The Presence of Incumbents Electoral Competition and Reelection in Brazil (1990-2014)

What are the effects of attempts at reelection on electoral competition? We have tested the assumption that attempts at reelection have a positive and significant effect on electoral competition. We measured electoral competition using two indicators: Imbalance index T (Taagepera) and generalized concentration index (Herfindahl-Hirscham (HH)). We analyzed data on the seven last Brazilian national elections (1990-2014). We employed descriptive (variable measures and times series) and inferential statistics (regression analysis with panel data). Our main findings: 01. the percentage and the concentration of incumbents has a significant and negative effect on the concentration of votes; 02. the interaction between the percentage of incumbents in a given dispute and in a coalition has a significant and positive effect on the concentration of votes.

In this work, we test the hypothesis that the concentration of incumbents in parties/coalitions has a negative and significant effect on electoral competition. To do this, we calculated the imbalance and concentration of votes using the indexes proposed by Taagepera (1979). We added control variables and tested the relationship between these indexes and the proportion and concentration of candidates for reelection. The results suggest that the concentration of votes is the most sensitive index for measuring the distribution of votes and that the proportion and the concentration of incumbents both have significant effects on electoral competition.
The text is organized in four sections. In the first, we present studies on the Brazilian electoral system with a focus on electoral competition. Next, we review the international debate that relates congressional reelection to electoral competition. The third section presents our methodology and empirical analysis.
Our final considerations are presented in the conclusion.

Competition and the electoral system
The effects of electoral rules on the functioning of the political system has become a recurrent theme in Political Science. Duverger's pioneering work (1954) revealed the impact of the electoral formulae on numbers of parties, thus blazing a trail for further research. Rae (1967) and Lijphart (1990) went on to investigate the effects of such formulae, electoral district magnitude, and voting structures on proportionality and party numbers. While not unanimous on every aspect of the topic, both researchers concluded that electoral rules have significant effects on election results. Carey and Shugart (1995) examined the effects of electoral rules on vote personalization. They found that there is often a tension between the collective interests of political parties and the individual ones of party members. Electoral dispute outcomes turn on both party and personal reputations. Carey and Shugart (1995) looked at four electoral system characteristics as a means of estimating this relationship: 01. party leader control over electoral lists; 02. vote aggregation units for conversion into seats; 03. vote type; and 04. district magnitude. They concluded that the most personalized systems are those where: The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990-2014 (2020) 14 (1)  districts are high-magnitude. Carey and Shugart (1995) also argued that legislators' efforts to build up their personal reputations increase during reelection campaigns. Fiorina (1977), Cox (1997), Katz (2005) and others believe that candidates tend to prioritize individual interests in reelection campaigns.
Brazil is no different. Overall, the literature demonstrates that the OLPR system combined with high-magnitude districts drives personalization of the vote (AMES, 2003;NICOLAU, 2006). On the one hand, the system allows all candidates to compete against each other. On the other hand, the vast majority of votes are cast on the basis of personality. Consequently, the focus of an election moves from political parties to the individuals representing them. This is reinforced by the fact that campaigns are organized and financed by candidates. In effect, the system drives personalization of the vote. Quite reasonably, candidates tend to engage in the creation/maintenance of their personal reputations at the expense of their party affiliations. Their prime imperative is to achieve the best possible position on the electoral list. Most elections involve a plethora of candidates distinguishable in voters' eyes by their names alone. To win, candidates not only need their party/coalition to win seats, but also to rank among 'the most voted for' (NICOLAU, 2006).
Furthermore, the Brazilian system treats incumbent candidates as it does all others. For this reason, some scholars argue that members of Congress (MoCs) seek to create/maintain zones of political dominance through allocation of resources. They believe that legislators seek to control these strongholds by swapping 'pork' for votes (AMES, 2003;BORGES, PAULA AND SILVA, 2016;PEREIRA and MULLER, 2003;PEREIRA and RENNÓ, 2007;SANTOS, 2003 SAMUELS, 2003). Incumbent candidates tend to maintain or expand their strongholds -and this in turn leads to reduced space for competition (AMES, 2003;MELO, 2004). Borges, Paula and Silva (2016) demonstrate that votes for incumbents are both more widespread and more dominant. Votes for incumbents tend to be spread widely across an electoral district and to dominate a larger number of key municipalities. It was our hypothesis that this configuration has a negative effect on electoral competition in Brazil's states.
By contrast, Kinzo et al. (2004) argue that the high degree of party fragmentation seen in the Brazilian Congress provides proof of the existence of competition. Others make similar arguments for the high levels of electoral volatility among parties (BRAGA, 2006;BOHN and PAIVA, 2009;MAINWARING, 1998). Limongi (2006) found competition present in a detailed analysis of municipal-level vote distribution in congressional elections. According to him, noncompetitive electoral coalitions were present in only 320 of the 5,665 municipalities analyzed 2 . Limongi (2006) shows that approximately 1.7% of the voters who came out to the polls in the 2002 election came from noncompetitive municipalities. Likewise, Silva (2013) demonstrates that the absence of competition is not the norm in congressional races. According to Silva (2013), competition levels have been on the increase since 1994. Avelino, Biderman and Silva (2016) investigated the concentration of votes among congressional candidates and found a generalized reduction in the magnitude of this measure.
According to them, both candidates elected to office and their defeated rivals reduced their vote concentration levels. Furthermore, they have shown that deconcentration is the dominant strategy among victorious candidates (AVELINO, BIDERMAN and SILVA, 2016). Finally, Silva (2017) shows that district magnitude is a central element in any explanation of candidates' vote ______________________________________________________________________________________________ 2 With effective number below 1.5. Electoral Competition And Reelection in Brazil (1990-2014 (2020) 14 (1) e0003 -6/36

The Presence of Incumbents
concentrations. Due to magnitude, where state and federal congressional candidates compete in the same district, votes for state representatives tend to be more concentrated (SILVA, 2017).
Furthermore, there is a body of work that questions whether MoCs have the capacity to and/or interest in creating/maintaining electoral strongholds (ABRÚCIO and SAMUELS, 1997;AVELINO, BIDERMAN and BARONE, 2012;CARNEIRO and ALMEIDA, 2008;FIGUEIREDO and LIMONGI, 2008;SAMUELS, 2003). In the first place, the electoral return of 'pork' in Brazil comes under attack.
According to Samuels (2001), the large number of actors with an interest in claiming credit for purported benefits makes it difficult to turn pork into votes, especially in the absence of support from the mayors and governors (ABRÚCIO and SAMUELS, 1997 They conclude that chances improve when a party controls the local city council (AVELINO, BIDERMAN and BARONE, 2012).
In summary, a part of the literature on the subject assumes that the Brazilian electoral system encourages personalism. The interest of candidates in the formation of political 'grotões' is a consequence of this. Therefore, it further assumes that: 01. electoral competition tends to be 'banned' from some strongholds (micro-regions/municipalities), and 02. incumbent candidates are the main culprits. As shown above, there also exists a strong reaction against this perspective. Critics call into question the very existence of electoral strongholds, as well as candidates' ability to maintain them and interest in so doing. They also seek to show that parties play an important electoral role, despite the incentives towards personalism inherent in the system.

Competition and reelection
The systematic examination of the relationship between electoral competition and congressional reelection has a long history. The emergence and development of a debate on the subject are largely confined to the North American context. An inverse correlation exists between incumbents' electoral advantages and electoral competition (ABRAMOWITZ, 1991;ABRAMOWITZ et al., 2006;FEREJONH, 1977;GLAZER, BERNARD and ROBBINS, 1987;LYONS and GALDERISI, 1995;MAYHEW, 1974;TUFTE, 1973). According to Mayhew (1974), the electoral advantage enjoyed by incumbents is the main cause of reduced competition in the US House of Representatives. Likewise, Tufte (1973) believes that pro-incumbent bias in redistricting is a key variable in explaining this reduction. Lyons and Galderisi (1995) and Abramowitz et al. (2006) show that incumbents' attributes and party polarization have a negative and significant effect on electoral competition.
Doubts exist about the mechanisms linking electoral competition to congressional reelection. The specialized literature usually considers the The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -8/36 relationship between three variables: 01. the swing ratio 4 ; 02. electoral competition; and 03. votes for incumbents. In general, these works identify a strong and significant correlation between the swing ratio, competition, and the electoral performances of incumbents. A reduced swing ratio is usually associated with reduced competition, and strengthened incumbents. Three explanations are constructed from this scenario. In the first, the main culprit is redistricting. In this scenario, incumbents benefit from a process that implies reduces both swing ratio and electoral competition (LYONS and GALDERISI, 1995;TUFTE, 1973).
However, there are those who cite incumbents' attributes as an independent variable with a causal role in the process (ABRAMOWITZ, 1991;ABRAMOWITZ et al., 2006;CAMPBELL and JUREK, 2003;COX andKATZ, 1996, 2004;KAZEE, 1983;MAYHEW, 1974). According to this point of view, the resources available to an incumbent, as well as his/her capacity to see off challengers are the key variables, and are responsible for such success as an incumbent may enjoy, as well as any reduction in the swing ratio and/or electoral competition. Finally, there is a strain of thought that postulates a moderate causal relationship between swing ratio, electoral competition, and votes for incumbents (BLACK and BLACK, 2002;JACOBSON, 2007;STONECASH, 2003STONECASH, , 2008STONECASH, BREWER and MARIANI, 2003). In this last scenario, a deepening of electoral cleavages is the main cause of the entire process. The growth of incumbent advantage reflects a secular, partisan realignment of the electorate (KEY, 1959) 5 . The argument is that over the last fifty years, the configuration of Republican and Democrat electorates has changed significantly, especially in geographic and ethnic terms (BLACK and BLACK, 2002;POLSBY, 2004;SINCLAIR, 1982). According to Stonecash (2007), this movement is responsible for what has been interpreted as increased incumbent advantage. The debate remains a live one, with no clear victor thus far.

______________________________________________________________________________________________
To our knowledge, few Brazilian studies on electoral competition have considered reelection as an independent variable 6 . The impact of incumbent advantage on electoral competition remains unknown. If such an advantage does exist, little is known about its effect. In general, works on this topic examine or explain the geography of votes for incumbents. Silva (2013)  There are also studies of the opposite relationship, i.e. on the effect of competition on reelection. According to Ames (1995), the vulnerability of candidates for reelection presents a positive correlation with competition. The main reason for this is the absence of an electoral list that would prevents intra-party competition, especially among incumbents themselves (AMES, 1995). For Samuels (2000), the nature of electoral competition in Brazil conspires against incumbents.
Disadvantages are faced both in the selection of candidates (direct competition with strong candidates) and in the electoral process (all starting from the same position on the list). Similarly, Santos (2003) believes that electoral competition is an important variable for explaining both a decision to stand for reelection, and

______________________________________________________________________________________________
The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -10/36 incumbents' electoral success or failure. For Pereira and Rennó (2001), however, electoral competition should not be considered as a variable that can explain reelection 7 . This is because the competition is endogenous to a model that considers only party presence, projects and pork 8 . In following works, Pereira and Rennó (2007) show that electoral competition does not have a significant effect on congressional reelection, contrary to their expectation that it would 9 .
In short, there is an international tradition that systematically investigates the relationship between electoral competition and congressional reelection. The literature seeks to identify not only the effects, but also the mechanisms that give rise to this phenomenon. Redistricting, candidate attributes, and electoral behavior are some of the factors cited. In Brazil, most studies cite electoral competition as an explanatory variable for incumbent reelection. The general view is that competition has a significant and negative effect on reelection. A systematic analysis on the relationship between the number and distribution of incumbents in an election, and electoral competition remains to be undertaken.

Methodology
In this section, we present our research strategy. Our main hypothesis is that 'attempts at congressional reelection have significant and negative effects on electoral competition'. According to the literature, the Brazilian electoral system favors personalism. In other words, it favors an electoral focus on individual candidates. In general, incumbents take advantage of this dynamic by their greater prominence (in terms of resources and name recognition) vis-à-vis challengers. Therefore, personalism is the mechanism linking congressional reelection with electoral competition. Our assumption is that personalism produces a positive, pro-incumbent effect that in turn causes imbalance in the competition.

VD: electoral competition
There is a lively debate over the most appropriate measure for electoral competition (AVELINO, BIDERMAN and BARONE, 2012;AVELINO, BIDERMAN and SILVA, 2016;CARAMANI, 2003;HOLBROOK and VON DUNK, 1993). According to Taagepera and Ray (1977), there are many indexes whose goal is to measure "the degree to which components of an entity vary in number and differ from one another in size" (TAAGEPERA and RAY, 1997, p. 275). Where competition is present, party fragmentation has been the most frequently observed dimension The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -12/36 (GOLOSOV, 2010;NICOLAU, 2006;PEIXOTO and GOULART, 2014). In this text, we 'define electoral competition as the level of dispute between the candidates or parties for votes'. We followed the suggestion of Caramani (2003) and Silva (2013) and considered both vote fragmentation and the degree of balance among competitors as measures of competition. To do this, we looked at two measures suggested by Taagepera (1979), namely, o 'Imbalance Index' (T) and the 'Concentration Index' (HH). The goal is to obtain a more complete picture of the competitiveness of elections 10 . The formulae follow:

The presence of incumbents
We gauged reelection using two measures: 01. concentration of candidates for reelection in parties/coalitions and 02. proportion of these candidates in an election. Like the concentration of votes, the degree of concentration of incumbents was determined using the Herfindahl-Hirschmam index (HH).
______________________________________________________________________________________________ 10 An anonymous reviewer gave us the suggestion of specifying competitions. The idea is to restrict the calculation to a specific set of candidates -for example, only the ones that are elected. Another idea was to measure competition using a summary measure of candidate voting. We decided not to accept the suggestions for two main reasons: 01. any candidate classification criteria seemed arbitrary to us and 02. summary measures imply a reduction in the variance of the phenomenon. In our view, the suggestions would bring biases that would methodologically weaken the analyses.
(2020) 14 (1) e0003 -13/36 Nominally, the concentration will correspond to the sum of the square of the proportion between the candidates for reelection of each party/coalition, the total sum representing the candidates for reelection in each state in each election. The proportion of incumbents is calculated from the total number of candidates in each state in each election. The concentration and percentage of incumbents are our independent variables. Our hypothesis is that both the concentration and the percentage have negative and significant effects on electoral competition. We have assumed that incumbents enjoy the advantage in the race and, therefore, subvert competition. Therefore, in a scenario where the incumbents are many and concentrated, competition tends to be reduced. However, there is a risk of concentration increasing internal competition, but we think it reasonable to think: 01. that de-concentration of incumbents by party/coalition generates even greater competition and 02. that political parties are able to coordinate campaigns very closely to avoid conflicts between their incumbent candidates.

VC: control
To bolster the reliability of our results we have added a set of control variables. Our goal is to engage in dialog with previous work and eliminate competing causes. First, we added to our model some of the variables cited by Ames (2003) and Silva (2013) 08. percentage of coalition incumbents; and 09. percentage of incumbents from the governor-elect's party 11 . The inclusion of revenue is due to the widely documented finding of its effect on voting (ABRAMOWITZ, 1991;JACOBSON, 1978JACOBSON, , 1989JACOBSON, , 1992LEMOS, MARCELINO and PEDERIVA, 2010;PEREIRA andRENNÓ, 2001, 2007;SAMUELS, 2001). The control variable of the number of small municipalities is due to references in the literature to 'grotões' (AMES, 2003). Finally, the inclusion of the percentage of incumbents from the coalition and of the governor-elect's party is an attempt to control for the dependency of the candidates on other actors ______________________________________________________________________________________________ 11 We would like to thank an anonymous reviewer for the idea of including the last three variables. Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -14/36 in the political system (ABRÚCIO and SAMUELS, 1997;AMES, 2003;SAMUELS, 2001) 12 .

Techniques
We used descriptive and inferential statistics to analyze the data. In the descriptive statistics, we presented the main measures of central tendency and the time series of the dependent and independent variables in our research design. The inferential component is performed from a regression analysis of panel data. In this way, we intend to measure the effects that cannot be observed in pure transverse cutting or by a simple time series. The objective is to identify the effect of reelection attempts on electoral competition and to make sure that they exist throughout the observed time series.

Results
To begin the descriptive step, we present some measures of central At the suggestion of an anonymous reviewer, we report that for the calculation of the incumbent percentages from the coalition and party of the governor-elect, we used the total number of candidates in the dispute as the denominator. Coalition here refers to national-level coalitions. 13 All analyses were performed using 'Gretl' (Gnu Regression, Econometrics and Time-series Library) open source software.
Therefore, it can be said that the behavior of the imbalance series is more regular.
The next step is to analyze the descriptive statistics of the independent variables. In the Tables 02 and 03 Table 05). However, even for this measure, only 2.5% of the variation in values can be explained by its immediate past. This means that the past is not a good predictor for the future of both measures, or that the series do not exhibit regular behavior.

PropInc PropIncPrev
The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990-2014 (2020) 14 (1) e0003 -18/36 the concentration average is low. To predict the variation in concentration and proportion of incumbents, the fit of the models is even lower. Therefore, the series of these variables are stationary but not very regular. The analysis by federal unit presents some variation in this diagnosis (more details in Tables S01 and S02 in the annexes). For example, in the Federal District, 15% (Adjusted R² = 0.152) of the variance of incumbent concentration in coalitions can be explained by its immediate past.

Dependent variable
After this descriptive analysis, the next step is to try to answer the  15 We would like to thank an anonymous reviewer for suggesting we report this analysis.

Models
The theoretical discussion shows that Brazilian literature prefers to treat electoral competition as an explanatory variable for congressional reelection.
Therefore, there has been no systematic analysis of the effect of attempted reelection on electoral competition. This is despite the unquestionable variation in competition and incumbent electoral performance with each election in each The Presence of Incumbents Electoral Competition And Reelection in Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -20/36 state. So, do elections become less competitive when more incumbents decide not to run for reelection? To the best of our knowledge, this question remains unanswerable. Therefore, our model tries to estimate this effect. In Brazil, the ability of incumbents to maintain/expand their constituencies is unclear. A sizeable part of the literature argues that incumbency is worth little, as it is not easy to achieve recognition from the electorate for benefits provided. We have restricted our analysis for the period 2002-2014 for data availability reasons.
The results are reported in the Table 07.  (1) and (2) results increases suspicions about the inadequacy of OLS models for estimating the effects (Table 07) 18 . One way to increase the robustness of the estimate is to weigh the observations 19 .
______________________________________________________________________________________________ 16 Model (1) has fixed effects for transverse and temporal observations. Therefore, it controls the unobserved variables that are constant over time and between states. 17 The Wald test tests for temporal dummies significance and error heteroskedasticity. The F test tests for intercept variance between transverse unit report inadequacy and reliability problems concerning the parameters estimated by the model. 18 According to the literature, when there are divergences between the fixed and random effect models in terms of statistical significance of the estimators, the first one should be relied on (GREENE, 2008;GUJARATI and PORTER, 2011;STOCK and WATSON, 2004). 19 As seen, the fixed effect model presents the residual heteroskedasticity problem (F = 9925.88 p <0.05). Therefore it violates the assumption of constant variance of residues (GREENE, 2008;GUJARATI and PORTER, 2011;STOCK and WATSON, 2004). In Graphs S03 and S04 there is a comparison between the distribution of residues by transverse groups. Clearly, the WLS model reduces the variance of residues in most states. To standardize the results, we decided to analyze the percentage rather than the proportion, a difference restricted to the scale. 21 In this paper, we have opted not to engage in a deeper discussion about the explanatory capacity of the variables. The reason for this is the focus on the direction and magnitude of the concentration effect and the percentage of incumbents. Electoral Competition And Reelection in Brazil (1990-2014 (2020) 14 (1) (AMES, 2003;NICOLAU, 2006). This is especially true if we observe that the percentage of large party candidates is not statistically significant. However, these findings bring us closer to the literature opposed to the existence of so-called 'grotões' (AVELINO, BIDERMAN and BARONE, 2012;AVELINO, BIDERMAN and SILVA, 2016;SILVA, 2017), seeing as it is incumbents that would be responsible for the creation of these redoubts 24  percentage of coalition incumbents. Therefore, we assumed that revenue concentration will have a more robust effect when associated with the percentage of incumbents. Given the above results, we assumed that this interaction would have a positive and significant effect on vote concentration, meaning it would contribute significantly to reducing competition. Thus, in scenarios where the percentages of incumbents inside and outside the coalition are high, competition will be significantly reduced. We are betting on both the effect of campaign revenue and that of incumbent advantage (PEREIRA andRENNÓ, 2003, 2007;SAMUELS, 2001SAMUELS, , 2003. We also plan to investigate the importance of coalitions more systematically (PEREIRA and MULLER, 2003;SANTOS, 2003).

The Presence of Incumbents
Model (4) presents three main results: 01. the interaction between the percentages of incumbents is positive and significant; 02. the interaction between the percentage of incumbents and the concentration of revenue is not statistically significant; and 03. the concentration of revenue loses statistical significance. The ______________________________________________________________________________________________ 25 We do not know how much of this result is due to the way this variable was operationalized. In fact, Borges, Paula and Silva (2016) recommend that the number of candidates in the governor's coalition be controlled as well; due to a lack of data, we cannot fully follow their suggestion. Nevertheless, we thank an anonymous reviewer for the suggestion we include this work in our bibliography. Brazil (1990Brazil ( -2014 (2020) 14 (1) e0003 -24/36 first finding reinforces the importance of participation in a coalition to electoral competition analysis. The presence of incumbents, especially from coalitions, tends to intensify competition. On the other hand, the disappearance of the statistical significance of concentration highlights the risk that this variable's effect is due to omitted variables.

Final considerations
In Brazil, the lion's share of the literature supports the view that the presence of incumbents has a negative effect on competition. Dissenters tend to restrict themselves to geographic analysis of incumbent vote, or testing for the existence of electoral strongholds. Our work investigates competition , with the independent variable being congressional reelection. Therefore, we combine a descriptive analysis, considering time series, with inferential OLS and WLS models.
Our main finding contradicted our main hypothesis. In our WLS models, the concentration and percentage of incumbents had significant negative effects on vote concentration, i.e. they contributed to increased competition. We had expected that reelection attempts would unbalance the dispute by reducing competition. This is because we believed the hypothesis that personal voting tends to favor incumbents. However, we underestimate the fact that personalism is linked to the fragility of parties in organizing campaigns. Ultimately, this allows for internal competition that we had not relied on.
In the end, the main finding of this work draws attention to the weight of intra-party competition. Hitherto underestimated in Brazil, intra-party competition seems to have an effect on congressional reelection. If parties worked as teams, the most obvious effect of reelection attempts would be competition imbalance. Our results reinforce the arguments in favor of vote personalism in Brazil, especially in the absence of a closed list. Thus, it adds a caveat to works on elections that fail to take into account party organization levels.
Above all, it highlights the prevalence of candidates over parties in the electoral dispute, especially when we observe that the percentage of large party candidates is not statistically significant 26 . Our main findings also corroborate arguments against the existence of so-called 'grotões', as the incumbents themselves are supposed to be the creators of such redoubts. Overall, the findings identify competition between incumbents as having an effect on electoral competition. Future research may choose to focus exclusively on votes for these candidates, and shed more light on when, where and how these disputes occur.
In summary, our contribution to the literature focuses on two major innovations: 01. levels of analysis and 02. independent variables. Firstly, we chose to investigate competition in state elections rather than in smaller districts.
The literature suggests construction of informal districts, but, in reality, elections are based on the boundaries of the states and the Federal District. Our analysis reveals that both the constant and the effect of a set of independent variables vary significantly between states . In other words, it gives evidence of the existence of electoral subsystems. Therefore, there exists a need to examine electoral results by observing variation according to federal unit.
Finally, our analysis estimates the effect of reelection on competition, controlled by other components of the electoral system . Therefore, it identifies the importance of coalitions for analysis of the Brazilian electoral system.
We believe these innovations show the way towards new research horizons, not least to fill in the many gaps we have left unfilled. These include the roles of governors, electoral behavior, and individual budget amendments.