Abstract
Do voters sanction incumbent parties collectively for the performance of the government, or do they hold politicians accountable individually? Most economic voting studies assume that voters hold the incumbent party collectively accountable and that candidate-specific factors do not have an independent effect on how voters punish and reward incumbents. This article argues that voters weigh the past executive’s performance differently, depending on the degree to which the presidential candidate is identified with the performance of the government. It differentiates between presidents running for re-election, successors—political allies of the outgoing president—and non-successors—party-internal opponents of the president. The article shows that re-running presidents are held more accountable than the other two and that successors are subject to greater electoral sanctioning than non-successors. Empirically, the article estimates multi-level models on an original data set that combines information on candidate types with individual-level observations from 51 election years in Latin America.
On the one hand, I’m the candidate of the governing party. At the same time, I don’t need to take on the burden of being the government’s candidate. People know that my candidacy developed not only outside the government but in spite of the government.
In a mass rally on October 2, 2010, Dilma Rousseff, the candidate of the incumbent Partido dos Trabalhadores (PT), closed her electoral campaign for the presidency of Brazil. Dilma, as she is popularly known in her country, chose the city of Sao Bernardo do Campo, the industrial center of the state of Sao Paulo, for her final campaign event. The choice of location was charged with symbolism—this was where her political mentor, Luíz Inácio Lula da Silva, had initiated his political career 30 years earlier. In the course of the campaign, the media had depicted Dilma as “Lula’s handpicked successor,” and essentially every media outlet had tied her name to Lula’s, the outgoing, term-limited president. According to a poll conducted days before the election, voters perceived that an administration headed by Dilma would guarantee the continuity of Lula’s economic and social policies (Hunter, 2012). This was not surprising given that Dilma had been until recently Lula’s Chief of Staff, responsible for drafting many of the government’s policies. On the night her campaign came to an end, she once more reminded her followers that she would continue the policies adopted by her mentor, in particular the popular Bolsa Família conditional cash transfer program that had granted money to 11 million families. Three days earlier, in the last televised political advertisement before the election, Lula had made a final appeal in favor of his protégée: “Like me, Dilma loves the poor. When you vote for Dilma you are voting for me” (MercoPress, 2010).
A clear counter-example to Dilma’s campaign was the 1998 presidential election campaign of the incumbent Partido Liberación Nacional (PLN) in Costa Rica. During the campaign, the media emphasized the fact that José Miguel Corrales, the PLN presidential candidate, was an internal opponent of President José María Figueres, who had “kept his distance” from the outgoing administration (Latin American Weekly Report, 1997). As a member of the PLN left, Corrales was a staunch critic of neoliberalism, whereas the administration of his co-partisan had continued the market reform process initiated in the early 1990s. According to media reports, it was well known that Corrales had “fundamental differences with President Figueres over economic policy” (Latin American Data Base, 1997). Corrales, a member of parliament, had not occupied a cabinet position in Figueres’ administration, which made his appeal for change credible. In the party primary, he had defeated the candidate favored by the president (Picado León, 2007).
The cases of Dilma and Corrales are two examples of how parties fight election campaigns, and how the media report on them. They are, by no means, exceptional. In contemporary democracies, parties center their national election campaigns around their leading candidate (King, 2002), and, depending on the leader’s trajectory, parties and the media emphasize different issues: If incumbents run for re-election, their campaigns emphasize their accomplishments, and the media hold their record in office under great scrutiny. In cases of no re-election, parties and the media underline the candidate’s 1 links—or lack thereof—to other prominent party figures, such as the outgoing head of government. As a result, voters are usually well aware of the identity and background of the candidate of the incumbent party. 2 As Felipe Calderón’s quote in the epigraph suggests, voters likely understand the difference between being the presidential candidate of the incumbent party and being the presidential candidate of the government.
Despite the differences in trajectories of the candidates, the electoral accountability literature has not explored this internal party variation in any detail. With few exceptions, economic voting studies assume that voters hold the incumbent party collectively responsible for the performance of the government. As Samuels’ (2004) study of presidentialism claims, “There is no reason why voters could not hold political parties as a whole and presidential governments as a whole [ . . . ] accountable in a retrospective voting fashion” (p. 426). Yet, for collective party accountability to exist, a key condition must apply: Candidate-specific variables cannot have an independent effect on the magnitude of the economic vote. As the examples above show, this is a very demanding condition, given that, in some contexts, the candidate of the incumbent party has a stronger connection to the economic policies of the government than in others. Thus, by holding to the strong party ideal, the electoral accountability literature failed to account for the possibility that incumbent parties may avoid responsibility for their time in office by nominating a candidate with weak ties to the executive.
This study constitutes the first systematic attempt at understanding if, and under what conditions, the relationship between the candidate of the incumbent party and the outgoing president matters for electoral accountability. Do voters sanction incumbent parties collectively for the performance of the government, or do they hold politicians accountable individually? The article theorizes that voters do not hold incumbent parties collectively accountable, and develops a framework of executive accountability in which voters use the candidate’s trajectory to assign responsibility for government performance. Voters assign more (less) responsibility for the executive’s past actions if the candidate is closely (weakly) linked to the policy outcomes of the outgoing government.
Empirically, the article focuses on presidential democracies in Latin America, but I expand on the external validity of my findings in the conclusion. While the prospect of evaluating politicians individually is not particular to presidentialism, focusing on presidential democracies in Latin America has three main advantages: First, due to the existence of term limits, presidentialism offers greater variation in the type of candidate. In contrast to prime ministers, most incumbent presidents are not allowed to run for re-election. In this scenario, incumbent parties can either choose a candidate who is closely related to the outgoing administration, or a candidate who offers a fresh start. Second, in Latin America, the nomination of a presidential aspirant is not endogenous to the evaluation of the economy or the popularity of the president. As I show below, when voters have positive perceptions of the performance of the government, candidates with close ties to the president are not more likely to be nominated than candidates without. This means that my results are not contaminated by endogeneity bias. Third, by focusing on Latin America, home to almost half of the world’s presidential democracies (Cheibub, Gandhi, & Vreeland, 2010), I can build a homogenous data set consisting of survey data based on an identically measured set of questions. This allows me to estimate a pooled multi-level model in line with Duch and Stevenson (2008).
The article demonstrates that in Latin American presidential democracies, re-running presidents, and successor candidates—that is, presidential candidates who have strong ties to the outgoing president—are subject to greater electoral sanctioning than candidates of incumbent parties who have no close links to the president—that is, non-successors. Thus, when there is a connection between the outgoing president and the candidate of her party—as there was between Lula and Dilma—leaders’ reputations are transferred more to their successors. Moreover, the article shows that voters are attentive to intra-party dynamics and, in particular, to the cues they receive from party leaders and the media.
The article is organized into six sections. I first discuss the existing literature on the effect of institutional and political contexts on the economic vote and develop new hypotheses regarding the effect of candidate type on electoral sanctioning. The second section provides a detailed definition of the three types of candidates and describes the criteria used to code each case. The third section presents the data set, which combines original data on the different types of candidates with survey and aggregate-level data. The fourth section presents the results and shows that the main expectations of the study are corroborated. The fifth section addresses threats to causal inference, including reverse causation and omitted variable bias. The final section summarizes the results and discusses the normative implications of the findings.
What Determines Electoral Accountability?
The question of what determines electoral accountability lies at the heart of democratic theory (Manin, Przeworski, & Stokes, 1999). Answers to this question are usually framed in terms of reward–punishment models of retrospective voting. In these frameworks, voters reward incumbents that perform well and punish those that perform poorly. As Stokes (2001) suggests, “Voters use the past performance of the government to predict future performance and see the government as responsible for that performance” (p. 13).
Yet, despite normative preconceptions that electoral accountability should exist, empirical studies show that the link between the economy and the incumbent vote is often weak and unstable (Lewis-Beck, 1988; Paldam, 1991). Scholars have been able to show that national differences in the magnitude of the economic vote arise from variation in institutional and political context: There is greater electoral sanctioning when there is high clarity of government responsibility (Powell & Whitten, 1993; Whitten & Palmer, 1999); when incumbent presidents “switch policies” 3 (Johnson & Ryu, 2010); when viable alternatives to the incumbent government exist (Anderson, 2000); when presidential and legislative elections are held concurrently in presidential systems (Samuels, 2004); and when the economy is dominated by domestic relations rather than foreign trade (Duch & Stevenson, 2008; Hellwig, 2001). 4 However, even after accounting for context, much instability remains (Anderson, 2007; Dalton & Anderson, 2011).
This article argues that at least some of the variation in magnitude and remaining instability of the economic vote emerges from a failure to account for the identity of the candidate of the incumbent party. With few exceptions, 5 the economic voting literature assumes that voters hold incumbent parties accountable in the same way, irrespective of the identity of its leading candidate (Erikson, 1989; Fisher & Hobolt, 2010; Hellwig & Samuels, 2008; Hibbs, 2000; Powell & Whitten, 1993). As parties benefit from projecting an image of unity during election campaigns, the assumption is that even term-limited presidents “ought to convince voters that they share a personal and political affinity with their successor-candidate” (Samuels, 2004, p. 426).
However, even if Samuels is right and outgoing presidents always try to persuade voters that they share an affinity with their parties’ candidate, it is not clear that they will be successful in persuading voters that this affinity is real if voters do not perceive the two politicians to share the same policy positions—most notably on the economy. Clearly, political parties are not monolithic entities in which every single party member holds the same views. We know at least since Ostrogorski (1910) that parties have different factions, and that these different factions are often a reflection of the cleavage structure that exists in society (Sartori, 1976). Faction leaders and their supporters fight for power, and their ultimate goal is to secure the prime ministerial or presidential nomination (Samuels & Shugart, 2010; Siavelis & Morgestern, 2008; Strøm & Müller, 1999). As a result, if the candidate who secures the nomination has no ties to the outgoing government—that is, she belongs to a faction that advocates different policies from the ones put in place by the executive—the magnitude of the economic vote may be significantly dampened.
There is, in sum, good reason to believe that the identity of the candidate of the incumbent party matters for electoral accountability. I argue that the magnitude of the performance-based vote varies depending on whether a president runs for re-election, or whether the incumbent party nominates a successor or a non-successor. When incumbent presidents re-run, voters can directly punish the person in charge of running the government. There is no discrepancy between the policies of the government and the individual politician in charge of them. As Nadeau and Lewis-Beck (2001) suggest, a “president running for re-election is the economic manager personified. Voters can look at his or her record in office and cast a ballot accordingly” (p. 169). Voters’ ability to attribute responsibility is facilitated by the way incumbents run their campaign for re-election, and by the way the media report their campaigns: The candidates emphasize their achievements, and the media put their performance under great scrutiny. In addition, when presidents run for re-election, voters can use the incumbents’ past performance to predict the outlook of the future government’s policies. 6 Discounting major external shocks, voters can expect the president to follow a similar policy trajectory in her consecutive term. 7 Thus, I expect electoral sanctioning to be highest when presidents re-run.
When presidents do not re-run, the nomination of a successor signals to voters that the presidential candidate has at least some association with the policies of the outgoing government. Even if a successor candidate wanted to distance herself from the president—due for example to a weak economic performance—this attempt would likely lack credibility and the media and the opposition would remind voters of the connection between both politicians. Furthermore, when assessing an outgoing government’s record in office, voters can use the information about past performance to predict the policy outlook of the new government more effectively when the candidate is a successor than when she is a non-successor. The nomination of a successor gives voters greater certainty that the new government will follow a similar policy trajectory to the one already in place. In sum, I expect greater electoral sanctioning when the candidate is a successor than when she is a non-successor.
When incumbent parties nominate a non-successor, it is not evident to voters that the presidential candidate has a clear association with the performance of the government. Non-successors are intra-party opponents of the president, and usually do not share the same policy preferences as the outgoing government. In fact, most non-successors secure their parties’ nomination by opposing the policies or the candidate backed by the president. Moreover, during election campaigns, the media usually highlight the fact that these candidates do not belong to the same faction as the president or that they do not have the president’s support. When assessing the executive’s record in office, voters cannot use the information about past performance as effectively to predict the policy preferences of the new government. I thus expect the lowest levels of electoral accountability when incumbent parties nominate non-successors.
The discussion about the type of presidential candidates leads to the following hypotheses:
Types of Candidates in Latin America
As suggested in the introduction, the measure of candidate type proposed in this article consists of three categories—re-running presidents, successors, and non-successors. I consider a presidential aspirant to be a successor when she is the preferred option of the outgoing president among the group of candidates seeking to clinch the presidential nomination of the incumbent party or coalition. Conversely, presidential candidates are categorized as non-successors when they are not the preferred option of the president. Two different scenarios describe how these candidates are nominated. In the first scenario, there is contestation for the presidential nomination, that is, two or more candidates face or are scheduled to face each other in an internal election. 8 If the candidate who has the president’s support wins the internal election, she becomes a successor. If, however, a candidate who is not supported by the president wins the internal contest, she becomes a non-successor. In the second scenario there is no contestation to determine the presidential aspirant. Successors are either handpicked by the president or emerge out of an overarching consensus among different factions. Non-successors, on the contrary, are handpicked by a faction that opposes the president.
Empirically, I code candidates as successors and non-successors based on five criteria:
Media and/or election reports document public statements of support from the outgoing president to a candidate during the nomination process. If the endorsed candidate clinches the nomination, she becomes a successor. If one of the candidates not receiving a public statement of support wins the nomination, she becomes a non-successor.
Media and/or election reports describe candidates seeking nomination as the preferred option of the outgoing president during the selection process. This is indicated by expressions like “favorite,” “protégée,” “preferred candidate,” “ally,” “confidant,” and so on. Alternatively, candidates seeking nomination are described as opponents of the outgoing president. This is indicated by expressions like “rival,” “critic,” “opponent,” “defeated the preferred candidate of the outgoing president,” and so on. If the former clinches the nomination, she becomes a successor. If the latter wins the nomination, she becomes a non-successor.
Even if expressions like “protégée,” “favorite,” or “rival” and “critic” are not used explicitly, media and/or election reports make it unambiguously clear whether the candidate is a successor or a non-successor. For each case I indicate the sources in online Appendix 2.
In cases in which coalition governments nominate a single candidate, candidates who belong to the party of the president are coded as successors unless Criteria 1 to 3 indicate otherwise. Candidates who do not belong to the party of the president are coded as non-successors unless Criteria 1 to 3 indicate otherwise.
Candidates who have occupied positions of trust, like chief of staff or minister of the presidency are coded as successors unless Criteria 1 to 4 indicate otherwise. Candidates who have not occupied such positions are coded as non-successors unless Criteria 1 to 4 indicate otherwise.
The measure of candidate type proposed in this article captures (a) the degree to which voters identify the presidential candidate with the performance of the outgoing government and (b) voters’ perceptions of continuity between two administrations of the same party. Focusing on the political relationship between outgoing presidents and candidates of the incumbent party in cases of no re-election has the advantage of providing a highly identifiable measure of intra-party variation, while still capturing differences in policy positions. Both presidents and candidates are highly visible figures who are heavily scrutinized by the media. When a candidate receives the support of the outgoing president during the nomination process, voters are provided with a highly informative cue. In addition, Wiesehomeier and Benoit (2009) demonstrate that presidents’ policy positions in Latin American presidential democracies differ systematically from their own parties’ positions by being closer to the median voter in the electorate and the median party in parliament. This suggests that the distinction between successors—who are closer to the presidents’ ideal points—and non-successors—who are closer to the incumbent parties’ ideal points—also capture important differences in candidates’ policy positions.
Table 1 describes the 51 cases analyzed in this study. Four main points emerge from this table. First, presidents seeking re-election (15 in total) had a much better electoral performance than successors (19 cases) and non-successors (17 cases). The average vote for re-standing presidents was 55.1%, for successors 38.9%, and for non-successors 37.5%. In addition, re-running presidents were re-elected in 93% of the cases, while both successors and non-successors won 53% of the time. Second, the operationalization criteria (fourth column) allowed for a straightforward coding of the candidates as successors or non-successors. In 28 of 36 cases, I relied on the first and second criteria, which means that the media consistently reported candidates as being presidents’ “favorites,” “protégées,” “preferred candidates,” “allies,” or, alternatively, “rivals,” “critics,” “opponents,” and so on. I was able to confidently code the remaining eight cases by relying on the remaining criteria, which similarly provided unambiguous information (online Appendix 2 includes a detailed description of each case). 9 Third, the fact that 84% of successors and 35% of non-successors held cabinet positions in the outgoing administrations further underlines that successors were much more closely connected to the outgoing government than non-successors. Finally, the last column indicates that the media followed the trajectories of the candidates and reported on their relationship to the outgoing president. Given the strong media effects identified in the literature on political communication (Fridkin Kahn & Kenney, 2002; Sanders & Gavin, 2004), it is not unrealistic to expect most voters to have been aware that successors were closely linked to the incumbent president and that non-successors were not. 10 If voters were unaware about the identity of the candidates, it would be more difficult to detect the patterns of electoral accountability that I find.
Types of Candidates in Latin American Presidential Elections.
R = re-running presidents; S = successors; NS = non-successors; PJ = Partido Justicialista; FPV = Frente para la Victoria; AND = Acción Democrática Nacionalista; MAS = Movimiento al Socialismo; PSDB = Partido da Social Democracia Brasileira; PT = Partido dos Trabalhadores; C-DC = Concertación – Democracia Cristiana; C-PS = Concertación – Partido Socialista; PC = Primero Colombia; U = Partido de la U; PLN = Partido Liberación Nacional; PUSC = Partido Unidad Social Cristiana; PRD = Partido Revolucionario Dominicano; PLD = Partido de la Liberación Dominicana; AP = Alianza País; ARENA = Alianza Republicana Nacionalista; PAN = Partido de Avanzada Nacional; GANA = Gran Alianza Nacional; PLH = Partido Liberal de Honduras; PNH = Partido Nacional de Honduras; PRI = Partido Revolucionario Institucional; PAN = Partido Acción Nacional; PLC = Partido Liberal Constitucionalista; ALN = Alianza Liberal Nicaragüense; FSLN = Frente Sandinista de Liberación Nacional; PRD = Partido Revolucionario Democrático; PA = Partido Arnulfista; ANR = Asociación Nacional Republicana; C-90 = Cambio 90; PC = Partido Colorado; FA = Frente Amplio; MVR = Movimiento Quinta República; PSUV = Partido Socialista Unido de Venezuela.
Data and Measurement
This section introduces the data and the method used to test the hypotheses outlined in the theory. I focus on presidential elections in Latin America, in the period between 1995 and 2012, for which individual-level data are available. I rely on data from four Latin American surveys: Latinobarómetro (34 studies), Latin American Public Opinion Project (LAPOP) (13 surveys), Centro de Estudios Públicos (CEP) (three surveys), and Comparative Study of Electoral Systems (CSES) (one study). Between 1,000 and 1,800 respondents were interviewed in each survey. 11 Sixty-eight presidential elections were conducted in Latin America in the period under study—that is, 1995 to 2012. However, 18 of these elections were left out of the final data set. 12 In 12 cases, the party of the incumbent president did not field a presidential candidate from its own ranks and did not join a coalition. 13 In a further six cases, surveys were conducted before the incumbent candidate was chosen, or after the new government had taken office. 14 In total, the data set comprises 62,041 individual observations from 51 different surveys. 15
The dependent variable in the main models is a measure of national vote intention for the incumbent party or coalition. Latinobarómetro uses the standard formulation of “If there was an election tomorrow/this Sunday, which party would you vote for?” The LAPOP, CSES, and CEP surveys ask for the names of the specific candidates. 16 The dependent variable is coded as binary in all models, in which 1 represents a vote for the incumbent party, and 0 represents a vote for any other party or candidate (following standard practice, undecided voters are excluded from the analysis). Forty-nine percent of respondents across all surveys planned to vote for the incumbent party.
The main predictors are measures of government performance, candidate type, and the interaction between these two variables. I use two different measures of government performance. First, I use a standard economic sociotropic retrospective indicator for all 51 election studies included in the sample. 17 The consensus in the field is that assessments of the national economy generally trump pocketbook considerations (Duch & Stevenson, 2008; Kiewiet, 1983). The choice of a retrospective over a prospective indicator reflects the theoretical importance of the retrospective model (Fiorina, 1981; Key, 1966) and is also based on the recent finding by Singer and Carlin (2013) that in Latin America, prospective voting predominates early in the election cycle but retrospective voting dominates as the presidential term nears its end. Following Duch and Stevenson (2005, 2008), I recode the economic indicator into three categories—worse, same, and better. 18 Twenty-two percent of respondents in the sample thought the economy had improved over the previous year, 38.6% thought it had stayed the same, and 38.9% thought it had deteriorated. The crucial factor is that these proportions are balanced between cases of successor and non-successor candidates, as I show below.
The second measure of government performance is constructed for a subset of 37 surveys. The Latinobarómetro and CEP surveys ask respondents whether they approve or disapprove of the president. Respondents who approve of the president are coded 1, all others are coded 0. LAPOP asks respondents to rate the job performance of the president on a Likert-type scale. Respondents answering good and very good are coded 1, all others are coded 0. 19 In total, 48.3% of respondents in the sample approved of the performance of the president. By using presidential approval in addition to economic evaluations, I avoid reducing government performance to a pure economic measure. In fact, Singer (2011) demonstrates that the economy does not always figure at the top of voters’ concerns. He shows that it is likely to dominate voters’ evaluations under conditions of economic recession, volatility, and economic underdevelopment. In addition, all the models control for gender, age, education, marital, and socioeconomic status at the individual level 20 (see online Appendix 1).
The key second-level predictors in all the models are the three different types of candidates—re-running presidents, successors, and non-successors. The models include further second-level predictors to account for variation in the political, institutional, and economic contexts of elections. Four dummy variables capture whether the party of the president has a legislative majority (majority situation), whether presidential elections are held on the same day as parliamentary elections (concurrent elections), whether presidential candidates are nominated via a primary election, 21 and whether incumbent presidents are policy switchers. In addition, I also account for the levels of trade dependence (measured as imports and exports as a percentage of gross domestic product [GDP]), and for the levels of party system fragmentation (measured as the effective number of parties that won seats in the previous election). Excluding these variables from the analysis could have an effect on the overall levels, though not on the nature of the performance-based vote (Singer & Carlin, 2013). Therefore, I include these variables and interact them with the government perceptions measures. 22
Analysis and Results
This study employs multi-level models to test whether performance-based voting is conditioned on known contextual variables such as the type of candidate. My approach follows Duch and Stevenson (2005, 2008), in that I pull multiple individual-level surveys from different sources. For the pooled strategy to be effective, variables need to be identically measured. In the case of this study, this is not a constraint, because the wording of the question and response sets in the different election studies are similar enough as to be represented by a single variable (Duch & Stevenson, 2005).
In this hierarchical structure, Level 1 units are individual respondents and Level 2 units are the surveys themselves. A set of individual characteristics, that is, perceptions of government performance, affects vote choice, while case-specific variables, for example, the type of candidate, predict differences in the effect of the individual-level variables across cases. I estimate random-intercept random-slopes models in which the individual-level evaluations of the economy and the performance of the president are allowed to vary from cluster to cluster. 23 Models like these have become standard in political science and have been mainly used to study vote choice (e.g., Duch & Stevenson, 2005, 2008; Huber & Stanig, 2011; Kayser & Peress, 2012), but also turnout (Ghitza & Gelman, 2013).
Table 2 reports the results of the full models (the coefficients are on the logit scale). 24 The dependent variable is the vote for the incumbent party or coalition (1, 0), and the main predictors are economic perceptions, presidential approval, and the interaction of these two variables with the type of candidate. The first model uses economic evaluations as the indicator of government performance. This variable is broken into three dummies: worse, better, and same. Two of these, worse and better, are included in the models. The second model uses the dummy variable presidential approval to capture government performance and includes economic evaluations as a control variable. The key second-level predictor measuring the type of candidate is also broken into three dummy variables: non-successors, successors, and re-running presidents. Non-successor is the reference category.
Candidate Type and Performance-Based Voting.
In both models in Table 2, the main effects of the coefficients on the government performance measures are for a non-successor candidate. As candidates of the incumbent party, non-successors are expected to be associated, at least to some degree, with the performance of the government. Therefore, in Model 1, the two dummy variables economy got better and economy got worse are expected to be statistically significant and have opposite signs, with the former being positive and the latter negative. However, the results do not fully conform to this expectation, as the coefficient on the economy got better dummy is positive but not statistically significant at any conventional level. This suggests that in cases of positive economic perceptions, voters do not identify non-successor candidates with the economic record of the government.
Model 1 provides a direct test of H1 and H2 using economic evaluations to measure government performance. Two main findings emerge from Model 1. First, re-running presidents are subject to higher levels of electoral accountability than successors and non-successors. 25 This constitutes evidence in favor of H1. Second, in line with H2, successor candidates are subject to more electoral sanctioning than non-successors. The successor dummy interacted with each of the economic perception variables, is statistically significant at the 1% level, and signed in the predicted direction.
The panels in Figure 1 show predicted probabilities of voting for the incumbent party by evaluations of the national economy (top panel) and presidential approval (lower panel) for a re-running president, a successor, and a non-successor. The purpose of the figure is to show how the differences in the probability of voting for each type of candidate change as a function of economic evaluations and presidential approval. The top panel, which is based on the results of Model 1, shows that the three types of candidates are subject to different levels of economic voting. The three slopes are statistically significantly different from each other. The slope for re-running presidents is clearly steeper than the slope for successors and non-successors, while the slope for successors is steeper than the one for non-successors. This illustrates the finding from the hierarchical model that re-running presidents, successors, and non-successors are subject to different levels of electoral sanctioning.
Incumbent vote, candidate type, and government performance.
Model 1 includes further institutional and contextual variables discussed in the economic voting literature. According to Samuels (2004), there is more economic voting when presidential and legislative elections are held concurrently. However, the results of Model 1 do not conform to this expectation. The concurrent elections variable, interacted with economy got worse, has a positive and statistically significant coefficient, suggesting that the impact of negative economic evaluations on the vote is lower when presidents are elected on the same day as the legislature. Johnson and Ryu (2010) claim that there is more retrospective voting when presidents switch policies. Yet, there is also no support for this claim, as the policy switcher dummy, interacted with each of the economic perceptions variables, is not statistically significant. However, there is partial support for the thesis that high clarity of responsibility increases the size of the economic vote (Powell & Whitten, 1993)—Majority situation is statistically significant and in the expected direction when it is interacted with economy got better, but not when it is interacted with economy got worse. Finally, in contrast to Duch and Stevenson (2008) and Hellwig (2001), there is no evidence that presidents are held less accountable for the economy when elections take place in contexts of higher exposure to global economic forces. Trade dependence is not statistically significant when it is interacted with economic perceptions.
Model 2 examines the hypothesized impact of candidate type on performance-based voting, using presidential approval as the measure of government performance (economic perceptions are included as a control variable). The economic evaluations coefficients are statistically significant and in the expected direction. The main effect for the presidential approval variable is positively signed and statistically significant at the 10% level, suggesting that voters identify non-successors only marginally with the performance of the president. More importantly, the coefficients on successor and re-running president, when interacted with presidential approval, are in the expected direction and statistically significant at the 5% and 1% levels, respectively. This is consistent with H1 and H2: Re-running presidents are subject to greater electoral sanctioning than successors and non-successors, while successors are held to greater account than non-successors. Model 2 also accounts for the other contextual variables. However, none of these variables, when interacted with presidential approval, reach statistical significance at the 5% level.
The lower panel in Figure 1 shows predicted probabilities based on Model 2. The line for re-running presidents is clearly the steepest, while the line for successor candidates is steeper than that of non-successors. Re-running presidents get on average 37% more votes than successors, and 54% more than non-successors, when voters approve of the president.
To provide additional confidence in the validity of my results, I estimate the election-level vote shares for each type of candidate by different levels of economic perceptions. In this aggregate-level analysis, the dependent variable, incumbent vote, is the percentage of votes received by the incumbent president’s party. I control for past election outcomes by including the party’s percentage of the vote in the previous presidential election (previous vote). The measures of government performance are economic evaluations—the election-wide average of responses to whether the national economy got worse, stayed the same, or improved 26 —and presidential approval—measured as the percentage of respondents who approve of the president. I interact these variables with candidate type.
I test for the impact of candidate type on electoral accountability by regressing incumbent vote on economic evaluations (Models 1 and 2 in Table 3), presidential approval (Models 3 and 4), candidate type (baseline non-successor), and the interaction of the two measures of government performance with type of candidate. I employ ordinary least squares regression in all the models, and control for potential heteroskedasticity within country-groups by clustering standard errors according to countries. Following Hellwig and Samuels (2007), the models control for the effective number of parties. 27 In addition, Models 2 and 4 include an additional set of controls that capture different dimensions of the political system: whether the incumbent party has a legislative majority in parliament (majority situation), whether a primary was conducted to nominate the presidential candidate, whether the president governs in a coalition, and whether the incumbent head of state was a policy switcher.
Electoral Accountability and Types of Candidates.
The results are consistent across models and in line with the main finding emerging from the individual-level analysis: Irrespective of the performance measure, non-successors are subject to lower electoral accountability than successors and re-running presidents. Economic evaluations and presidential approval interacted with successor and re-running president are positive and statistically significant. However, in contrast to the individual-level results, presidents running for re-election are not held to higher electoral standards than successors. Successor interacted with government performance is not statistically significantly different from Performance × Re-run.
Conditional probabilities better illustrate the degree to which the type of candidate conditions the effect of economic and government perceptions on election results. As Model 1 shows, successors and re-running presidents profit more from good economic evaluations than non-successors. A 5-point increase in the measure of economic perceptions increases the vote share for successors relative to non-successors by 1.55% (0.31 × 5), and by 1.6% for re-running presidents (0.32 × 5). Similarly, Model 3 shows that a 5% increase in the percentage of respondents who approve of the president is associated with a 3% increase in the vote share for successors (0.59 × 5), and a 3.6% increase in the vote share for re-running presidents (0.72 × 5).
Addressing Threats to Causal Inference
In this section, I address two threats to causal inference: reverse causation and omitted variable bias.
Reverse Causation
Retrospective voting models usually treat economic evaluations or government approval as exogenous with the implicit assumption that they reflect some objective measure of government performance. However, in recent years scholars have raised the concern that voters’ political preferences may influence their perception of government performance depending on whether a co-partisan is in power (Duch, Palmer, & Anderson, 2000; Evans & Andersen, 2006; Stanig, 2013; Zaller, 2004). If evaluations of the government are not exogenous to party support, vote choice is potentially contaminated by problems of reverse causation. This could inflate or dampen the magnitude of the performance-based vote (Duch & Stevenson, 2008), which in the context of this study would mean that no clear conclusions can be drawn about overall or average levels of retrospective voting (Fisher & Hobolt, 2010).
Yet, even if government evaluations are to a certain extent endogenous, it remains possible to make inferences about the micro–macro interaction terms found in my models. However, there are two different interpretations to these interactions. Fisher and Hobolt (2010) hold that
[i]f the association between government evaluations and vote choice appears to be conditioned by a macro variable, it either means [ . . . ] that the strength of retrospective voting is affected by the macro variable, or that people’s partisanship affects their evaluations of government differently according to the level of the macro variable. (p. 362)
In the case of this study, this means that either performance-based voting is weaker when the candidate is a non-successor, or voters’ partisanship or personal affinity with the president affects their evaluations of the government less when the candidate is weakly connected to the president. Either way, my results indicate that voters assign different levels of importance to government performance depending on the type of candidate. For clarity of exposition, I follow Fisher and Hobolt in reporting and discussing my findings in terms of the direct interpretation.
While the political bias in retrospective evaluations has received most attention in the literature, there also are two “second order” concerns with reverse causation in this study. The first concern is that the (expected) incumbent vote may cause the type of candidate and not the other way around. This would be the case if successors (non-successors) are only nominated when the outgoing president is positively (negatively) evaluated or if only presidents who expect to win re-election decide to re-run. The second concern is that the institution of re-election may be endogenous to the performance of the president. Ten out of 15 presidents in my data set were able to re-run only after passing constitutional reform to allow immediate re-election. 28 The process of changing the constitution may focus the public’s attention on the president’s record, increasing the overall salience of the economy.
First, I address whether the nomination of a successor or a non-successor is endogenous to the popularity of the president and to the evaluations of the economy. Table 4 shows average economic evaluations and levels of presidential approval in cases in which presidential candidates are successors and cases in which they are non-successors. The differences are negligible. Economic evaluations are slightly better in election years with successors—the difference lies in the lower proportion of respondents who think the economy deteriorates—while the percentage of respondents who approve of the government is actually higher in election years with non-successors—47% against 41.4%.
Economic Evaluations and Presidential Approval (in Percentages).
Economic, Political Factors, and Successor Candidates.
In addition, I run a series of logistic regressions on the 36 cases in which there is no incumbent president re-running. The dependent variable is a dummy coded 1 if the candidate is a successor (19 cases), and 0 if she is a non-successor (17 cases). Model 1 has only one predictor, economic evaluations—the election-wide average of responses to whether the national economy got worse, stayed the same, or improved. The coefficient is not statistically significant, suggesting that economic perceptions do not predict whether a candidate is a successor or not. Model 2 includes further aggregate-level economic and political controls. The former are election year GDP growth per capita, change in the percentage of unemployed, and the log of the inflation rate. The political controls are party system fragmentation, whether a primary is scheduled to nominate the presidential candidate, whether the incumbent party is a personalist vehicle, 29 the age of the incumbent party, and the level of institutionalization of the party system—measured as the average age of all parties that gained more than 10% of the vote in the previous election. 30 None of the additional controls achieve statistical significance. 31
Second, I look at potential endogeneity problems in the decision to run for re-election. When the economy is underperforming, presidents may anticipate a defeat and may choose not to re-run. There is, however, no evidence of such calculations taking place. In fact, incumbent candidates like Menem in Argentina, Cardoso in Brazil, and Mejía in the Dominican Republic re-ran in the midst of severe economic crises. The only president in the data set who chose not to seek another term despite being allowed to do so by the constitutional arrangements already in place is the Argentine Néstor Kirchner, who according to all indications would have been easily re-elected in 2007. 32
Third, I address concerns regarding the endogeneity of the re-election institution. Ten out of 15 presidents in this study were only allowed to re-run after passing a constitutional reform in their first terms. It is difficult to imagine a situation in which poorly performing presidents would have been able to pass such reforms. As institutions in Latin America are generally fluid, most popular presidents could have attempted to change the constitution to re-run. Moreover, after 1995—the year in which my data set begins—Latin American presidents could look back at the examples of Fujimori and Menem, who had been successful in passing reform and achieving re-election. Why would other popular or well-performing presidents not follow the same course of action?
I first address this problem by testing whether incumbent presidents who re-run after a constitutional reform are subject to different levels of performance-based voting than presidents who do not need to pass reform to re-run. I re-estimate the models from Table 2 adding a fourth category to the type of candidate variable for cases in which presidents re-run after a constitutional change. Model 1 in Table 6 reports interactions of the four different types of candidates with economic evaluations, while Model 3 reports interactions of the four types of candidates with presidential approval. In both models, the baseline performance indicator is for an incumbent president who seeks re-election without changing the constitution. The baseline performance variables when interacted with constitutional change are not statistically significant in either model. This provides evidence that incumbent presidents who change their constitutions to re-run are not held more accountable than presidents who do not need to make such changes to run again.
Constitutional Reform and Presidential Re-Election.
To provide an additional robustness check, I re-estimate the models from Table 2 only taking into account the six presidents who would have been allowed to re-run by the constitutional rules already in place—that is, Néstor Kirchner and Cristina Fernández in Argentina, Lula in Brazil, Leonel Fernández in the Dominican Republic, Alberto Fujimori in Peru (for his third term), and Hugo Chávez in Venezuela (also for his third term). As Kirchner did not seek a second term in Argentina in 2007, Models 2 and 4 in Table 6 repeat the full multi-level models from Table 2 with only five re-running presidents instead of 15. The substantive results do not change. Models 2 and 4 show that government performance still matters more when presidents re-run and when the candidate of the incumbent party is a successor rather than a non-successor. This suggests that the main findings hold, assuming that institutions such as term limits are exogenous and presidents cannot change them for their own advantage.
Omitted Variables
Omitted variable bias occurs when an omitted variable is a determinant of the dependent variable and is also correlated with one or more of the predictors. To rule out the possibility that an omitted variable could be causing the relationship between candidate type, economic evaluations, and the incumbent vote, the models estimated in Table 2 are repeated with additional controls for various features of the political system and the economic context. Personalism, age of incumbent party, party system institutionalization, and coalition are four features of the political system that may be correlated with the type of candidate, while also explaining the incumbent vote (see online Appendix 1 for a codebook). As incumbent presidents leading personalist vehicles and/or recently established parties are less likely to face intra-party opposition, they may be in a better position to either change the constitution to run for a consecutive term, or to impose the nomination of a successor candidate. In a similar way, due to the absence of consolidated parties capable of putting a stop to their re-election attempts, incumbent presidents in weakly institutionalized party systems may be more successful in their attempts to change the constitutional rules to re-run. In addition, incumbent presidents may not be able to secure the nomination of a successor if their parties are in a coalition and, as a result, have to cede the presidential nomination to one of the coalition partners. Finally, three features of the economic context—per capita GDP growth, change in unemployment, and the log of the inflation rate in the year of the election—are expected to be correlated with economic evaluations.
To examine how these contextual variables condition individual-level voting behavior, I interact the various features of the political system with the individual-level measures of government performance. I also include interaction terms for the aggregate-level economic controls and the type of candidate. For reasons of space, I present the results of these models in online Appendix 4. The coefficients on the interactions between candidate type and individual-level government performance are essentially unaffected by the addition of these new variables—they remain in the same direction, are statistically significant, and are very close to the values reported in Table 2.
Discussion and Concluding Remarks
A central claim of democratic theory is that democracy induces governments to be responsive to the preferences of the people. In modern representative democracies, governments are responsive because they are elected, and citizens exert influence over policy makers by holding them electorally accountable. As political parties structure elections in most democracies, observers claim that voters sanction incumbent parties collectively for the performance of the government. This view of elections has become the dominant model of electoral accountability and has remained virtually unchallenged for more than 50 years.
This study contests the notion of collective party responsibility. It develops a framework of executive accountability, in which voters use the candidate’s identity to assign responsibility for government performance. Voters assign more responsibility for the executive’s past actions if the candidate is closely linked to the policy outcomes of the outgoing government. Empirically, the article focuses on presidential elections in Latin America between 1995 and 2012. The results show, first, that re-running presidents are subject to higher levels of electoral sanctioning than successors—that is, political allies of the outgoing president—and non-successors—that is, party-internal opponents of the outgoing president. Second, they demonstrate that successors are held more accountable for the performance of the government than non-successors.
Figure 1 displayed a graphic illustration of the main results. The slope for non-successors can be interpreted as the lower bound of party accountability, that is, accountability associated with the party brand. The fact that this slope is not statistically significantly different from zero for positive economic evaluations suggests that when the economy is performing well, voters do not give credit to candidates without close ties to the executive. The slope for successors represents the upper bound of party accountability, and the lower bound of individual accountability, while the slope for re-running presidents represents the upper bound of individual accountability. When a president runs for a consecutive term, her individual performance contributes to the accountability equation. In sum, the differences between the three slopes, and in particular between the slopes for successors and non-successors, confirm that voters do not hold incumbent parties collectively responsible for the performance of the government.
Taken together, these findings have important implications for our understanding of electoral accountability. The results suggest that presidents’ reputations are not equally transferred to all the presidential candidates of the incumbent party or coalition. Voters clearly consider whether candidates have close ties to the outgoing president, and take these ties into account when deciding their vote. Furthermore, the fact that different types of candidates are subject to different levels of performance-based voting is indicative of voters’ sophistication. It suggests that voters are attentive to intra-party dynamics and, in particular, to the cues they receive from party leaders and the media. This is true even in Latin America, where voters have been described as poorly informed (Carreras, 2014).
The results presented in this study also indicate that voters do not attribute responsibility for good and bad economic evaluations in equivalent ways: Voters punish all three types of candidates—even non-successors, to some degree—when they perceive deteriorating economic conditions, but only reward successors and re-running presidents when their evaluations of the state of the economy improve. This finding is consistent with previous literature that identifies a cognitive phenomenon called “negativity bias,” by which individuals react more to negative than to positive stimuli of similar magnitude (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Taylor, 1991). Along these lines, Stanig (2013) shows that retrospective evaluations respond to economic downturns more strongly than to recoveries. In downturns, citizens of different ideological persuasions and partisan affiliations tend to agree that the economy is underperforming, whereas in recoveries, evaluations are polarized along partisan and ideological lines.
The foregoing analysis suggests that voters’ evaluations of the economy and assessments of how good a job the president has done are not fully equivalent measures of “government performance.” Voters hold incumbent parties in majority situations more accountable for the economy, but only marginally for the perceived performance of the president. This suggests that the context in which elections take place matters more when the measure of government performance is economic evaluations than when it is presidential approval. Why this is the case is not fully clear and leaves room for further research. However, one hypothesis comes to mind. It is likely that many voters approve of the president for reasons that are not directly associated with the policy performance of the government, but with other characteristics, like the president’s charisma, or whether her party hands out particularistic benefits (Stokes, 2005). For these voters, the political context in which elections take place may not matter as much as for voters concerned with the actual policy performance of the government. The latter may better understand the complexities of politics and may therefore consider that presidents who have less political control over economic policies—for example, because their party is in a minority in the legislature—cannot be held equally accountable as presidents who have more discretion over the policies that their governments can implement.
The results of this article also have direct implications for the debate about the advantages of allowing presidents to run for consecutive terms (Carey, 2003). I show that the re-election rule is an important tool for enhancing electoral sanctioning. As long as elections are not conducted in an “uneven playing field” (Levitsky & Way, 2010), the advantages of allowing presidents to run for re-election seem to be clear on normative grounds. Thus, if the re-election provision enhances accountability when used in the right context, we should start thinking about the stage in the process of democratic consolidation at which a country can start contemplating its introduction.
To conclude, regarding the external validity of the findings presented in this study, it needs to be pointed out that because presidential candidates campaign and are elected separately from the legislature, presidentialism may provide additional incentives for voters to hold individual politicians accountable independently from their parties (Samuels & Shugart, 2010; Wiesehomeier & Benoit, 2009). Moreover, Latin America is a region in which several party systems are weak (Mainwaring, Bejarano, & Pizarro Leongómez, 2006; Mainwaring & Scully, 1995) and individual politicians are very influential. Yet, there are no theoretical reasons to expect that the connection between the outgoing head of government and the candidate of the incumbent party would not matter for electoral accountability in parliamentary and mixed democracies as well. In fact, the growing literature on the personalization of politics in Western parliamentary systems shows that leader images have an impact on the vote independently of party images in countries such as the United Kingdom (Andersen & Evans, 2003) and Canada (Johnston, 2002). Thus, if voters in established parliamentary systems also take candidate-level factors into account when casting a vote, it is likely that they will assign responsibility for past performance differently depending on the identity of the candidate of the incumbent party.
Footnotes
Acknowledgements
The author is grateful to Piero Stanig, Simon Hix, Francisco Panizza, John Carey, Ken Benoit, Scott Desposato, Francisco Cantú, Ryan Carlin, Randy Stevenson and the anonymous reviewers for comments on previous versions of this article. All remaining errors are the sole responsibility of the author.
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.