 in the state of São Paulo. Thirdly, Brazilian parties are considered less institutionalized, with a low level of nationalization, which means that the national party system is composed of a variety of regional parties whose size varies from one state to another. 11 Figure 2.2. Mean Electoral Volatility by State, 2006-2014 50 40 30 20 10 00 RS SC MG SP BA PA PR PE MS PB PI RJ GO RO ES CE AC MT MA AP TO SE RN RR AL AM Source: Author s calculations based on data from the TSE. 11 In fact, the national parties are still to a considerable extent federations of state parties. (Mainwaring, 1999, p. 266). For similar views that stress the influence of Brazilian federalism over parties organization, see Ames (2001) and Samuels (1998).

15 How has electoral volatility evolved over the period from 1998 to 2014? Has it been stable? Figure 2.3 shows that when we average the score for each election across states, volatility seemed to be on a decline until 2010, but picked up again during the last round of elections. Figure 2.3. Mean Electoral Volatility by Period 35 30 25 20 2002 2006 2010 2014 Source: Author s calculations based on data from the TSE. Five new parties were created in Brazil between 2010 and 2014. Could the increase in volatility during this period be attributed to the arrival of the new parties? The answer is, in a large part, yes (Mainwaring, Power, and Bizzarro, 2017). Following Sikk's method (2005), we will first quantify how much of the overall votes these new parties received and then illustrate a few cases of votes that clearly migrated from the traditional parties to the newly created ones. Then, we show that among the traditional parties, the more institutionalized ones were better prepared to deal with the emergence of new parties.

16 Table 2.1. Electoral Volatility in 2014, by State New parties (votes % ) New parties (seats % ) State Volatility AC 32.8 6.9 0.0 AL 45.7 16.5 22.2 AM 49.0 20.7 25.0 AP 35.1 14.4 0.0 BA 22.2 15.2 12.8 CE 35.7 20.3 18.2 ES 36.2 9.1 20.0 GO 33.9 14.0 17.6 MA 32.8 10.5 11.1 MG 20.7 9.4 7.5 MS 20.0 7.8 0.0 MT 45.5 19.4 12.5 PA 30.1 20.9 29.4 PB 23.5 8.7 16.7 PE 15.9 4.3 4.0 PI 35.1 11.7 10.0 PR 19.6 7.4 6.7 RJ 26.6 14.1 21.7 RN 43.2 25.9 25.0 RO 35.0 9.6 12.5 RR 38.5 9.9 0.0 RS 14.2 4.6 3.2 SC 28.3 17.9 18.8 SE 42.5 22.3 25.0 SP 23.4 8.3 8.6 TO 42.5 12.7 12.5 Brazil 31.8 11.8 12.7 Source: Author s calculations based on data from the TSE. The new parties received a substantial portion of the overall votes in most Brazilian states (Table 2.1, column 3). In some cases, such as the states of Amazonas, Ceará, Pará, Rio Grande do Norte, and Sergipe, new parties received more than one fifth of the total votes for federal congressional representatives. Even in states where they did not do so well, they still obtained a significant share of votes. This was the case in Pernambuco, which was the state where new parties had the smallest amount, with only 4.3% of the votes, and in Rio Grande do Sul, where they only had 4.6% of the votes.

17 In terms of representation (Table 2.1, column 4), the new parties did not win seats in Congress in only four out of the 26 states analyzed. In most states, they achieved a sizable number of seats. Furthermore, in seven states, new parties elected 20% or more of state representatives 12. Do new parties obtain a higher share of votes (or seats) in states where electoral volatility is higher? A simple correlation of volatility and new parties votes yields 0.67, indicating that where new parties received more votes, volatility tended to be higher. The correlation between volatility and seats is weaker, but still positive, at 0.47. To understand how these parties became so important in such a short period, it is worth analyzing a few cases. Did more institutionalized parties lose votes to the new parties, or did the new parties grow at the expense of smaller and less affluent parties? Did the new parties win support because their candidates were new to politics and not affiliated to any other party beforehand? In the state of Alagoas, the new party Solidariedade (SD) received more than 130,000 votes in the 2014 election. The majority of the votes went to a candidate who ran for Congress for the first time, but who was the son of a congressman from the state who had been in office for years and was about to retire. Another new party, PROS, won over 85,000 votes. Most of them were for a candidate who came from the traditional PSB 13 ; the PSB, in turn, received 90,000 votes less in 2014 than it did in 2010. In the state of Rio Grande do Norte, the new party PSD obtained over 190,000 votes in the 2014 election. In this case, a candidate running for reelection who had switched to the PSD from a small party, the PMN, won the majority of the votes. The PMN received almost 90,000 votes less in 2014 than in 2010. Here, PROS received more than 180,000 votes thanks to a new candidate who was, however, a former municipal councilor, the grandson of a former congressman and the son of a state deputy. 12 Registered in 2011, the PSD was the fourth largest party in Congress in 2014; many of its new leaders came from the DEM and the PSDB. The other new parties created after 2010 were still rather small in 2014 (Mainwaring, Power, and Bizzarro, 2017). 13 The PSB was created in 1985 (during the return to democracy) and had 13 municipal sections in the state of Alagoas in 2010.

18 In the state of Santa Catarina, the PSD received more than 550,000 votes. The three PSD candidates who got a seat in Congress had left the DEM (former PFL), a traditional party 14, which lost over 460,000 votes in the 2014 election. These few cases described above show that the sources from which new parties got their votes in the 2014 election varied greatly. A more systematic analysis is needed to understand the relation between political parties level of institutionalization and electoral volatility. The goal of the next section is to provide such an analysis. 2.5 Measuring Party Volatility Our main hypothesis is that more institutionalized parties are better able to cope with the entrance of newcomers on the political scene than less institutionalized parties are. More institutionalized parties are assumed to have a stronger party organization in all municipalities of the state, which allows them to build loyal followers and provide their candidates with accurate information on their chances of winning office. Consequently, these parties should have lower electoral volatility. We thus treat electoral volatility as the result, not the cause of lack of party organization. This point was made by Wolinetz (2006). Tavits (2008a) also proved that causality runs in the direction of electoral volatility - that is, electoral volatility is the result for changes in the supply of parties, and not the other way around. As discussed earlier in this chapter, in Brazil, the internal organization of political parties mirrors the structure of the federation, meaning that parties structure themselves at three levels: municipal, state and national. Federal policies of the national government hinder or help small parties to survive on the national scene and, therefore, affect the nature of party coalitions and party systems. It is the local party structures, not national ones, that select legislative candidates. The voting district, the state, is an important political arena (Ames, 1995). Under these circumstances, it is tricky to determine a party s level of institutionalization at the national level. A party can have a strong presence in one state, with solid local organizational bases in terms of municipal sections, members, and tradition, and be weak in others. For this reason, we treat each party in each municipality as an independent unit of analysis. Our 14 The PFL was created in 1985 and had 220 municipal sections (as DEM) in the state of Santa Catarina in 2010.

19 measure of party organization is the percentage of municipalities in each Brazilian state that had a party municipal section in 2010 and 2014 15. We used the percent change in an individual party s vote share in a municipality as our measure of party volatility. The main difference between our measure and the Pedersen index is that ours weighs the change in the vote share by the size of the party s vote. The Pedersen index is an unweighted systemic measure that does not allow for the comparison of individual parties. In order to compare parties volatilities, we needed to adjust to take into account the importance and size of the party in the municipality. Our measure of party volatility is presented in equation 2.2 below. Equation 2.2. Measure of Party Volatility PV i,t = / 20+ ( vote &,2,( vote &,2,(*+ )/vote &,2,(*+ where PV i,t stands for Party Volatility of party i at time t ; vote &,2,( is the share of votes of party i in municipality m at time t ; vote &,2,(*+ is the share of votes of party i in municipality m at time t-1. 15 Appendix A presents aggregated data on municipal sections and temporary commissions of the 27 traditional parties.

20 Figure 2.4. Mean Party Volatility by State, 2006-2014 (in percent) 20 15 10 05 00 RS RJ BA MS SP ES CE PI TO SE MG PE SC MT RO PB GO PA PR AL AC RN RR AP MA AM Source: Author s calculations based on data from the TSE. As expected, our measure of party volatility across states (figure 2.4) closely follows the distribution of the Pedersen index of electoral volatility, described in the previous section. The mean value of party volatility by period (figure 2.5) also replicates the trend observed earlier, in the systemic measure, where volatility fell until 2010, and then increased more recently. Figure 2.5. Mean Party Volatility by Period (in percent) 7.5 7.2 6.9 6.6 6.3 6.0 2006 2010 2014 Source: Author s calculations based on data from the TSE.