Candidate Experience and Electoral Success

Abstract Studies of how previous political experience affects a candidate’s electoral success have overlooked the experience that candidates get from running campaigns even if they lose. This article argues that experience running for office, whether successfully or unsuccessfully, could give candidates several benefits, such as expertise in running strong campaigns, a network of connections, and visibility among the electorate. As a result, candidate experience, not just office-holding experience, should be positively correlated with electoral success. The article tests this expectation in Brazil using a database of candidates for seven types of elected offices between 1998 and 2018. It finds that candidates who ran for, but lost, elected offices are more likely to win when they run in future elections for the same and lower-ranked offices, compared to candidates with no experience running for office. Thus, candidate experience, not just office-holding experience, is important for explaining electoral success in politics.

experience" as having won previous office (i.e., office-holding experience) rather than having run for prior office (i.e., candidate experience), regardless of whether the candidate won or lost. Where studies have controlled for candidates who ran but lost, they have not theorized the reasons why-and under what conditions-running and losing would have an effect on electoral success. Studies also have not often focused on candidate-centered elections in Latin America.
In this study, we theorize why candidate experience may have a positive association with future electoral success and test this theory with data from Brazil. We argue that running for office, whether successfully or unsuccessfully, gives candidates extensive experience that can help them and their parties run for and win future elections. Candidates gain knowledge and expertise in how to run campaigns, they develop a network of political connections that can provide campaign resources, and they attain name recognition and visibility among the electorate. The experience of having previously run, even unsuccessfully, can benefit candidates and increase their likelihood of electoral success.
To test this, we created a dataset of Brazilian candidates and their electoral success from 1998 to 2018. Brazil is an ideal scenario in which to test our theoretical expectations in Latin America because it has a candidate-centered electoral system that rewards candidates, not just parties. 1 It also provides a clear hierarchy of elected offices, whereby some offices offer more experience in areas like running future campaigns, building networks, and name recognition than others. Specifically, we rank-order seven types of Brazilian elected offices-governor, senator, mayor of a large municipality, federal deputy, mayor of a small municipality, state deputy, and city council member-according to the advantages they give to candidates running for office, and then test whether candidates who previously ran for offices that provide these benefits and lost are electorally successful in the next election.
By comparing candidates with previous experience running for an office and losing against candidates who do not have this experience-either because they are newcomer candidates or because they have experience running for other types of offices-we find that candidate experience does affect electoral success in Brazil, as we hypothesize. Candidates who have run for higher-level offices but lost the election are more likely to win when running for lower-level offices at national and state levels. Those that have previous candidate experience (running and losing) in the same office that they are running for also are more likely to win. Those with lower-level candidate experience, however, are no more likely to win higher-level office. Our results illustrate the importance of having previous election experience for a candidate's electoral performance, even if the candidate did not win the office. These findings should be generalizable beyond Brazil to other candidatecentered elections in Latin America and other regions of the world.
This study is important for several reasons. First, much of the literature on candidate quality has focused on prior experience running for and winning an office (office-holding experience and incumbency), not running and losing. Studies that have accounted for prior candidate experience have focused on running and losing an election to the same office a candidate is running for again and have not theorized exactly why experience might matter.
Second, this is one of the first studies to show empirically that strategically choosing where to run for office after losing correlates with electoral success. In a country like Brazil, where turnover is high and candidates move from office to office frequently, we demonstrate that a pattern exists for where candidates and parties focus their efforts. A run for one type of office can have a positive benefit when running for another office if candidates and parties correctly assess how candidate experience can benefit them.
Third, while many assume that candidates enter politics at the local level and work their way up, we show that in Brazil the benefits of running for office accrue in the other direction. Candidates who run and lose in higher-level elections are more likely to win future elections at lower levels. Like De Magalhães and Hirvonen (2014), we find that candidates at lower levels who run and lose are not more likely to win higher-level elections.
Finally, this study explores the role of candidate experience in the Latin American context, which has often been overlooked. Brazil may be the most applicable Latin American case for the theory, but the findings may generalize to candidate-focused elections in other Latin American countries. Our findings highlight the strategic role of candidate quality in elections, which benefits not only candidates but also parties and democratic processes.

Previous literature on candidate quality and electoral performance
An extensive literature in the United States has focused on candidate quality in congressional elections (Banks and Kiewiet 1989;Bond, Covington, and Fleisher 1985;Buttice and Stone 2012;Cox and Katz 1996;Krasno and Green 1988). These studies have sought to explain why the quality of challengers (compared to incumbents) varies across districts and over time and whether candidate quality contributes to an incumbency advantage (Cox and Katz 1996;Carson, Engstrom, and Roberts 2007). They found that candidate quality in the form of office-holding experience (i.e., incumbency) does affect electoral success.
Candidate quality does not have to be limited to experience in elected office, however. Some US scholars have recognized that nonelected political experiences, such as being a congressional aide or a party official are also useful for candidates and that it is important to distinguish between the previous types of elected offices held (Bond, Covington, and Fleisher 1985;Krasno and Green 1988). Kulisheck and Mondak (1996), for example, conduct an experiment and measure candidate quality as a combination of the skill and integrity of a candidate (as conveyed by the opinion of a fictitious organization presented to the subjects of the experiment). They find a positive effect of candidate quality on a voter's evaluation of the candidates and on their final vote choice. More recently, Buttice and Stone (2012) find evidence of the positive effect of a candidate's quality on vote share in US House elections with a measure of candidate quality based on informants' ratings of candidates on issues such as "personal integrity," "competence," "ability to work well with other leaders," "qualifications to hold office," and "overall strength as a public servant." These studies, however, have overlooked the fact that unsuccessful candidates also gain experiences that can give them an advantage when running for future office.
The comparative literature on candidate quality has similarly understated the importance of previous candidate experience. Studies largely have focused on candidate quality much in the way it is studied in the United States, defined as having previous experience in elected and nonelected public offices (e.g., Bohn 2007;Hobolt and Høyland 2011;Langston and Aparicio-Castillo 2011;De Magalhães and Hirvonen 2014;De Magalhães 2015), and finding that it has a positive effect on electoral performance. For example, De Magalhães and Hirvonen (2014) study how competing and winning office for mayor and state deputy in Brazil affects the electoral performance of these candidates when running for other offices. They find that running for state deputy and winning has a positive and significant association with the electoral performance of these candidates when they run for federal deputy (although this relationship disappears over time), and running for mayor and winning increases the probability of winning future state deputy elections (although this only holds for a subgroup of municipalities).
Some comparative studies have taken into account the effect of running for office and losing on electoral success. Early studies on political recruitment argued that running and losing might be important because being a long-standing party member and having a record of service in the party organization increases the chances that a candidate is selected to compete in the first place (e.g., Czudnowski 1972;Gallagher and Marsh 1988). Studies also found that political drive contributes to a candidate's success. Focusing on the UK, Norris and Lovenduski (1995) developed a measure of drive based on the campaign experience of candidates. They found that competing in elections and losing is an important element of future electoral success for a seat in parliament.
Only a few comparative election studies have considered candidate experience. Focusing on national legislative elections in Poland, Górecki and Kukołowicz (2014) measure candidate experience as whether a candidate had run unsuccessfully for the same office in the preceding election, and they find that it increases the probability that a candidate wins elected office. Schwindt-Bayer, Malecki, and Crisp (2010) also test the effect of candidate experience-as part of a battery of indicators of a candidate's political background-on electoral success in Malta, Australia, and Ireland's lower houses. The authors find that the effect of having previously run and lost is positive in Ireland and Australia but negligible in Malta. Focusing on Brazil, India, and Canada, Anagol and Fujiwara (2016) find that among candidates who ran unsuccessfully for election, those who were the runner-up (i.e., finished the election in second place) were more likely to run again for office and to be elected when they did. 2 Although important, these studies have largely considered candidate experience as secondary variables in their analyses 3 and have not thoroughly theorized why and how previous candidate experience should matter or tested those mechanisms.

A theory of candidate experience
Using existing literature, we posit that previous candidate experience may improve electoral success for several reasons. First, prior experience gives candidates extensive knowledge and expertise in how to run a strong campaign for political office. This knowledge about how the election process works allows candidates to make strategic decisions when they run for office. One of the most important decisions affecting electoral success involves the allocation of scarce resources such as money, people, and time (Sudulich and Wall 2011;Sudulich 2013). Making good decisions about resource allocation is crucial to a candidate's chances of being elected. For example, Sudulich and Wall (2011) find that Irish legislative candidates who diversify their campaign spending (by spending on a variety of activities such as advertisements, posters, campaign workers, travel and transport, and marketing and research) perform better in elections than candidates who do not diversify their spending. Guimarães et al. (2019) find that different types of spending can produce different electoral results in mayoral elections in Brazil, with structure-related spending (i.e., personnel, transportation, third-party services, and campaign committees) being more relevant for electoral success than strategy-related spending (i.e., polls and advertising). When candidates are running their first campaign, they may not know how best to allocate their resources. When candidates have experience running for office, they can allocate resources better and increase their chances of success.
A second reason that previous candidate experience may matter in winning future elections involves the networks candidates build when running for office (even if unsuccessfully). Previous experience running for office provides candidates with a network of politicians, donors, and activists from which they can build a stronger and better campaign in subsequent elections (e.g., Boas et al., 2014;Dumas and Shohfi 2020). These actors can be key for the fundraising efforts of candidates and can play an important role in mobilizing supporters on election day (e.g., Desmarais, La Raja, and Kowal 2014;Layman et al. 2010). In turn, mobilization and fundraising efforts might improve the electoral performance of candidates (Arceneaux and Nickerson 2009;Jacobson 1978;Desmarais, La Raja, and Kowal 2014;Rosenstone and Hansen 1993;Samuels 2001b;Snyder 1990;Speck and Mancuso 2014). Party leaders and gatekeepers are also important members of these political networks as they may support (or at least not prevent) the nomination of candidates (especially if candidates develop strong ties with these leaders at both the national and state level) when they want to run again, given their greater electoral potential in comparison to newcomer candidates. Indeed, Wojcik and Mullenax (2017) find that Brazilian legislators with larger and more diverse networks are more likely to win reelection. When candidates have had experience running for office, they have existing networks to draw on and have an advantage compared to newcomer candidates in the next election.
Third, previous candidate experience can increase a candidate's visibility among the electorate. This may occur because of the benefit of name recognition. 4 Research has shown that name recognition can affect candidate support through two different mechanisms (Abramowitz 1975;Ferejohn 1977;Stokes and Miller 1962). Mere exposure increases name recognition (Zajonc 1968). According to this argument, familiarity with a stimulus leads to warmer feelings toward it. Research on exposure and name recognition in consumer marketing argues that when individuals have been previously exposed to a brand, they are more willing to consider purchasing a product of that brand (e.g., Berry 2004, 2006;Holden and Vanhuele 1999). Extrapolating this finding to our research, candidates who previously ran may be electorally advantaged in the future among voters who were already exposed to them. In addition, name recognition might affect candidate support by an indirect pathway. It may work as a heuristic that voters use to make inferences about candidate viability (Stokes and Miller 1962;Kam and Zechmeister 2013), increasing the probability that they will support the candidate.
In this article, we argue that the way in which these things might matter run through both parties and voters. Party leaders are likely to value candidates with previous electoral experience more than those who lack this experience (Chiru and Popescu 2017;Put, Muyters, and Maddens 2020). Party leaders prefer experienced candidates who are more familiar with the campaign cycle, have developed strong intra-party networks and fundraising skills, and require less training than complete novices. Parties can benefit from the electoral capital developed by candidates during prior candidacies. As a result, parties may be more willing to help candidates coordinate their campaigns, fundraise, reach others in the party network, and improve their visibility. Voters may be more likely to vote for experienced candidates than for newcomers because of the better campaigns they are able to mount and their fundraising skills, which together allow them to spend more money (and spend more efficiently) to convey information about themselves and develop better networks for mobilizing supporters. As a result, experienced candidates become more visible in the minds of voters, increasing the likelihood that they will cast a vote for them. 5 In the next section, we describe how we operationalize these concepts and develop specific hypotheses for the Brazilian context.

Expectations for Brazil
Brazil offers a candidate-centric political system and hierarchy of political offices at the local, state, and national level, which allows us to proxy the theoretical mechanisms just described. In Brazil, electoral rules provide candidates with incentives to cultivate a personal reputation as a means to attract votes (Carey and Shugart 1995;Ames 1995Ames , 2001. Even though the control of party leaders over candidate nomination and recruitment varies across parties, levels (national/subnational), and types of offices (executive/legislative) (Álvares 2008;Bolognesi 2013;Braga 2008;Samuels 1998aSamuels , 2008Power and Mochel 2008), self-recruitment is frequent for many offices (Power and Mochel 2009). Candidates for legislative offices often mount their own campaigns (Samuels 2003(Samuels , 2008, and this gives them experience running campaigns, allows them to build networks of politicians, donors, and other relevant political actors, and makes candidates, in addition to parties, known among the electorate. Even in parties or races for political offices where party leaders may have more control of the nomination process, such as the Partido dos Trabalhadores (PT) and executive offices, parties often help coordinate personalized campaigns, and the election itself is still focused on candidates (even if party label also matters). Candidates with prior experience should have an advantage in the election. Brazilian elected offices also have high turnover, as many incumbents lose reelection and "face a pool of extremely strong challengers" (Samuels 2000a, 481); thus politicians move from elected office to elected office, making experience an important part of a political career (Santos and Pegurier 2011). It is in this type of political context that we would expect candidates' prior experience running for office to influence their likelihood of winning elections.
The specific hierarchy of elected offices in Brazil (Ames 1995;Samuels 1998bSamuels , 2003) allows us to explore whether it is candidate experience running for any office or candidate experience running for different types of offices that matters. Existing literature shows that the hierarchy of offices in Brazil ranks executive offices over legislative ones, and within each branch of government, it places national offices over state and local ones. 6 The most important office (next to the country president) is governor. In Brazil, governors control sizable budgets and are typically well-known locals who have developed deep networks in their districts (Samuels 1998b(Samuels , 2000b. Senators are located below governors, and mayors from large municipalities are below senators. 7 The Senate is associated with power and prestige in the form of longer terms in office and more staff funding (Pereira and Rennó 2013). Mayors from large cities have strong budgetary powers and significant control over bureaucracy, and politicians have come to regard holding this municipal-level executive office as a career goal in and of itself (Samuels 2000a(Samuels , 2003. Federal deputies are ranked below mayors from large municipalities but are above mayors of small cities. Mayors from smaller municipalities are further down the hierarchy in power and authority because they have less control over political appointments and budgets than mayors from large municipalities, but we rank them slightly ahead of state deputies and city councilors because they are executive offices. Subnational legislative offices, specifically state legislatures followed by city councils, round out the bottom of the hierarchy. While state deputies and city councilors have the power to tax residents, pass laws, support a police force, and "supervise" the mayor and governor to some degree (Klašnja and Titiunik 2017), they have less authority than do governors and mayors and less policy-making power than those in the national legislature. 6 We also explored two empirical rankings of offices: the average minimum number of votes needed to win an election and official salary for the office. Appendix Tables A12 and A13 provide those data. The ranking matches our theoretical ranking quite well. Small municipality mayors and state deputy offices could be reversed in order, but overall, the rankings correlate well. 7 We define large municipalities as those that have at least two hundred thousand voters for mayoral elections.
We argue that this hierarchy of offices represents elected offices that are likely to confer greater candidate experience (campaign experience, political network access, and name recognition/visibility) than others (Borchert 2011). 8 This occurs for two reasons. First, higher-level offices have more power, authority, and prestige associated with them (Samuels 1998b(Samuels , 2000bPereira and Rennó 2013), making them more attractive for candidates and campaign donors. Candidates who run for higher-ranked offices develop stronger political networks that allow them to raise more money for their campaigns in comparison to those candidates who run for lower-ranked ones (Samuels 2001a). Candidates running for offices higher in the hierarchy may therefore have a strategic advantage in their access to campaign resources from these same donor networks when they run for office again, in comparison to newcomer candidates and to candidates who have previous experience competing for lower-ranked offices.
The second way that offices higher in the hierarchy confer more benefits to candidates is through the number of candidates running for those offices. For the top offices in the hierarchy (governor, senator, and mayor from big municipality), fewer candidates compete because the district magnitude (M) is smaller, and only one or a small number of candidates run from each party. 9 With fewer candidates, each one has greater visibility. For example, both governors and mayors from large municipalities are elected under a majority runoff system, and one candidate typically runs from each party for executive offices. For Senate elections, the electoral system is a simple plurality with district magnitudes of one or two (two senators are elected in one election and one in the next), and the number of candidates per party is again small. 10 Lower down the hierarchy, electoral rules are such that any given candidate has numerous competitors both within and across parties. Federal deputies are elected via open-list proportional representation in districts that range from eight seats to seventy seats, and parties can have 50 percent more candidates on the ballot than the number of deputies elected from that district. 11 Small municipality mayoral elections are conducted via plurality rules rather than majority runoff (which is used in large municipalities). State legislatures and city councils are also elected via open-list proportional representation and use state-or city-wide districts where the number of representatives elected is the size of the state legislature or city council. 12 The number of competitors matters for visibility of candidates; the fewer people competing, the more attention each one is likely to receive, and the greater the ability of voters to remember candidates who previously ran but lost. It also matters for the amount and quality of campaign experience a candidate gets. Higher-level offices with fewer competitors should result in more serious campaigns, more time and effort going into the campaign, and thus more experience to mount future campaigns. Those candidates should also be better able to build networks and raise money because of the smaller number of competitors.
We test whether candidate experience is linked to candidate success in Brazil with three specific hypotheses. First, we hypothesize that a previous candidacy for a higher-level office will have a positive effect on the probability of winning a lower-level office (H1). 8 We cannot adjudicate which of the mechanisms are at work here, but we think the hierarchy of offices provides some indication of where visibility, campaign experience, and networks are stronger or weaker and allows a better test of the theory. 9 Brazil allowed electoral alliances across parties during the period studied in this article. 10 Senators serve eight-year terms with one-third and two-thirds of the Senate elected every four years. Three senators are elected from each of the twenty-six Brazilian states and the Federal District for a total of eighty-one seats in the Senate. 11 The chamber has 513 members elected from the twenty-six states and the Federal District. They serve four-year terms. 12 Both state legislatures and city councils are unicameral.
The campaign experience, better networks, and name recognition associated with these higher-level offices should produce greater chances of electoral success in a lower-level office. Second, we hypothesize that having previous candidate experience in the same office that candidates are running for will increase their probability of success (H2). 13 Candidates running again for the same office will already have experience campaigning, developing networks, and raising money that could help them win a future election to that office. Third, we hypothesize that having run for a lower-level office previously should have no effect on the probability of winning a higher-level office, because the benefits of campaign experience, networks, and visibility are smaller for lower-level offices compared to higher-level ones (H3).

Data and methods
To test our hypotheses, we use candidate data obtained from Brazil's Tribunal Superior Eleitoral (TSE, Supreme Electoral Court). The TSE, acting jointly with the Tribunais Regionais Eleitorais (TRE, Regional Electoral Courts), is directly responsible for the administration of the elections at the state and local level. Our dataset includes candidate-level data from the five local elections (i.e., mayor and city council) between 2000 and 2016, and from the six national and state elections (i.e., governor, senator, federal deputy, and state deputy) between 1998 and 2018. 14 Table 1 shows the number of candidates running for each office in each election year. 15 The main dependent variable is whether a candidate won a seat in the election under study. This is a dichotomous variable that takes the value of 1 if the candidate won a seat in the election and 0 otherwise. 16 Because we are analyzing elections for seven types of offices, we have seven different dependent variables: whether the candidate won a seat for governor, Senate, mayor from a big municipality, federal deputy, mayor from a small municipality, state deputy, and city council.
Our primary concept of interest is candidate experience. Yet candidate experience can be attained by running for an office and winning as well as running for an office and losing. 17 Because those who win also have office-holding experience that additionally benefits them when running in future elections (Perissinotto and Bolognesi 2010), it is a confounded measure of candidate experience. Thus, we measure candidate experience as running for and losing an election and use this as our main independent variable. Our analyses also include office-holding experience (running and winning) as a control. We operationalize candidate experience using seven dichotomous variables-one for each 13 De Magalhães (2015) examines this for Brazilian mayors and finds no incumbency effect and no impact of mayoral candidate experience on election to a federal office or state office. He studies only those candidates who won and the runners-up. As we address in our own analysis, this might be because of the wide variety of prestige associated with different municipalities; large municipalities are much more prestigious than small ones. 14 We use the 1998 and 2000 election years for national/state and local elections, respectively, as the baseline for constructing the previous experience variables. We should note that the TSE has incomplete candidate data from 1998 to 2002, thus we might be missing some observations. After contacting the TSE, we have no reason to believe that the missing data follow a systematic pattern. For more information on this see http://www.tse.jus.br/ eleicoes/estatisticas/repositorio-de-dados-eleitorais. 15 We include all candidates regardless of whether the candidate or party was competitive or not in the election, following Klašnja and Titiunik (2017). The appendix includes additional models with only competitive candidates (Tables A5-A8). Results are similar. 16 We also estimated the main models using an alternative specification of the dependent variable-the vote share obtained by each candidate (see Appendix Table A2). Our results are largely robust to this alternative specification. 17 Running and winning usually means that the candidate is an incumbent for the current office, but not always. Some current officeholders will run for a different office while still holding their current office. office under study-that take the value of 1 if the candidate ran and lost for that office in the immediately previous election, and 0 otherwise. 18 We create similar variables for the office-holding experience controls. Figure 1 presents information on our main independent variable: candidate experience measured as running and losing. The figure has a panel for each office and shows the percentage of candidates who ran unsuccessfully for that office. 19 The first panel for governor shows that candidates for this office had experience running previously for governor (5 percent), big municipality mayors (11 percent), and city council (9 percent). Senator, federal or state deputy, and small municipality mayor candidate experience was less common. Variation in type of candidate experience exists across offices; the most common type of experience is city council. Around 20 percent of candidates in federal deputy, state deputy, and city council races had previous experience running for city council.
Our statistical models include several control variables in addition to office-holding experience. First, we control for the Federal District with a dichotomous variable that takes the value of 1 for the Federal District and 0 otherwise. We expect that candidates competing in the Federal District could face a different set of incentives because the political opportunity structure is different from other states (given that the Federal District does not elect a mayor or city council) and therefore the role of experience could be different there. Second, we control for district magnitude (for the legislative offices).  18 We use a dichotomous measure of candidate experience instead of a continuous measure because the data on elections is truncated at 1998 and 2000. Data from years prior to 1998 and 2000 are incomplete, but we know that candidates have been running for office since the transition to democracy in 1985. The dichotomous measure provides an unbiased and more valid measure than a continuous one because we know who ran in the baseline 1998 and 2000 elections years. Additionally, the dichotomous measure avoids any problems of nonlinearity, in that the experience gained from not running to having run once is likely much greater than the experience gained from running in five previous elections compared to six. Appendix Tables A3 and A4 present models with a cumulative measure of previous candidate experience, one measured linearly and one measured nonlinearly. The results are similar on the whole, but as just noted, the measure is less valid so we present the dichotomous measure in our main results. 19 Appendix Figure A1 presents the same information for prior office-holding experience in each of the seven offices under study.
We control for this because, in races where a candidate faces more competition, his or her probability of winning will decrease as a function of the number of competitors. Third, the models include a series of individual-level controls that could affect the probability of running and winning an office, such as the candidates' marital status, occupation, education level, gender, and age (Aguilar et al. 2015;Langston and Aparicio-Castillo 2011). 20 Fourth, we control for unmeasured variables that might make all candidates running in the same election year more or less likely to win office by including fixed effects for year. We also consider unmeasured variables that might make candidates running for the same party and/or district more or less likely to win office by including fixed effects for party label and using clustered standard errors where appropriate. 21 Figure 1. Percentage of candidates running for office who had previous candidate experience in that or other offices. 20 More information on these variables, such as their descriptive statistics and codification, can be found in Appendix Table A11. Models in Appendix Table A9 include controls for candidate race, wealth, and campaign expenses as these could also affect candidate success (Jacobson 1978;Sudulich and Wall 2011). We only had data for a reduced sample of candidates (national and state elections in 2018), so we include these as robustness analyses only. 21 Specifically, for the federal deputy model, we include clustered standard errors by party in a district to control for any correlation between candidates running on the same party ballot in a district. For the state deputy and city council models we clustered on party only, because multiple candidates run for each party but the elections take place in one multimember district. For the Senate model, we clustered standard errors only on district, because elections for this office have at most two candidates on each party ballot but candidates do run against each other within districts. Finally, for local executive elections (mayors from big and small municipalities and governor) we did not use clustered errors at all because each party only fields one candidate and elections are carried out in only one district (either a city or a state).

Results
Our main analyses are seven independent logit models (one for each office being contested in an election year), estimating the relationship between prior candidate experience and the probability of winning an election with controls as noted previously. Appendix Table A1 presents the results of the full models for each of the offices under study. Here we present the results of the analyses with a series of figures that show the differences in the predicted probability of electoral success for those who ran unsuccessfully for an office previously compared to those with no prior candidate experience. Appendix Table A10 shows the raw predicted probabilities associated with the figures.
Overall, we find support for our hypotheses that candidate experience is associated with electoral success. Figure 2 shows the difference in the predicted probability of winning a lower-level office between those who previously ran for a higher-level office and those with no prior experience (H1). The figure shows that having previous experience running for higher-level offices yields positive probabilities of winning compared to having no experience for all offices except mayor of a small municipality. These relationships are often significantly different from zero. For example, having previous experience running for governor, for senator, or for mayor of a big municipality significantly increases the probability that a candidate wins a federal deputy seat by 0.29, 0.23, and 0.43, respectively (compared to a candidate with no such experience). Candidates running for legislative offices at the state level also benefit from having previous experience running for higher-level offices, with the largest increase seen in having previously run for mayor of a big municipality. Candidates running for the office lowest on the hierarchy, city council, benefit from having experience as a candidate for all other higher-level offices. Experience running for (but losing) a governorship increases one's probability of winning by 0.65, and candidate experience for a Senate seat increases the chance of winning a city council position by 0.56, compared to candidates with no experience. For small-municipality mayoral elections, experience running for the higher-level offices of senator, big municipality mayor, or federal deputy does not increase the probability of show the difference in predicted probability of winning a seat (i.e., first difference) for the office on the x-axis between a candidate with previous experience running for the higher-level office indicated and a candidate with no such previous experience. Differences were calculated using MoreClarify in Stata. Bars show 95% confidence intervals.
winning compared to candidates with no experience, but prior experience as a gubernatorial candidate does. These findings largely support H1. Figure 3 presents results for candidates who previously ran for the same office that they are running for now (H2). The figure shows that having candidate experience in the same office one is running for increases the probability of electoral success for federal deputy, small-municipality mayor, state deputy, and city council, as compared to candidates without experience. For senator, this relationship is positive and near significant. The positive impact of having previous candidate experience in the same office one is running for (compared to not having it) is highest for mayors of small municipalities: having previous experience running for this office increases the likelihood that a candidate wins the election by almost 17 percent, in comparison to a candidate with no such experience. The impact of having previous experience running for the same office on the other offices varies between an increase of 9 percent for city council, 10 percent for state deputy, and 4 percent for federal deputy. These findings support H2. Figure 3 does show, however, that previous experience running for (and losing) elections for governor or big-municipality mayor does not translate into better chances of if one runs again. The difference for governor elections is positive but not significant. Prior candidate experience confers no benefit for elections; however, prior office-holding experience as a governor does. Our models showed that candidates who were previously governors did have an incumbency benefit in future elections (Appendix Table A1). They just do not get a similar boost if they had run but lost a previous election. For big-municipality mayoral elections, candidates who ran previously and lost appear to be punished to some degree compared to those with no similar experience. Figure 3 shows the probability of Figure 3. The differences in predicted probabilities of winning election to the same office. Predicted probabilities were calculated holding all continuous variables at their mean values and categorical variables at the mode. Circles show the difference in predicted probability of winning a seat (i.e., first difference) between a candidate with previous experience running for the office indicated and a candidate with no experience running for that office. Differences were calculated using MoreClarify in Stata. Bars show 95% confidence intervals.
winning is −0.11. Yet big-municipality mayoral candidates who ran and won previously (i.e., have previous office-holding experience) are more likely to win a future election, as are incumbent governors (Appendix Table A1). Thus candidate experience does not matter for these races, but office-holding experience does. 22 We explore our final hypothesis (H3) with Figure 4. This figure shows the differences in predicted probabilities of success in higher-level offices for those having experience running in lower-level offices compared to those with no experience running for those offices. In general, having previous experience running for lower-level offices has no significant relationship with electoral success when running for higher-level offices, as expected. A few cases are even statistically significant in a negative direction. This means that candidates who have experience running for lower-level offices and later run for higher-level ones are actually disadvantaged compared to candidates with no experience. This could be because running for a higher-level office involves a more competitive election and a larger set of donors and voters who may perceive the candidate as a loser unable to win a higher-level office. They may prefer to contribute to another candidate's campaign (perhaps one who ran for a higher-level office before) or vote for another candidate, thus disadvantaging a candidate who previously ran unsuccessfully for lower-level office. Another possible explanation of the negative coefficients could be the fact that higher-level offices are more attractive to strong newcomer candidates (candidates who are wealthy and have access to resources to finance their campaign, and/or outsiders who are well-known among the electorate), making candidates with previous lower-level candidate experience look worse even than candidates with no experience. . The differences in predicted probabilities for winning higher-level offices. Predicted probabilities were calculated holding all continuous variables at their mean values and categorical variables at the mode. Circles show the difference in predicted probability of winning a seat (i.e., first difference) for the office on the x-axis between a candidate with previous experience running for the lower-level office indicated and a candidate with no such previous experience. Differences were calculated using MoreClarify in Stata. Bars show 95% confidence intervals. Note that candidate experience variables for federal deputy, state deputy, mayor of small municipality, and state deputy were not included in the governor model because they predicted failure perfectly. City council candidate experience is excluded from the mayor of a large municipality analysis for the same reason. Figure 4 shows two instances where the effect of running for a lower-level office goes against our expectation and is, instead, positive and statistically significant. These cases are candidates with previous big municipality candidate experience running for senator, and candidates with previous experience running for mayor of a small municipality who later compete for federal deputy. These two cases make sense because they are instances where the offices are relatively similar to one another. The experience gained as a mayoral candidate in a large municipality translates well into a run for a Senate seat from the state where that large municipality is located (or is the capital). The visibility, campaign experience, and name recognition that could be gained in a race for a large municipality mayor would be attractive to parties and voters, many of whom would also be voting in that Senate race. Similarly, running for mayor of a smaller municipality generates experience that could help win over many of the same voters who vote for federal deputy in a district including that municipality.

Conclusion
In this study, we set out to test whether having prior candidate experience has a positive relationship with the electoral success of candidates, using Brazil as a test case. Most election studies have focused on whether having experience in elected and nonelected office affects candidates' electoral performance, but they have tended to overlook the experience that candidates gain even if they lose. As a result, our understanding of the factors that can help candidates win elected office remains limited, particularly in Latin America, which has not been heavily studied in this literature. In addition, by theorizing the mechanisms for why running for office might contribute to electoral success, we seek to extend the literature on political recruitment while connecting it to that on political careers.
We argued that previously running as a candidate provides advantages such as knowledge and expertise in running strong campaigns, a network of connections that can provide resources for future campaigns, and visibility among the electorate via name recognition. As a result, we expected prior candidate experience to affect electoral success positively. To test our main argument, we used candidate-level data from Brazil from 1998 to 2018 and developed three specific hypotheses about how campaign experience for seven different offices (governor, senator, large municipality mayor, federal deputy, small municipality mayor, state deputy, and city council) should affect a candidate's electoral performance when running for each one of them.
Our findings largely support our main expectations. First, candidates who have previous experience running for office at a higher level than the office they are currently running for are more likely to win. Second, candidates who have previous candidate experience in the same office they are running for have a higher probability of winning that office when they run for it again. This applies to candidates running for all offices except mayor of big municipalities. Third, candidates with experience in lower-level offices are in general no more likely to win higher-prestige offices. These findings offer evidence that, contrary to common assertions that candidates should start locally and work their way up the office hierarchy to national and more prestigious offices, candidates may actually be more successful when they have experience running for higher-level office (even if they lose). This may be because a run for a higher-level office provides more visibility, larger networks, and more sophisticated campaign knowledge than a run for lower-level office.
These findings are important because they underscore the role of not just office-holding experience but also candidate experience, and they show its importance in Latin America. Having experience running in an election, even if one does not win, increases the probability that a candidate will win subsequent elections if that experience carries advantages such as campaign knowledge, better networking opportunities, and visibility. Future studies about the individual-level determinants of electoral success should thus consider candidate experience explicitly in their analyses. This is particularly true when studying other candidate-centered systems, in which candidates rather than just parties are important to both running (via selection and recruitment) and winning elections. This includes Latin American countries such as Colombia, El Salvador, and Mexico, and myriad countries around the world with similar systems, such as the United States.
The findings also suggest that candidates and parties should think carefully about where to run when they are running for office with previous candidate experience. Using the knowledge, networks, and visibility created in their previous candidacy, they will be best suited to run for the same office or for an office lower in the hierarchy. Selectively choosing a race for which the benefits of prior candidate experience can come into play will increase a candidate's chance for electoral success because candidate experience provides distinct campaign advantages for success. This is particularly important in multilevel systems such as Brazil, Mexico, or the United States, where politicians might not necessarily have static ambitions.
Furthermore, this study has implications for democratic processes and outcomes and opens research agendas related to them. First, the fact that candidate experience matters for electoral success could have negative implications for political recruitment of women and minorities. If candidates who have a higher probability of getting elected are those with the energy, time, and motivation to run repeatedly for office (i.e., drive) then certain social groups might be disadvantaged in the political system. For example, research shows that women may be less politically ambitious and less likely to be recruited as a candidate (Rule 1981;Fox and Lawless 2004;Elder 2008), making them less likely to run multiple times. Women's election to office might suffer as a consequence (Schwindt-Bayer 2018). Second, while much research has focused on how the career ambitions of successful candidates affects their behavior once in office (e.g., Chasquetti and Micozzi 2014;Desposato 2005;Victor 2010), understanding the career trajectories of unsuccessful candidates could provide further insights into how politicians behave once in office.
Finally, this study has implications for the strategies that political parties and politicians use for winning elections in Latin America and other countries around the world. In particular, knowing that prior experience running for office can increase election success rates could incentivize candidates to not give up after the first try. The experience, networks, and greater visibility achieved in that effort may increase chances of winning another election contest. Candidates (and parties) should draw on those advantages and highlight them in future elections if they want to be competitive. This experience makes for better candidates and improves electoral success.