Entries andWithdrawals : Electoral Coordination across Different Of ices and the Brazilian Party Systems ∗

In this article, we present new data on electoral alliances (coligações) that were formed to contest Brazilian general elections between 1986 and 2014. We present evidence to show that alliances formed for gubernatorial and lower house elections are connected to one another. These joint alliances are part of complex coordination strategies for managing the entry and withdrawal of candidates for concomitant elections regulated by different rules. As we shall show, these joint strategies result in: 01. interlinked processes of party concentration in subnational executive elections, and party fragmentation in national lower-house elections; which is the result of 02. the emergence of political parties that specialize in contesting elections for different political of ices.


bpsr
Entries and Withdrawals: Electoral Coordination across Different Offices and the Brazilian Party Systems I n this article, we present new data that includes all the electoral alliances 1 formed for general elections in Brazil between 1986 and 2014.We show that electoral alliances for state governor and for the House of Representatives (HR) express complementary aspects of a same coordination process of entries and withdrawals in concomitant electoral contests conducted according to different principles (majoritarian and proportional).The immediate consequence of this process is the reduction of the number of viable competitors in sub-national executive elections with a corresponding multiplication of competitors in elections for the HR.This results from the emergence of political parties that are specialized in contesting elections for different political of ices.The inal result is, simultaneously, a reduction of party fragmentation in sub-national executive elections alongside an intensi ication of fragmentation in HR elections.
While such research has tended to privilege alliances formed to contest seats for the HR, this is changing.Research on alliances for other of ices has grown, including research that links alliances formed for different types of of ice (BORGES and LLOYD, 2016;CARREIRA O and NASCIMENTO, 2010;DANTAS, 2007;KRAUSE and GODOI, 2010;MIRANDA, 2013;SOARES, 2013).This article follows this new line of research, from a perspective that has been largely neglected since Lavareda (1991): we investigate the link between the alliances formed for gubernatorial elections and those formed to contest the seats in the Lower House.However, we are not interested in the ideological or partisan consistency of these alliances, and much less the extent to which these decisions affect voters via coattail effects between different of ices (see BORGES and LLOYD, 2016;SAMUELS, 2005;SHUGART and CAREY, 1992;SOARES, 2013).
Rather, our emphasis constitutes a return to the original strategic analysis of Soares (1964).We stress, though, that strategic decisions are made in a game where alliances for governor and for representatives are negotiated and formed simultaneously.As we will show, this pre-coordination of entries for different electoral contests is the backbone of bpsr Fernando Limongi & Fabricio Vasselai the national party system in Brazil.As such, we endorse the interpretations of Abrucio (1998) and Samuels (2003), for whom gubernatorial elections are the epicenter of politics in Brazil.However, we adopt a different perspective.Our focus centers on a stage prior to the electoral process: that of launching candidacies.We place the gubernatorial race at the center of this process for two reasons.
First, given their political and inancial resources, governorships are the most important elected of ices in the country after the presidency.Second, electoral legislation itself makes elections for state governments the central axis for the formation of electoral alliances.It determines that all alliances for of ices elected by proportional representation (as is the case of HR elections) in a given state must be a subset of the alliance for majoritarian of ices (which is the case of gubernatorial elections) in the same state.This means that party elites engage in a typical electoral coordination game (COX, 1997) within each district (state), but they also negotiate entries and withdrawals for all concomitant electoral contests -for governor, senator and representatives (national and state).That is to say, the optimal entry strategy for a given party in a given state has to consider the combined effects of this decision over all majoritarian and proportional elections taking place simultaneously.Also, in Brazil decisions about entries are not dichotomous.A party may decide not to have its own candidate for governor, but it may still participate in the election indirectly by supporting the candidate of another party.The party that withdraws from a gubernatorial election and then formally enters into an electoral alliance led by another party, almost always does so in exchange for being included in the alliance for the House of Representative elections.
The reason for that is well known: the Open-List Proportional System adopted in Brazil allocates seats per electoral alliances, hence votes cast for a party can -and usually do -bene it an alliance partner.In fact, it has already been shown that electoral alliances for the Lower Chamber are one of the causes of the high level of party fragmentation in the Brazilian legislature (BRAGA and ROMA, 2002;DALMORO and FLEISCHER, 2005;NICO-LAU, 1996), speci ically because alliances for proportional elections in Brazil end up transferring seats from big parties to small ones (see, among others, CALVO et al., 2015;DANTAS andPRAÇA, 2010, 2010;MACHADO, 2005;NASCIMENTO et al., 2016;RODRIGUES, 1995).However, the consequences of alliances for gubernatorial elections are rarely discussed, much less the relationship between these and alliances for proportional elections.
Considering the fact that both types of alliances are formed through joint, interdependent decisions, we propose that the notorious level of party fragmentation in the HR is actually the other face of the alliance formation process which seeks to maximize the chances of success for gubernatorial candidates.In other words, we argue that the partisan concentration in sub-national executive elections is interwoven with the fragmentation of the HR, and that they are both de ined by parties' strategies of entry, which, depending on their relative strength, privilege different of ices.To address this point empirically, we start by showing that the dominant strategy of electoral entry adopted by Brazilian parties is to build alliances.It is already known that this happens in proportional elections, but we show that this is also the rule for gubernatorial elections.Almost all Brazilian political parties participate in gubernatorial contests in all states; a small minority with their own candidates and the majority as members of alliances.Hence, while the number of parties with candidates for the HR has been increasing, the number of parties with nominal candidacies for state government has either decreased or at most remained steady.The number of parties that lead an alliance, that is, those that have their own candidates to gubernatorial races, is limited.
Next, we show that this scenario is also re lected in inal electoral results.To do this, we run a simulation that demonstrates that the parties who most lose HR seats due to the existence of alliances are precisely those who are at the head of gubernatorial alliance.Conversely, the parties who bene it the most from proportional alliances are those which opt to not run their own candidates for governor.With this analysis we go beyond existing research to more clearly identify the winners and losers from proportional alliance and thus, we substitute the usual typology-big versus small parties-for another that refers to the state-wide contexts within which elections take place.This leads naturally to the question identi ied by Lavareda (1991): why would big parties accept this equilibrium-which seems sub-optimal for them-in which smaller parties take their seats?His hypothesis about the elections of the period 1950-1962 was that the big parties negotiated political support for gubernatorial elections, "in which even a small contingent of votes directed by a small party or leadership in isolation could be vital at the polls (or even before) to the extent that they would give the appearance of strength and help make candidacies viable" (LAVAREDA, 1991, p. 116).Nicolau (1996) suggested that, because of heterogeneity at the state level in terms of party size, forming alliances for the HR were in the interest of both small and big parties, thereby privileging the contextual rationality introduced by Lima Jr. (1983) in his study of the pre-1964 period.Other authors such as Braga (2006) and Melo and Saez (2007) also followed this line of argument and suggested, in passing, that campaign time on radio and television could offer a viable explanation.
We revive these arguments and make Lavareda's suggestion (1991) more concrete.For parties with a chance in gubernatorial contests, what is at stake when an alliance is built (besides getting potential opponents out of the way) is the competition for (and the transfer of) electoral resources like fractions of time in the Free Electoral Advertising Slot (Horário Gratuito de Propaganda Eleitoral -HGPE) on TV and radio.Forming alliances increases the amount of time available for candidates while reducing that of their direct competitors.We present new data that provide evidence for this strategy.Based on electoral results and data supplied by electoral courts, we estimate the percentage of the HGPE that each candidate for governor in Brazil has had since 1986.With this, we show bpsr that in gubernatorial elections there has been a strong correlation between the candidate's fraction of the HGPE and her electoral performance in the irst round.We also show that alliances assure the predominance of big parties and reduce the effective number of direct competitors in gubernatorial elections.
The rest of the paper is organized as follows.The second section introduces our database and the simulations that we carried out.The third section de ines the object of study-strategies for launching candidacies for gubernatorial and HR elections-and shows how these strategies are circumscribed by the legislation that regulates both the formation of alliances and the distribution of HGPE time.The fourth section brings in a simulation of HR seats in the absence of proportional alliance, as well as cross-checking HGPE distribution with the electoral performance of gubernatorial candidates.In addition to this, we discuss the consequences of the strategies identi ied in the previous section for the structure and development of the Brazilian party framework.The last section presents our conclusions.

Data and methodology
Of icial datasets from the Electoral Repository of the Superior Electoral Court (TSE)2 do not contain information about the composition of the electoral alliances, in terms of parties, that were formed for the elections of 1994, 1998, 2006, and 2010.For some of iceyears, the of icial source only had the name of the alliance.For 1994, all of icial data, including nominal electoral results, were restricted to only 16 states.The parties that made up each alliance are identi ied only for the elections of 1986, 2002, and 2014, but in the case of 1986 (and, in part, 2002 as well), the data contain inconsistencies and are incomplete.
Such limitations have made researchers either restrict the timespan of their analyses, privilege big parties and big alliances, or use unof icial data.Originally built for tests conducted in Vasselai (2015), the database that we use solves these problems.To create the database, we complemented the online repository data with internal TSE data obtained over several years through direct contact with TSE technical staff.Some sources, such as electoral reports, are digital, whereas others were obtained personally in print.In some cases, information was collected directly from state-level Regional Electoral Courts.Through these primary materials, we have manually reconstructed information about which parties were members of each electoral alliance for every non-municipal election between 1986 and 2014, including complete coverage of the 1994 elections.The analyses presented in this article therefore take into account all alliances and candidacies during the period 1986-2014.
We irst use our data set to construct descriptive statistics about electoral entries and alliances.Next, we use our data to simulate the party composition of the HR if electoral alliances had not been allowed.In other words, we implement an algorithm that replicated the TSE's calculations for the distribution of seats-in accordance with the legislation that was in effect in each year-but without considering the transfers of votes between parties that takes place within alliances 3 .In addition to this, for each year-state-party, we identi ied the type of electoral entry in the gubernatorial election (e.g. head or member of alliance, solo candidate).As a result, we make it possible to link the participation of parties in gubernatorial elections to gains and losses of seats in the HR.
Finally, we use our data to calculate the distribution of HGPE time for every candidate for governor in every state since 1986.We follow all the nuances of each of the many laws that regulated these elections, and even manually identify the number of seats held by every party in the national legislature for the of icial dates used by the TSE to allocate HGPE.Finally, we refer to information obtained from various Regional Electoral Courts (TREs), such as media plans and HGPE distribution records, as well as consultations with the press, in order to validate these estimations.

Running candidates: parties' entry strategies
In every election, parties come face-to-face with a basic choice: to enter an electoral contest or not.Obviously, this decision takes into account the costs and bene its involved in running candidates.Since Duverger (1954), it has been assumed that parties tend to prudently withdraw from races in which they believe they have slim chances of success.In these cases, expected costs exceed expected returns.The optimal decision of each party depends, however, on the decisions made by others potential entrants.Thus, before the beginning of the electoral process, party elites ind themselves involved in a complex game of entry coordination.This inter-elite electoral coordination (COX, 1997) becomes even more complicated when there are multiple simultaneous elections for different of ices, many of which are held under different electoral principles (majoritarian or proportional representation, one or two rounds, etc), as is the case in Brazil.
As a result, in Brazilian general elections, parties must decide, in every state, if they will enter each race or withdraw from it-this applies to elections for governor, federal and state representatives, and senator.These decisions about entry, however, are not dichotomous since parties have the possibility of forming electoral alliances.If a party joins an alliance for an of ice whose election is regulated by the majoritarian principle, it may not 3 To simulate precisely the number of legislators that would be elected by each party in the House of Representatives if there were no alliances, it is necessary to know not only the number of votes for each candidate, but also the number of votes that each party received solely for the party itself.In 1986 and 1990, the of icial electoral results in some states (including records from the results of vote counts) do not differentiate between votes for speci ic parties in the case of alliances.In these few cases, we imputed the number of votes for each party in the electoral alliance based on the percentage of votes for individual candidates within each party.
( 2018) 12 (3)  e0001 -6/27 bpsr have its own candidate but still participate in the election by supporting the candidate(s) of other parties.For PR elections, all members of an alliance provide candidates for the joint list.Therefore, in this scenario, electoral coordination gives rise to a multi-level game between parties that hinges on decisions over where and for which of ices each will offer candidates.Figure 01 shows the possible forms of entry for Brazilian parties.

Graph 1. Decision trees for electoral entries and withdrawals
Source: Authors' elaboration.
Since the removal of the "authoritarian rubble" in the 1986 election, legislation has allowed for the formation of electoral alliances, re-instating the practice of the Third Republic (1945)(1946)(1947)(1948)(1949)(1950)(1951)(1952)(1953)(1954)(1955)(1956)(1957)(1958)(1959)(1960)(1961)(1962)(1963)(1964).One point that is often overlooked about alliance formation, however, is that under current legislation 4 alliances for majoritarian and proportional elections are no longer independent from one another.Article 06 of Law 7493, which regulated the 1986 election, allowed "political parties to establish alliances to register candidates for majoritarian elections, proportional elections, or both", but paragraph 12 of this article established that it would be "forbidden for political parties to establish different alliances for majoritarian elections and proportional elections".The intention of this rule was clear: alliances at these two electoral levels should maintain a certain consistency.However, the rule was strategically interpreted as: whoever forms an alliance to contest elections for the state government should also make alliances to dispute proportional elections -that is, elections for the HR or state-level Legislative Assemblies.As important, if not more so, was the logical complement to this: whoever did not form an alliance for governor would not form an alliance for proportional elections.By linking the two entry decisions to one another, the rule discouraged solo entries (single-party candidacies) in gubernatorial elections.
As always, the application of the rule was much more complicated than expected.The law contained some ambiguities and loopholes that led to some legal squabblings over its meaning and concrete application.The law, for example, does not seem to take into account the existence of two simultaneous majoritarian elections -one for governor and another for senator.In the 1990 and 1994 elections, which were regulated by speci ic 4 Electoral alliances were regulated by Article 140 of the Electoral Code of 1950.This article said nothing about the relationship between electoral alliances formed to contest elections for different of ices.legislation and regulations5 , new loopholes and ambiguities led to slightly different applications of the rule across different states, but the rule that alliances had to be identical (or that a party in alliance for a gubernatorial election had to run alone for Senate and proportional elections) was generally in effect.In practice, the alliance formed for the election of the state government subordinated the others, establishing the reference point around which other alliances had to maintain consistency.
With the enactment of Law Nº 9504/97, the general law of elections, a legal framework was established, and the subject became regulated by its Article 06, according to which: "Political parties are, within the same electoral restrictions, allowed to establish alliances for majoritarian elections, proportional elections, or both, making it possible, in this last case, to form more than one alliance for proportional elections among the parties that make up the alliance for the majoritarian election".
We argue that the interpretation of this principle that has prevailed in practice is that the majoritarian alliance should contain the proportional one.In other words, suppose that there are N parties p 1 …p N in the country and that, in a given state, p 1 runs a candidate for governor while p 2 and p 3 opt not to, forming an alliance for governor G = {p 1 , p 2 , p 3 }.In this same state, for any alliance D for the HR that contains any one of these parties, it should be true that D ⊆ G.As such, the possible combinations of alliances for the HR would be: What this means is that, on the one hand, an alliance to contest a proportional election must be formed between parties that are not competing in majoritarian elections (not even as members of different alliances).On the other hand, it also means that parties who do not form an alliance for the gubernatorial election cannot be in the same alliances for proportional elections.
The consequence of this directive is evident: it increases the overall cost of running for governor.A party that is not sure of its electoral viability for governor receives an obvious incentive from the electoral legislation to withdraw from the gubernatorial race and to seek shelter in an alliance that would help it in the proportional contests.For a party with this pro ile, the withdrawal of its own candidate for the gubernatorial race appears prudent not only in order to avoid wasting resources with a long-shot candidacy, but also to avoid the even greater waste of not being able to count on allies who could transfer votes in the election for the HR.Consequently, as opposed to the traditional Duvergerian framework in which parties only consider the consequences of their decision for one race, in the Brazilian case, they make a joint decision, that is, they consider the effects for majoritarian and proportional elections together.This reinforces Duverger's idea (1954) of a prudent withdrawal effect on the majoritarian election, while decreasing the likelihood of a coattail bpsr effect over the proportional.
However, none of this would, by itself, prevent parties with a real chance of victory in the gubernatorial election from simply waiting passively for the prudent withdrawal of those candidates that did not have a real chance, as is the case in a more traditional Duvergerian scenario.If they were to do so, they would not need to pay the cost of incorporating the withdrawers into their alliances in the HR.The detail that transforms this process into an active negotiation about withdrawing candidates for governor is the link, created by the electoral legislation, between the number of seats that parties hold in the HR and the distribution of the ixed HGPE time allocated for all elections.Any candidate who wants to increase her exposure on radio and TV will have to obtain this time from an ally.From Law 7493, which regulated the 1986 election, to the General Law of Elections (Law Nº 9504/97), an alliance is entitled to the sum of the fractions of the HGPE controlled by its members 6 .Thus, to buid an alliance amounts to obtain additional, scarce and valuable resources to run a guvernatorial campaign.
Since the redemocratization, the HGPE has been distributed in accordance with the performance of parties in legislative elections.More speci ically, the electoral laws that governed the 1986 (Law 7508/1986) and 1990 (Resolution 16402/1990) elections established that, for any of ice being contested, 1/3 of the HGPE was to be distributed equally between all parties, approximately 41.67% of the HGPE in accordance with the size of the parties in the National Congress (Upper and Lower Houses), and the remaining 1/4 of the HGPE in accordance with the size of the parties in the Legislative Assemblies in the respective states 7 .Starting with the law that governed the 1994 election (Law 8713/1993), the Senate and the State Assemblies ceased to enter the calculations: for all of ices being disputed, 50% of the HGPE was to be distributed equally between all the parties and 50% in accordance with the proportion of the seats occupied by each party in the HR.Since the establishment of the general law of elections (Law 9504/1997), this percentage has been changed to 1/3 distributed uniformly and 2/3 divided proportionally in accordance with the size of the parties in the HR.
Since 1986, therefore, the relative value of the seats in the HR, measured in terms of the fraction of the HGPE that they guarantee, has only increased.It should be noted 6 Resolution 16402 from 1990 explained in Article 27, line VI that 'the time to be distributed to the electoral alliance will correspond to the sum of individual times of the parties that constitute it'.In the relevant legislation, the references are always to parties or electoral alliances, which means that they treated the same in legal terms. 7In 1986, the exception was the Federal District (Article 1º, line II, paragraph and Law nº 7508 from 1986) and in 1990 the exceptions were the Federal District, Amapáand Roraima (Article 27º, line II, paragraph e and Resolution 16402 from 1990): in all these cases, 2/3 of the HGPE was distributed in accordance with the size of the representation of the parties in the National Congress and 1/3 was distributed equally to all parties with the right to receive the HGPE.The parties with the right to receive HGPE time were those who presented candidates for at least 1/3 of the seats up for election in the House of Representatives and State Assemblies (Article 1º, line VIII of Law 7508 from 1986 and Article 27º, line V from Resolution 16402 from 1990).though, that fractions of the HGPE are calculated by taking the national distribution of seats as a baseline.As a result, even if a party does not have a representative elected in a given state, it will have a fraction of the HGPE time to negotiate in that state.Hence, for a party without a real chance of victory in a gubernatorial election, the incentives to negotiate its withdrawal and enter the HR race in all states also increases, since the latter also guarantees larger fractions of the HGPE in all disputes in the next electoral cycle (BRAGA, 2006;MELO ans SAEZ, 2007;SPECK and CAMPOS, 2014).The result of this is that, as Melo (2010, p. 19) suggests, each party can choose which races to prioritize.
The specialized literature has overlooked the fact that, from the point of view of its effects on party fragmentation, alliances have radically different effects on the two different types of electoral contest.In majoritarian elections, parties in an alliance present just one candidate and, therefore, an alliance acts to concentrate (and not distribute) votes.In proportional elections at state level, an increase in the number of alliances also reduces the number of lists, but not the number of parties that have their own candidates and, consequently, that may win seats.As Vasselai (2015) shows, since the composition of the HR is the result of the aggregation of each state's electoral results, the increasing number of party entries across states further intensi ies the fragmentation of the House's party system.
Figure 02 presents data relating to the entry strategies of Brazilian parties for post-democratization general elections for governor and for the HR.More speci ically, this igure shows the percentage of times that each of the possible strategic decisions was taken by the parties able to compete each year.For a given of ice in a given year y, the total number of decisions made in Brazil (which is the denominator used to calculate the percentages that make up the y-axis of the graph) corresponds to the product N y * S y , where N y is the number of parties legally allowed to ield candidates in year y, and S y is the number of states that exist in year y.It is worth noting that both the number of parties and the number of states vary over this period.
The most important point shown by this graph is the fall in the number of complete withdrawals and the simultaneous growth of entries as members of an alliance.This movement is pronounced in the irst elections we analyzed, stabilizes in the next elections, and gains momentum again in the last two elections when entering into alliances became the norm.Complete withdrawals are extremely rare.In the case of majoritarian elections, however, the percentage of candidacies remains stable.The norm is to enter into an alliance in both majoritarian and proportional elections.
To get a better idea of what these numbers represent, let us consider gubernatorial elections.In 2014, 34 parties were allowed to run candidates.Given that there are 27 concomitant elections, if all parties ran their own candidates in all states, we would have a total of 918 candidates for governor.If we add together the three types of entry, we get very close to this: 844, or almost 92% of parties participating in the gubernatorial elections in Graph 2. Cumulative histogram: decisions relative to electoral 'entries' for Governor and House of Representatives Source: Dataset_Limongi_Vasselai. some form.However, if we consider the number of candidates for governor who effectively ran, this number falls to 166, with 93 supported by an alliance and 73 running alone.As such, the number of entries as a member of an alliance was 678.The fact of the matter is that a large number of candidates did not have their own candidate for governor in any of the 27 electoral districts.
For both positions, the scenario develops in a practically identical way.Complete withdrawals were common in the irst elections, but as more have been held, a gradual development can be observed towards the current situation.In recent elections, almost no party stays out of the race.This growth in participation clearly stems from the dominance of the strategy of entering the race as a member of an alliance.
To better disaggregate this information, the graphs in Figure 03 detail the entry strategies of each party in gubernatorial elections over time8 .At the beginning of this period, the vast majority of the parties did not even participate in gubernatorial elections, withdrawing completely from the race in pure Duvergerian fashion.Over time, however, this strategy became rarer: in general, the parties that abstained from having their own candidacies for governor began to negotiate their entry into the alliance of one of the strong candidates.As can be seen, this did not change the fact that few parties were capable of leading gubernatorial alliances in a large number of states.
In fact, there have been few parties capable of running a signi icant number of candidates for governor in more than one election; basically, DEM, PDT, PMDB, PP, PSB, PSDB, PT and PTB9 .Besides, this is clearly a heterogeneous group.If we consider only those capable of heading large alliances, the PMDB, PT, and PSDB stick out from the rest, as they are the only parties consistently leading large alliances in different states during the entire period and, as we will show, a large alliance is a pre-condition for electoral succcess.
The presence of the DEM, PDT, PP and PTB in the gubernatorial races, traditional parties that were considered big in the past, is clearly dwindling.As they have done in presidential elections, these parties have also been gradually abstaining from participating in the struggle to elect a governor, opting instead, in most districts, to seek shelter in alliances led by the PT, PMDB, or PSDB.At the same time, since at least 2002, the PSB has emerged as something of a fourth power, ielding candidates for governor on a more consistent basis.
Other parties, such as, for example, the PCdoB, PPS, PR and the PV, rarely run candidates for governor.Therefore, they generally do not even try their luck or test their strength in this arena10 .With the decrease in gubernatorial candidacies from mediumsized traditional parties, it is not surprising that fewer and fewer parties are ielding candidates, meaning the increase in parties' electoral participation in gubernatorial elections is accompanied by a net reduction in the number of candidacies.
On the other hand, Brazilian parties are broadening their participation in elections for the HR, spreading out across the territory, as Figure 04 shows11 .At the beginning of the period, few parties had a presence in a signi icant number of states, and it was not uncommon for the bigger ones to run with their own lists.Small parties concentrated their entry efforts on speci ic districts where, for whatever reason, they had some comparative advantage (a historical link with a given state, prominent regional leadership, dominance of speci ic mayoralties, etc).The offerings of candidates, as the next graph shows, became nationalized; in other words, practically all parties have at least one candidate in every state today.'Entering' is now the norm.Almost no one stays out of elections, but trying one's luck alone, without joining an alliance, is a strategy that few use.The exceptions con irm the rule: only the small ideological parties on the left insist on running alone.
To the extent that a given party has candidates across the 27 electoral constituencies (26 states and 1 federal district), each one of them opens up the possibility that the party will bene it from the transfer of votes that occurs within their alliance.Small parties, therefore, have an incentive to spread out across the country and to try their luck in all districts.As Vasselai (2015) shows, the larger the number of candidates presented by a given party, and the more these candidates are distributed across a large number of districts, the better the party's chances of winning a seat.
The list of parties that has presented candidates for HR in almost all states over the entire period is similar to that of those that have dominated the offerings for governor, featuring the PMDB, PSDB, and PT, in addition to the secondary tier of parties at the gubernatorial level, such as the DEM, PDT, PSB and PTB.The most visible trend, however, is the dramatic decline of solo lists, that is, the fact that fewer and fewer parties run candidates for the HR without being in alliances.The expansion of electoral alliances as a dominant strategy led to the multiplication of candidacies.By 2002, the vast majority of parties already effectively ran for a seat in practically every state.As a result, given that electoral alliances have been reducing the number of candidates for governor (or maintaining them at a low level) at the same time that they have been increasing the number of candidates for the HR (both intra-and inter-state), it is not surprising that, from the point of view of electoral results, gubernatorial results are becoming more concentrated while HR results are fragmenting.To measure the concentration or fragmentation of an electoral race, we use the classic index proposed by Laakso and Taagepera (1979), of the effective number of electoral parties, which measures how many parties concentrate the votes in a given election.One can calculate this measure for each district (that is, for each state in Brazil) or as a national aggregate (by using the aggregate number of votes for each party in all states).Figure 05 displays both pieces of information: the distribution of the effective number of intra-state (through a boxplot for each year) and national (a value per year and the trend line between these values) electoral parties for gubernatorial and HR elections over time.
In the gubernatorial races, the effective number of viable candidates-those that actually compete for the preferences of the voters-has luctuated between two and three.This con irms the suspicion of Melo (2010), for whom candidacies for governor had "the tendency to form two big blocks around which local political elites distribute themselves" (MELO, 2010, p. 10).The most common, let's say, is two and a half candidates: two stronger candidates supported by big alliances and either a third force or a myriad of conigurations of smaller alliances.In other words, just like "presidentiable" parties are rare exceptions in Brazil (in the sense that SAMUELS and SHUGART, 2010 seem to consider the rule), there are few parties that are "governorable".This scenario contradicts what, for Spoon and West (2015), should be expected in two-round executive elections.For them, just like for Duverger (1954) in his original propositions, there is an incentive for parties to try their luck in the irst round, leaving electoral coordination for the second round and thus increasing the number of candidates.This is not what has been happening in gubernatorial elections in Brazil.In fact, the coordination has been happening at the start, that is to say, the reduction to around two effective candidates has been happening in the irst round.The reason for this is that parties that could only "try their luck" in the irst round could not enter alliances with larger parties for proportional elections-this is a risk that is too high to take for those who know in advance that they have little chance of winning an executive of ice.
As opposed to what happens in presidential elections, however, there is no clear crystallization of the main contenders; that is, we do not observe the same two or three parties dominating the gubernatorial elections over time and space.If we look at any given state, the identity of the main competitors does not usually remain the same between one election and the next.However, the possible variations and alternatives are limited.The range of parties that actually run for governor "with a chance of winning" (that is, parties that lead large alliances) is limited in most contests to the three mentioned previously: the PMDB, PT, and PSDB, along with, every once in a while, a fourth force.Occasionally, parties can win individual elections in a state, but without this representing their consolidation in the long term.In other words, it is possible to relativize Melo's (2010) thesis that the repetition of gubernatorial elections was not capable of establishing a pattern of interaction and a speci ic structure for gubernatorial elections.
While gubernatorial elections have a small effective number of competitors, elections for the HR are becoming more and more fragmented, both within the states and at the aggregated national level.In other words, the strategy of having candidacies for the HR in almost all states has been reaping dividends for small parties, which always "hang on" to alliances in the hope that they will be able to bene it from the within alliance transference of votes to gain a seat.Naturally, with more alliances and more members within each of them, this increases the probability that more alliances, in each state, will pass the electoral quotient and the parties that belong to them will end up winning at least one seat.
The fact that party concentration in gubernatorial elections and party fragmentation in HR elections happen simultaneously, however, is not enough evidence to claim that they are related.To verify this relationship, it is necessary to pin point the link between the two processes, that is, it is necessary to show that the parties that lead gubernatorial alliances are exactly the same parties that also are harmed by these alliances in elections for the HR.Likewise, it is necessary to show that the parties that withdraw candidates from gubernatorial elections are those who win seats for the Lower House because of their participation in alliances.In the next section, we show that this, in fact, does indeed occur.To the best of our knowledge, this link has not been demonstrated in the extant literature.

The link between alliances for governor and for the House of Representatives and its effects
Investigating whether a given party won or lost seats because they had established an electoral alliance for the HR means speculating upon a counterfactual scenario: what would the electoral performance of this party have been if it had not entered into an alliance?The response to this question will never be de initive since the "ceteris paribus" condition would not hold.One can, however, simulate what the distribution of seats would have been in the absence of electoral alliances 12 .
Using the electoral outcomes for individual candidates and party lists for the Lower House elections, we therefore re-calculated the party quotients in each state and the distribution of seats without considering electoral alliances (following the details of speci ic legislation for each year).Using this simulated result, it is possible to know how many seats each party loses and/or wins in each state thanks to its participation in a given alliance.As emphasized earlier, we want to identify how many HR seats are transferred from the 'head' of each gubernatorial alliance to the other members.Thus, instead of the typical distinction between big and small parties, our calculations take into account the type of entry for the gubernatorial election (whether "head" or "member" of an alliance).Figure 06 shows the number of seats transferred in accordance with participation in the gubernatorial election.
The result of this exercise leaves little doubt: in general, leading gubernatorial al- 12 We know of course that political actors would have adapted to this new rule, ruling out the possibility of precisely estimating of the counterfactual scenario based on observed results.The behavior would change, as would the results.The simulation serves, however, as an approximation of the gains and losses in each year that resulted from the existence of alliances in elections for the House of Representatives.liances has led parties to lose seats in the HR.Likewise, not running one's own candidate for governor has a positive effect, producing a net gain in seats.The result is also supported by the fact that solo candidacies and full withdrawals from gubernatorial elections have no effect on the distribution of seats in the HR.This result makes sense: if we expect participation in alliances for governor to alter the performance of parties in elections for the HR (even if the effect depends on the role of each party within the alliance), it is signi icant that the graph shows that non-participation in a gubernatorial alliance does not affect the distribution of seats in the HR.
The effect of gubernatorial alliances on the distribution of seats in the HR has been becoming more accentuated over time.In the 1986 election, being the head or a member of a gubernatorial alliance had no discernible effect on parties' access to the HR.From 1990 on, however, this situation changes.The effect grows gradually, reaching signi icant levels in the 2006 election.In 2010 and 2014, the negative effect on the number of seats in the HR for the parties leading alliances is remarkable, bringing with it the opposite effect: a positive balance for members of alliances.
The specialized literature notes that alliances for the HR offer the expedient that small parties use to circumvent the electoral threshold implied in Brazilian electoral legislation (see, among others, DANTAS andPRAÇA, 2010, 2010;MACHADO, 2005;NASCI-MENTO et al., 2016;RODRIGUES, 1995).However, alliances in proportional elections, in and of themselves, do not have an unambiguous and stable effect on fragmentation in the HR.Without paying attention to the morphology, size, and internal composition of alliances, it is not possible to assess their effects.Furthermore, the contribution of alliances to the fragmentation of the HR grew insofar as alliances began to contain more and more parties.Calvo, Guarnieri and Limongi (2015), for instance, have shown that the direction of this transfer depends on the concentration-distribution of votes within the parties that make up the alliance and that this-because the small parties have more concentrated voting patterns-is why the small parties end up being the direct bene iciaries of alliances.
As can be seen in Figure 07, the speci ic contribution of proportional alliances to fragmentation takes shape and starts to become visible in the 1994 elections.In the 2002 elections, the process gained new momentum, and in 2010 and 2014, it became blatant.Without electoral alliances for the HR, party fragmentation would have remained practically constant from 1990 until 2014-and stood still between 1994 and 2010 13 . 13One should note that parties can also be induced to form electoral alliances as a response to the strategies of their opponents: if their opponents form alliances, it might be understood as necessary to pursue the same strategy.It remains to be seen, however, what the incentives or gains are for the parties that lead the gubernatorial alliances.As we pointed out initially, this crucial question was raised for the irst time by Lavareda (1991), who believed the big parties gave up seats in the HR in exchange for strengthening their candidacies for governor.However, the meaning of "strengthening" here can have different levels.A candidate for governor may have ambitions to build a big alliance for various reasons.As many authors suggest (see BRAGA, 2006;MELO and SAEZ, 2007;SAMUELS, 2003), she may be interested in convincing other parties to campaign for her, she may want to attract the support of mayors and political leaders from other parties, or she may simply wish to guarantee the withdrawn of opponents that could split the votes.Lastly, she may be interested in increasing her access to public resources such as electoral advertising.
As we have discussed before, we believe that alliances for executive elections look to maximize the HGPE time available to the "head" of the alliance.However, although this is understood as essential for campaigning and winning elections, little is known about the real effect of the HGPE on candidates' actual chances in elections for different of ices.The empirical literature on the topic is scarce, given that data about the distribution of HGPE time for several election years are dif icult to compile and organize (see SCHMITT et al., 1999;SPECK and CERVI, 2016;SPECK and CAMPOS, 2014).To assess the correlation between HGPE time and electoral performance, we use our new database to estimate the distribution of HGPE time for all gubernatorial candidates in all states since 1986, adjusted for the speci ic legislation surrounding each election14 .Since it is dif icult to obtain information about the effective representation of each Assembly on a speci ic data, we used the number of city councilmen elected in the elections that immediately preceded these dates as a proxy.In both these years, we also took into account the different rules about the distribution of the HGPE in the Federal District in 1986, and in the Federal District, Amapá, and Roraima in 1990 that were mentioned above.We also took into account the fact that, in both these elections, only the parties that ielded candidates for 1/3 of the slots in the House of Representatives and the Assemblies participated in the proportional distribution of votes.In the case of the subsequent elections, the rules were much simpler.Law 8713/1993, which regulated the 1994 election, determined that, for the Division of HGPE time, the representation of parties in the House of Representatives would be based on data regarding the promulgation of the law.We therefore used the Database of Party Migrations from NECI-USP to identify party representation on September 30, 1993.In the case of the 1998, 2002, and 2006 elections, the general law of elections, Law 9504/1997, determined that they would consider the representation inaugurated in the House of Representatives based upon the election immediately prior.Again, we used the In Figure 08, we present two series of graphs.One shows the relationship between the size of gubernatorial alliances and the electoral performance of the candidacies.The other shows the correlation between the HGPE time added by members of an alliance (that is, not taking into consideration the HGPE time of the party that led the alliance) 15 and the candidate's electoral performance.
As expected, we can see that larger alliances are frequently related to higher levels of votes while solo candidacies and alliances with few members tend to be associated with weak electoral performance.For gubernatorial candidates, broadening the alliances seems to be a good bet.Analogously, one can see a very similar result when we consider the HGPE time that each gubernatorial candidate adds by joining an alliance-that is, the fraction of the HGPE gained through the inclusion of an alliance member.While there is variation around a non-parametric lowess-smoothed line, the trend is clear: larger quantities of HGPE time added to a candidacy tend to be correlated with better electoral performance.As would be expected, this result, while evident for all years of the graph, is particularly marked after 1990.After all, from 1994 onwards, the allocation of HGPE time stopped considering party representation in the Senate and State Assemblies, which strengthened the mechanism that we describe: the negotiation of withdrawals of gubernatorial candidacies with the consequent increase in HGPE time for gubernatorial candidates in exchange for transfers of votes for HR elections.
Clearly, future investigation the relationship between the HGPE and electoral performance will require more sophisticated statistical tests.Identi ication strategies and causal inference methods are particularly needed for dealing with the problem of the direction of causality: do candidates with more HGPE time receive more votes, or do stronger candidates attract more allies, and therefore, end up having more HGPE time?Our new data, our analysis and the graphics above merely shed some light on this matter.Preliminary results from other studies, though, point in the same direction (SILVEIRA and MELLO, 2011).

Conclusion
Electoral alliances are one of the most studied objects in Brazilian political science.Since the pioneering contribution of Soares (1964), the specialized literature on the topic has only grown.As is often the case, our work has bene ited from the knowledge that has accumulated from this dense area of study.We follow diverse threads in the literature Database of Party Migrations from NECI-USP to obtain this information.Finally, after the alteration introduced by Law Nº 11300/2006, for the 2010 and 2014 elections, the distribution of HGPE time was based on representation the elections immediately prior-and we therefore reconstructed this representation using the results contained in our own database. 15By leaving out the HGPE time of the party that led the alliance, we can assess the relevance of alliance formation members.Otherwise, one would never know whether the correlation showed in the igure is dominated simply by the HGPE of the candidate's party. in formulating our proposal for an integrated understanding of alliances that are formed simultaneously in elections for governor and for seats in the House of Representatives.
Even if they do not need to be identical, the strategic decisions taken by the parties in these two races are interdependent, if for no other reason than that the legislation makes it so.Upon joining an alliance to contest a race for governor, parties limit their options for alliances in House of Representatives races.More speci ically, if they want to join an alliance for the proportional elections, the party must be part of that alliance for the gubernatorial election.The result is a reduction in candidates running for governor.
Following the suggestion of Lavareda (1991), we show that the strategy that guarantees electoral success in majoritarian elections has proportional alliances as a counterpoint.We offer much needed evidence that the "heads" of the alliances for majoritarian elections are the ones who receive resources from their allies (HGPE time) and who transfer seats to those same parties in legislative elections.As a result, instead of the usual typology (big parties/small parties), our analysis shows that the party system is organized around the roles of "head" and "members" of the gubernatorial alliance.As such, alliances contribute to the reduction of the number of entries in gubernatorial elections while simultaneously increasing the number of entries in proportional elections.The result is a process of specialization in which a few parties (the PMDB, PT, and PSDB, and to a lesser degree, the PSB or some other fourth party) contest majoritarian elections while the others dedicate themselves exclusively to contesting legislative seats.
Many questions remain unanswered, awaiting studies equipped with better strategies for identifying causal dynamics.It seems evident to us that competition for HGPE time is a central axis on which the Brazilian electoral party system rests and develops.Its speci ic weight in the agreements that lead to the formation of alliances requires speci ic and more in-depth studies.However, there is already evidence that politicians understand some of these aspects and have tried to alter them.Law 13165/2015 altered the criteria for the distribution of HGPE time in favor of larger parties so that in 2018, 10% of the time would already be reserved for equitable distribution while the 90% remaining would re lect the proportion of seats in the House of Representatives.Another innovation of the same law concerns the fraction of the HGPE that each alliance for majoritarian elections receives, given that what now determines this distribution is the time held by the six largest parties within each alliance.
In other words, the bargaining power of smaller parties was clearly reduced.It remains to be seen how much this will succeed in altering the logic of negotiations explained in this article.Suppose that there is a state in which there are three relevant candidates for governor in a given election, with each one at the head of an alliance of at least six parties.In this case, at least 18 parties would make effective contributions to the inal calculation of the distribution of the HGPE time, a number that is not insigni icant and that perhaps does not profoundly alter the logic upon which the current system operates.In any case, the modi ication of the law proves the importance of the mechanisms that we identify in this article.It also shows that politicians noticed it before academics had explored its importance in full.If anything, we hope that this example will serve to encourage new studies on the topic.
Revised by Matthew Richmond Submitted on August 19, 2017 Accepted on August 08, 2018

bpsr
Entries and Withdrawals: Electoral Coordination across Different Offices and the Brazilian Party Systems

bpsr
Entries and Withdrawals: Electoral Coordination across Different Offices and the Brazilian Party SystemsGraph 3. Cumulative histograms: no. of states where parties entered into elections for governor (by type of electoral entry) Source: Dataset_Limongi_Vasselai.

bpsr
Entries and Withdrawals: Electoral Coordination across Different Offices and the Brazilian Party Systems Graph 4. Cumulative histograms: no. of states where parties entered elections for the House of Repr.(by type of electoral entry) Source: Dataset_Limongi_Vasselai.

Graph 5 .
Effective number of electoral parties in elections for Governor and House of Representatives, in states (boxplots) and nationally (red line) Source: Dataset_Limongi_Vasselai.
Number of seats redistributed if electoral alliances for the House of Representatives were not allowed, in accordance with type of entry for governor SourceHouse of Representatives: actual party fragmentation versus party fragmentation simulated with electoral alliances not allowed Source: Dataset_Limongi_Vasselai.

Graph 8 .
Votes obtained in the 1st round and size of electoral alliance for governor and Votes obtained in the 1st round and HGPE in elections for governor, 1986-2014 Source: Dataset_Limongi_Vasselai. of the Senate in 1986 to identify manually the representation of the different parties in the House of Representatives and the Senate on July 04, 1986.In the case of the 1990 elections, Resolution 16402/1990 identi ies in Article 27º, paragraph 1º, that the the division of HGPE time was based on representation on April 03, 1990.As such, we manually identi ied representation within the House of Representatives and Senate through the Of icial Journals on the closest possible dates.As we mentioned before, these two elections distributed ¼ of the HGPE in accordance with the representation of the parties in the State Assemblies.