Introduction
Why have moderate electorates elected and reelected leftist governments in Latin America over the last twenty years? Data from the Comparative Study of Electoral Systems (CSES) show that the average Latin American voter in five countries during twelve electoral processes is located at 5.8 on the 11-point scale representing the classic left-right continuum. In only two out of twelve elections (for Uruguay in 2009 and Chile in 2005) was the average voter slightly skewed to the left of the spectrum (4.7 and 4.8, respectively), while in the remaining ten elections voters were slightly skewed to the right. For instance, in three consecutive elections registered by the CSES in Brazil (for 2002, 2006, and 2010), the average voter was located at the 5.7 of the scale. In Mexico three consecutive elections (for 2000, 2006, and 2012) show a skewed electorate at 6.7 of the ideological scale. Other scholars using surveys conducted by Latinobarometro (Reference Arnold, Samuels, Levitsky and RobertsArnold and Samuels 2011) and LAPOP (Reference Zechmeister, Kitschelt, Hawkins, Luna, Rosas and ZechmeisterZechmeister 2010) have reached basically the same conclusion: Latin American voters are moderate and, to some extent, slightly skewed to the right. How these conditions have led to the electoral success of the left is not evident.
Scholars analyzing voting behavior have primarily relied on the classic median voter theorem as one of the most powerful explanatory instruments in studies of party competition (Reference DownsDowns 1957; Reference Hinich and MungerHinich and Munger 1994; Reference CoxCox 1990). In this theorem, the median voter should always prevail regardless of where the median is located on the policy space. Assuming a normal distribution of voters, the basic theoretical expectation is that a moderate median voter should produce moderate parties in office. Instrumentally, vote-seeking parties will locate close to the position of the median voter to win elections. Empirically, however, the median voter theorem has been under question, beginning with the US case, where a moderate median voter sharply contrasts with highly polarized elections and partisan behavior in Congress (Reference Fiorina and AbramsFiorina and Abrams 2008; Reference Abramowitz and SaundersAbramowitz and Saunders 2008). Additionally, the last two decades of elections across Latin American countries has also left a picture far from moderation, even in the presence of a moderate median voter (Reference Arnold, Samuels, Levitsky and RobertsArnold and Samuels 2011).
Scholars have explained the success of the left in Latin America in different ways. Some authors have argued for a combination of democratic expansion and structural factors related to poverty, income distribution, and social exclusion, and how neoliberal policies drove voters to punish right-wing parties in office (Reference Baker and GreeneBaker and Greene 2011; Reference Blanco and GrierBlanco and Grier 2013; Reference Debs and HelmkeDebs and Helmke 2010; Reference ClearyCleary 2006; Reference Beasley-Murray, Cameron and HershbergBeasley-Murray, Cameron, and Hershberg 2009; Reference ArditiArditi 2008; Reference CameronCameron 2009; Reference CastañedaCastañeda 2006; Reference PanizzaPanizza 2005, Reference RobertsRoberts 2014; Reference QueiroloQueirolo 2013; Reference Weyland, Madrid and HunterWeyland, Madrid, and Hunter 2010; Reference MorganMorgan 2011; Reference SeawrightSeawright 2012). Others have also pointed out the existence of international factors and the demonstration effect produced by some leading cases like Venezuela, Bolivia, and Ecuador (Reference Levitsky and RobertsLevitsky and Roberts 2011; Reference RobertsRoberts 2007; Reference CastañedaCastañeda 2006). However, while these explanations have contributed to understanding the success of the left at the polls, they underestimate the scope of political agency involved in the strategies followed by partisan elites to reach power.
In contrast to existing studies, this article argues that the electoral success of the left has taken place through a clear partisan provision of programmatic cues to voters. We rely on a directional model in the spatial modeling tradition to demonstrate that left-wing parties have won on left-wing programs (Reference Rabinowitz and MacdonaldRabinowitz and Macdonald 1989), rather than a classic proximity model that predicts the predominance of moderate parties in elections. The argument and empirical evidence indicate that the strategy of successful parties does not always attend to the preferences of the median voter. Comparative and case-oriented data coming from the CSES for Brazil, Chile, Mexico, Peru, and Uruguay provide empirical support to our main argument.
This article makes two main contributions. First, while scholars have suggested different explanations of the electoral success of the left in Latin America, there are no attempts based on the spatial theory of voting. Although this article does not provide a new theoretical framework, it contributes to those scholars interested in new theory building by making use of well-known theories of voting behavior. By focusing on how voters react to the supply of programmatic options, we refocus the discussion on the electoral success of the left beyond those contributions that assess variations of left-wing parties in the region (Reference CastañedaCastañeda 2006; Reference Weyland, Madrid and HunterWeyland, Madrid, and Hunter 2010; Reference Levitsky and RobertsLevitsky and Roberts 2011). More specifically, while the predominant discussion of “social-democratic” and “populist” versions of the left are relevant for understanding the policy preferences and regime attitudes marked by different types of left-wing parties, those studies are unable to understand the electoral strategies of parties and voters during elections. Second, some elections in countries known for lacking stable programmatic linkages seem to indicate that voters are also able to provide clear programmatic mandates. More specifically, we demonstrate how the 2002 presidential election in Brazil shows a clear programmatic commitment of voters, electing a leftist party for the first out of four consecutive victories. This programmatic commitment fades away by 2010, when voters based their preferences on valence issues such as government performance rather than the programmatic cues that drove them to the polls in 2002. Arguably, the factors that account for the first electoral success of the left in Brazil are not necessarily the same that those behind further electoral victories.
Ideological Voters in Latin America?
The third wave of democratization has generated a large number of studies on party competition in Latin American politics (Reference DixDix 1989; Reference JonesJones 1995; Reference Mainwaring and ScullyMainwaring and Scully 1995; Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010; Reference LunaLuna 2014; Reference Carlin, Singer and ZechmeisterCarlin, Singer, and Zechmeister 2015; Reference RobertsRoberts 2014; Reference LupuLupu 2015). Yet, while these works rely on a variety of theoretical perspectives, they have failed to consider directional models of voting. It is common to rely on the classic proximity or Downsian model, but most if not all of these references are reduced to the use of a basic logic or parlance rather than a full application and measurement of spatial theory (see Reference León GanatiosLeon Ganatios 2013 for an exception). Arguably, this limited expansion of spatial models is largely based on the presumption that programmatic linkages between voters and politicians are exceptional in Latin American politics. As spatial models assume programmatic linkages, the lack of clear ideological preferences among voters underestimates the utility of these models to account for electoral behavior (Reference RuthRuth 2016; Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010).
This presumption is complemented by the prevalence of other forms of political exchange before and after the third wave of democratization (Reference LunaLuna 2014). Important segments of the electorate are mobilized to vote not simply according to their policy preferences but rather according to different forms of particularistic exchanges and charismatic linkages (Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010). These patterns of political behavior in numerous Latin American countries have been largely driven by elites without any interest at building party organizations on the basis of stable and clearly identifiable programmatic structures (Reference Levitsky, Loxton, Van Dyck and DomínguezLevitsky et al. 2016; Reference MainwaringMainwaring 1999). As a rule of thumb, where particularism and charismatic leadership prevails, voters disregard programmatic cues when making their voting decisions. If this is the case, spatial models lack fertile ground to account for voting behavior in Latin America.
However, there are at least two main reasons to question the idea that Latin American voters have no programmatic commitments. First, if voters were nonideological, important portions of the electorate would not be able to place themselves in the classic left-right continuum (Reference RuthRuth 2016). Yet longitudinal data challenge the argument against the lack of ideological commitment of voters and the inappropriateness of using left-right scales in the study of parties in the region (Reference QueiroloQueirolo 2013). Beyond the cases of Uruguay and Ecuador, with a maximum of 92 percent and a minimum of 71 percent, respectively, 82 percent of Latin American voters are able to locate themselves in the ideological spectrum during the period 1995–2010. This average is far from being in contrast to some of the most stable and developed democracies like Germany or Sweden (for 1981–2008), with 85 percent and 91 percent, respectively (see Eurobarometer). Furthermore, even in countries with noninstitutionalized party systems like Peru, 88 percent of the population is able to express ideological identification.
Second, if Latin American voters had no programmatic commitments, we should not expect any congruence between the average ideology of party supporters and the party’s actual ideology. However, Figure 1 shows that congruence between the voter’s ideology and their location of candidates on the ideological spectrum correlates at 0.77. Thus voters seem to know not only what their preferences are but also where the parties and candidates stand on the supply side of politics. Moreover, Figure 1 also discards the possibility that voters moved to the left (or even right) because of the lack of moderate options in most party systems, as there is an important concentration of candidates around the center and center-right of the spectrum. As a matter of fact, moderate voters have had an important set of options in most elections across Latin America. So why have they supported the left?
Voter and candidate ideological congruence in Latin America.
This question becomes even more intriguing as the rule in the region reveals a low level of leftism, as can be observed in cases like Argentina or Chile (Reference Arnold, Samuels, Levitsky and RobertsArnold and Samuels 2011). In Brazil, another case where the left won four consecutive presidential elections between 2002 and 2015, only about 20 percent of the electorate has leftist ideological preferences. Furthermore, in Costa Rica, where only 9.8 percent of the electorate reveal leftist preferences, the Partido de Acción Ciudadana (PAC), a moderate-left party, reached the presidency in 2014 and 2018. That is in a country with consistent moderate preferences and the largest share of voters with rightist ideological preferences in Latin America. What conditions favored the orientation of voters toward leftist parties in the region?
We argue that the classic proximity model predicts a centripetal orientation of parties that does not correspond with the electoral success of the left over the last two decades in Latin America. Instead we rely on a directional model to support the idea that leftist parties followed a rather centrifugal orientation, signaling clear cues to voters about their location on the policy space. Thus, in facing a predominantly moderate median voter, left-wing parties opted to seek a strategy of providing clear ideological cues rather than diluting their preferences with moderate policy stances.
A Sketch of Spatial Models of Voting
The emergence of directional models of voting provoked an intense debate around the ability of different spatial models to predict the electoral strategies of voters and parties in developed democracies (Reference Rabinowitz and MacdonaldRabinowitz and Macdonald 1989; Reference Rabinowitz, Macdonald and ListhaugRabinowitz, Macdonald, and Listhaug 1991; Reference IversenIversen 1994; Reference WestholmWestholm 1997). Until the late eighties, the median voter theorem largely dominated the spatial theory of voting, entailing the quintessence of what the literature calls the proximity model of voting (Reference HotellingHotelling 1929; Reference DownsDowns 1957; Reference BlackBlack 1958). This model assumes that partisan elites have no policy preferences and that parties maximize their vote by locating themselves close to the position of the median voter, which in a normal distribution in the policy space is right at the center of the spectrum. The classic Downsian model is a proximity model, to the extent that voters maximize (or minimize) utility when they move closer (or away) from their own position. Thus, when parties face an electorate whose distribution is symmetrical around the location of the median voter, getting closer to this position becomes a dominant strategy.
Directional models, in contrast, hold that voters do not maximize their utility by voting for the most proximate party. Two main steps compose the rationale of directional voting. First, voters are able to discern the direction that policies have to take. Then, voters will be attracted by those parties that match such direction and particularly by those parties displaying the most intense signals in that direction. In this way, utility comes not from the closer proximity of the voter to a certain candidate but from the direction and intensity of the voter’s policy preferences. Unlike proximity models, directional voters located at the center of the distribution are not necessarily moderate but are voters without a major intensity within that preference in the left-right continuum. Thus, the center is understood as a neutral point in the scale.
Notice that the distinction between proximity and directional models has been virtually unrelated to prospective and retrospective theories of voting (Reference FiorinaFiorina 1981; Reference MacKuen, Erikson and StimsonMacKuen, Erikson, and Stimson 1992). The main reason for this divorce is the different nature of these two sets of accounts of voting behavior. Prospective and retrospective models are largely based on egotropic and sociotropic considerations of the recent past or future condition of individual citizens, as well as other behavioral expectations based on gender, race, or income, among other possible factors. Instead, proximity and directional models depart from the assumption that citizens vote for parties or candidates on the basis of policy or ideological considerations, such as how near a certain party is to the policy preferences of voters (in proximity models) or which party provides the clearest ideological cues within the preferred side for the voter (in directional models). Therefore, the difference between the two sets of explanations revolves around the role of ideology in explaining voting behavior (Reference Adams, Merrill and GrofmanAdams, Merrill, and Grofman 2005).
Assuming a normal distribution, the preferences of moderate voters will always prevail under the proximity model. However, the expectations of ideological positioning in proximity and directional models are completely opposite. While the proximity model expects a location of parties and candidates close to the center, the directional model expects parties to locate far from the ideological center in order to attract and mobilize voters to the polls. Thus, whereas proximity predicts moderation, directionality predicts polarization. However, the expectation of directional models that voters will feel attracted by parties that provide clear policy preferences has a fundamental limitation. So far, we should expect that candidates at the extremes would be the most attractive for ideological voters, which implies that radical parties would have the higher chance of winning. Yet the directional model poses a limit to this tendency with the region of acceptability.
The region of acceptability suggests that citizens will punish extremist candidates, rewarding parties located within a region that poses a limit to extreme polarizing strategies. This region received no major theoretical elaboration by Rabinowitz and Macdonald (Reference Rabinowitz and Macdonald1989) and has received criticism by several scholars as the Achilles heel of their directional model (see Reference IversenIversen 1994; and Reference WestholmWestholm 1997). To keep this important concept, we argue that the region’s boundary is set by strategic voting considerations made by individual voters. As most voters know in advance their candidates’ electoral fates, strategic voters will always seek to get rid of radical candidates in favor of those with more chance of winning. These candidates are not far from the extreme left or right but still provide clear policy or ideological stances to their voters. Assuming a normal distribution of voters, strategic voting from both directional and proximity-oriented voters make most extremist candidates electorally unviable. Thus strategic voting does not collide with the argument posed by Rabinowitz and Macdonald (Reference Rabinowitz and Macdonald1989) but rather provides a plausible addendum to the region of acceptability.
Our expectations contend that parties may succeed by attracting the median voter using different programmatic strategies. This raises a crucial theoretical and empirical point. Theoretically, the rationale behind directional and proximity models of voting is different at the individual level of analysis. This is the reason why the simultaneous presence of both kinds of voters within a single electorate is plausible. In other words, electorates characterized by the presence of voters who cast their votes either on a directional or proximity basis. From an empirical point of view, the two models of voting are testable through our utility functions specified in the next section, which use the same information to calculate the magnitude of each voting model. Which one could define an election is an empirical question that we do not answer in this article but that remains open to future research.
Parties and voters may have different programmatic motivations over time. Although parties change their platforms and the main focus of their policy interests, these complex institutions tend to be reluctant to make programmatic change. However, even facing the same program across different elections, voters may not have the same motivations to vote for the same party. Therefore, capturing within country variation in terms of proximity and directionality in voting behavior will not only help us to unveil the voter’s motivations to support certain parties but also show how stable those preferences are over time. In addition to cross-sectional data for a sample of Latin American countries, our empirical section provides evidence of temporal variation between the national elections of 2002 and 2010 in Brazil. We expect to find directional models to have a stronger explanatory power compared to proximity models.
Data and Methods
Our empirical analysis uses data from the Comparative Study of Electoral Systems (CSES). This survey was conducted in Brazil (in 2002, 2006, and 2010),Footnote 1 Chile (in 2005 and 2009), Mexico (in 2000, 2006, and 2012), Peru (in 2000, 2001, and 2006) and Uruguay (in 2009), which allows us to pool data with eleven elections in five countries (see Tables A1–A2 in Online Appendix for descriptive statistics). Although this sample is relatively small for the whole region, the pool of cases provides a reasonable variation in several ways. For instance, while Chile and Uruguay have been traditionally considered as two highly stable party systems, Peru has turned into a fluid party system with minor signs of stability. The same can be observed in terms of the scope of programmatic orientation of parties, where both Chile and Uruguay have been observed as mainly programmatic, several important parties in countries like Brazil, Mexico, and Peru are well known for being strong providers of different forms of particularistic goods and services to voters (Reference MagaloniMagaloni 2006; Reference AmesAmes 2001; Reference Muñoz, Dargent, Levitsky, Loxton, Van Dyck and DomínguezMuñoz and Dargent 2016). Similarly, the strength of party organizations also varies a great deal. While Mexican parties display strong levels of party organization, several parties in Brazil and especially in Peru show high levels of personalism and the lack of investment in party organizations (Reference Levitsky and CameronLevitsky and Cameron 2003; Reference MainwaringMainwaring 1999).
Dependent variable
Given our expectations, it is important to note that most if not all works testing the explanatory power of directional and proximity models use candidate evaluation as a dependent variable, rather than vote intention (Reference Rabinowitz and MacdonaldRabinowitz and Macdonald 1989). We capture our dependent variable, the voters’ party or candidate evaluation, with a thermometer that varies from a minimum of 0 to a maximum of 10. We use these data to construct our dependent variable, which is the evaluation of the most voted left-wing candidate, for two reasons. First, we are interested in explaining the success of leftist candidates only. Second, although most countries in our sample have minor leftist parties that may or may not be part of governmental coalitions, most of these parties belong to the radical left and can be considered electorally small.
Arguably, the gap between candidate or party evaluation and vote intention is not irrelevant. First, strategic considerations may account for a gap between the evaluation of a certain candidate and the vote intention of respondents. Voters are usually informed about the electoral chances of different candidates, driving strategic voters to vote for candidates with higher chances of winning. Second, the gap between candidate evaluation and vote can be mediated by other factors like side payments and a long list of material benefits in exchange for the vote. Some voters may like the policy preferences of a certain candidate, but their ties to another with which they hold an exchange of political support for goods or access to specific policies may drive them to vote for the latter. Furthermore, there is always a span of time between the survey’s fieldwork and the act of voting. Many voters change their preferences in short periods of time, especially in party systems where partisanship is low or in fluid systems where party supply varies substantially between elections. All said, given that our research question explicitly refers to voting behavior, we also provide results for vote intention as ancillary information to this article. Our results using vote intention remain highly robust to our findings with candidate evaluation under different models and specifications (see Table A3 in Online Appendix).
Independent and control variables
Following Rabinowitz and Macdonald (Reference Rabinowitz and Macdonald1989), the assumptions about the rationale of voting under directional and proximity models can be translated into two different utility functions. In the case of proximity models, utility is calculated as the distance of voter i regarding party k, within a Euclidean space defined by an 11-point scale representing the left-right ideological spectrum, where 0 is defined as extreme left and 10 is the extreme right. As can be seen in equation 1, xij is the position of voter i on issue j, akj is the average position of candidate k on issue j, and ni is the number of issues on which voter i locates herself and party k, which allows us to calculate the same function for n-dimensional spaces.
Unlike the proximity model, the utility function for the directional model is calculated as the simple scalar product made by voter i for party k (Reference Rabinowitz and MacdonaldRabinowitz and Macdonald 1989), using the same ideological scale discussed above. In the following notation, both xij and akj remain identical to equation 1, and
is the neutral point of the scale, corresponding to 5 on the 11-point scale.
The debate around directional and proximity models has brought controversy to the way to estimate each utility function (see Reference WestholmWestholm 1997; Reference IversenIversen 1994; and Reference Macdonald, Rabinowitz and ListhaugMacdonald, Rabinowitz, and Listhaug 1998). In particular, there is controversy around the way to measure the spatial position of the candidate being evaluated by each voter on the ideological scale. The interpersonal comparison is based on the average of all respondents, assuming that all candidates have a unique location in the space (as the average perception), known by all voters in the sample (Reference Rabinowitz and MacdonaldRabinowitz and Macdonald 1989; Reference Macdonald, Rabinowitz and ListhaugMacdonald, Rabinowitz, and Listhaug 2001). The intrapersonal comparison, however, is based on individual perceptions, assuming that the position of candidates on the scale is not a constant but a value that varies across individuals.
The difference between the aforementioned measures of inter- and intrapersonal comparison is not trivial. The original directional model uses interpersonal comparisons, because the approach requires a unique position of candidates against which the position of every individual is compared in the scale. That is, the best way to measure the candidate location on the ideological scale is the average value assigned by voters. However, both Westholm (Reference Westholm1997, 868–869) and Iversen (Reference Iversen1994, 161) have criticized this measurement, as it is not likely that all voters share the same opinion around the location of candidates and that such a point is well represented by the average perception of all voters. In their defense, Macdonald, Rabinowitz, and Listhaug (Reference Macdonald, Rabinowitz and Listhaug2001) argue that intrapersonal comparisons produce biased estimations in favor of proximity models, given that voters tend to locate their most-preferred candidates close to their location (as an assimilation bias) and their less-preferred candidates in more distant locations (as a contrasting bias). For this reason, and following Rabinowitz and Macdonald (Reference Rabinowitz and Macdonald1989), our use of a directional model to account for the success of the left relies on interpersonal rather than intrapersonal comparisons, which have been left as ancillary information to this article (see Tables A4 and A6 in Online Appendix for results using intrapersonal comparisons for our panel data and Brazilian case).
We calculate the utility functions with a scale representing a single dimension of politics. We do not deny that there are other dimensions of politics in one or more countries in our sample. However, left-right remains as the single most important dimension of politics in the region, even in countries where other dimensions might be relevant for important portions of the population. There are at least three reasons to use the left-right continuum. First, the CSES questionnaire does not collect information on several policy issues on which we could calculate the utility functions. Second, following extant research (Reference Altman, Luna, Piñeiro and ToroAltman et al. 2009; Reference Wiesehomeier and DoyleWiesehomeier and Doyle 2012), there are multiple correlations in our dataset between different policy issues and the ideological dimension of politics measured with the left-right continuum.
We added several control variables to our models to test the validity of alternative explanations. In particular, we added political, economic, and sociodemographic factors. First, following Rabinowitz and Macdonald (Reference Rabinowitz and Macdonald1989) we control for closeness to a party as a proxy of partisanship. We expect that when a citizen feels closer to a party (0 = respondent does not report closeness to any party; 1 = not very close; 2 = somewhat close; 3 = very close) the evaluation of that party’s candidate will be more positive. Second, we also control for different income levels, measured on a 5-point scale of quintiles, assuming that the lower the quintile the higher the probability to vote for the leftist candidate. Third, following Campbell and collaborators (Reference Campbell, Converse, Miller and Stokes1960), we also control for education levels (measured with an 8-point scale going from low to high), expecting higher levels of education to increase the probability of voting for a leftist candidate.
Further, we estimate additional regression models to capture the scope of retrospective voting, using government performance. This variable ranges from a minimum of 1 (very good job) to a maximum of 4 (very bad job), thus it was recoded in inverse order to facilitate the interpretation of coefficients. Since this variable does not capture whether the incumbent belongs to the left or the right, we also introduced incumbency as a dummy variable coding cases as 1 for elections held under left-wing governments and 0 for all other cases (see Table A1 in Online Appendix). In this way, we classified four out of eleven elections as having left-wing incumbents (Brazil 2010, Chile 2005 and 2009, and Uruguay 2009), while the remaining seven were equal to 0. Subsequently both government performance and incumbency were interacted to measure the effect of government performance when the incumbent belongs to the left.
Results
Table 1 contains the results of linear regression models with fixed effects by country. While Hausman tests support the use of fixed effects, we also estimated all models using random effects with very similar results (see Table A7 in Online Appendix).Footnote 2 We ran separate regressions for directional and proximity models for the most voted left-wing candidate, plus a set of basic available controls. Unfortunately, although some of these controls are important to test the validity of our main argument, our purpose to maximize the number of countries and elections collided with a larger number of controls in our models. Thus, to reach more leverage in our empirical analysis, our next section provides more nuanced evidence using the Brazilian case, including valence issues and other available controls.
Explaining the electoral success of the left in Latin America (fixed effects regression with standardized coefficients).
Note: Standardized beta coefficients; robust standard errors in parentheses.
* p < .05; ** p < .01; *** p < .001.
As can be observed in Model 1, both directional and proximity models achieve statistical significance at 99 percent of confidence. The use of standardized coefficients to ease comparability allows us to observe that the size of the coefficient for the directional model is about 50 percent bigger than the proximity model. Model 2 adds government performance, incumbency, and the interaction term government performance × incumbency to the specification of Model 1. In general, Model 2 displays very similar results. First, although both directional and proximity models remain statistically significant, the size of the coefficient of the directional model is 93 percent bigger than the proximity model. Second, both income and partisanship achieve statistical significance in Model 2, but education does not. Third, government performance is statistically significant under directional and proximity models, indicating that reputable governments have a negative impact on the evaluation of leftist candidates. However, the sign of the coefficient for the interaction of government performance with the ideological orientation of incumbents indicates that candidate evaluation is positive when left-wing government performance is high. As a rule of thumb, valence issues like government performance are more likely to go hand in hand with the proximity model in advanced democracies, but this does not seem to be the case in our sample of Latin American countries.Footnote 3
Figure 2 shows marginal effects of both directional and proximity spatial utility, coming from Models 1 and 2 in Table 1. The figure shows that a marginal increase in spatial utility produces a larger impact on the evaluation of the most voted leftist candidate under our directional model, compared to the proximity model. As for Model 1, an increase in one standard deviation in the directional utility produces an increase of 0.64 points on the evaluation of left-wing candidates and a decrease of 0.42 points in the case of proximity. As for Model 2, an increase of one standard deviation in the spatial utility of the directional model produces an increase of 0.51 points on the evaluation of the most voted leftist candidate and a decrease of 0.31 points in the case of proximity. In line with our theoretical expectations, the substantive effects at the comparative show that the directional model outperforms the proximity model, suggesting that voters were mainly motivated by clear programmatic cues from leftist candidates.
Marginal effects of directional and proximity models on leftist candidate evaluation.
The Brazilian Case
The Workers Party, or Partido dos Trabalhadores (PT), is one of the most successful left-wing parties in Latin America’s third wave of democratization. Born in 1980 as a typical mass-based party (Reference MeneguelloMeneguello 1989; Reference KeckKeck 1992; Reference HunterHunter 2010; Reference SamuelsSamuels 2004; Reference Levitsky, Loxton, Van Dyck and DomínguezLevitsky et al. 2016; Reference Van DyckVan Dyck 2016), the PT emerged as a platform for the labor movement and other social organizations and intellectuals fighting against the authoritarian regime installed in 1964. From the outset, its programmatic base featured radical rather than moderate socialist ideology but was clearly linked to the expansion of social rights and participatory democracy (Reference Samuels, Zucco, Levitsky, Loxton, Van Dyck and DomínguezSamuels and Zucco 2016). Ironically, the fall of real socialism during the early nineties led the PT toward the left of the spectrum rather than the center. This move proved to be electorally inefficient and the party lost three consecutive elections against the right-wing populist candidate Collor de Mello in 1990 and the moderate Fernando Henrique Cardoso in 1994 and 1998.
Nevertheless, there is a wide consensus among scholars that the PT deliberately moved toward the center to win the presidential race of 2002. As Samuels (Reference Samuels2006) argues, both endogenous and exogenous factors drove voters to support the PT, electing Luiz Inácio Lula da Silva (Lula) for his first term in office. Among the many exogenous factors stressed in the literature, none has gained more attention than the implementation and social impact of neoliberal policies during the nineties (Reference Hagopian, Gervasoni and MoraesHagopian, Gervasoni, and Moraes 2009). For this explanation, the electoral success of the PT was driven by voters affected by the structural adjustment and several reforms entailing state retrenchment in the economy, in a country largely based on state-led economic development.
Such structural and favorable conditions in the environment did not help the PT to crystalize electoral victories in 1994 and 1998. During this period, the Partido do Movimento Democrático Brasileiro (PMDB) and particularly the Partido da Social Democracia (PSDB) led by Cardoso started to move toward the right of the spectrum, leaving space for the PT’s move toward the center. It was in this context that the PT followed a process of programmatic change driven by the grass roots and partisan elites, and particularly by its main leader and presidential candidate. This process of ideological moderation was mainly endogenous, to the extent that the party organization itself was responsible for strategizing a change for the sake of winning the presidency.Footnote 4 The outcome is unexpected, as the rule of thumb shows that grassroots organizations behave as the programmatic reservoir with which party leaders and party adherents have to deal with to strategize in elections (Reference MayMay 1973; Reference NorrisNorris 1995). Despite the PT’s moving toward the center (3.6 on the ideological scale in 2002), scholars agree that Lula won its first presidential election of 2002 under a left-wing platform and voters perceived this programmatic outlook.
Table 2 shows OLS results for candidate evaluation during the 2002 elections in which Lula and José Serra from the PSDB were the main presidential race contenders. Unlike our cross-sectional data analysis in Table 1, this section of our data allows us to use two valence issues (corruption and government performance) as well as other important controls such as previous vote in elections, partisanship, and dummy variables to account for important regional effects in a federal country. As can be observed, the proximity model only achieves statistical significance for Serra. In contrast, the directional model performs very well with both candidates, with a larger coefficient compared to the proximity model.Footnote 5
Explaining the electoral success of the left in Brazil 2002 (OLS estimates).
Note: Standardized beta coefficients; standard errors in parentheses.
* p < .05; ** p < .01; *** p < .001.
Other independent variables are also statistically significant, particularly for Lula under both directional and proximity models. First, in line with our results in Table 1, affluence strongly affects candidate evaluation, such that higher-income voters have a negative evaluation of Lula. Second, corruption seems to have a large impact on both directional and proximity models for Lula in 2002. However, the results for this variable seem counterintuitive, because the lower (the higher) the level of perceived corruption the better (the worse) the evaluation of Lula. The same is true for Serra, though “corruption” for this candidate is not statistically significant. Arguably, voters who perceive widespread corruption seem to punish all candidates, and vice versa, voters who perceive lower levels of corruption tend to be more indulgent with politicians.
Government performance is also statistically significant across models for both candidates, so that the worse the government performance of president Cardoso, the better the evaluation of Lula; the opposite is true for Serra. Fourth, unlike our comparative panel data, partisanship is statistically significant, indicating that voters who feel close to a party are more likely to have a positive evaluation of that party’s candidate. This finding is in line with Samuels (Reference Samuels2006) and with Lupu’s (Reference Lupu, Levitsky, Loxton, Van Dyck and Domínguez2016) assessment of partisanship in Brazil and particularly the role played by the PT during the last three decades. Finally, the variable vote in previous election captures the larger coefficient and achieves statistical significance across models. The sign of the coefficient indicates that those who voted for Lula in 1998 were predisposed to reiterate their vote for the same party, as their evaluation was largely positive and the opposite is was true for Serra: those who voted for the PT in 1998 have, on average, a negative evaluation of Serra.
Table 3 displays almost the same models for the 2010 elections, in which Dilma Rousseff from the PT ran against the PSDB candidate José Serra. All models are identical to those shown in Table 2, with the exception of corruption, which was not available in the questionnaire for 2010. The results for this election are substantively different from those observed in 2002. First, the coefficient for the directional model of the winning candidate is not statistically significant. Rather, this election shows that the candidate who lost the election (Serra) performs better under the directional model. Therefore, the results show that Serra was perceived by directional voters as the candidate with the more clearly polarized outlook. Second, the proximity model is statistically significant for both the winner and loser of the presidential race. However, the sign of the coefficient for Rousseff is not as expected and, in fact, its interpretation turns out to be incoherent. With the exception of this coefficient, the results are consistent with the ideological location of the two candidates during the 2010 elections. While Serra was on average located at 6.55 on the ideological scale, Rousseff was placed at 4.82. Therefore, directional voters clearly saw Serra as a more extreme candidate, while Rousseff turned toward a moderate location to capture proximity voters, which also allowed her to capture the benefits of a highly positive government performance attributed to the PT in office.
Explaining the electoral success of the left in Brazil 2010 (OLS estimates).
Note: Standardized beta coefficients; standard errors in parentheses.
* p < .05; ** p < .01; *** p < .001.
Table 3 displays other important results that help to account for the electoral result of 2010. First, education becomes statistically significant with a negative sign under the directional model with Rousseff, showing that educated voters (unlike what we observed in 2002) are not supportive of the PT candidate. Second, partisanship or closeness to a party remains statistically significant across all specifications. Voters who feel closer to a party are more likely to support Rousseff, while those who do not are more prone to support Serra. Arguably, closeness to a candidate could be understood as a measure of militancy, and Rousseff still seemed to capture this tradition among left-wing voters. Last but not least, both government performance and vote in previous election remain highly significant, with the largest coefficients across models. In the case of government performance, a positive evaluation of Lula’s government has a positive impact on Rousseff’s evaluation, and the opposite is true for Serra. Finally, vote in previous election shows that the incumbent party is still able to retain captive voters, which helped the PT to stay in office.
In a nutshell, comparing the results for 2002 and 2010 yields at least two important implications. On the one hand, while the PT led by Lula was able to capture the attention of ideological voters until 2002, it was not that successful at attracting this profile of voters in 2010. During this election, voters were less persuaded by the PT’s programmatic outlook led by Dilma Rousseff, being more attracted by the PSDB candidate (Serra) and his right-wing coalition of supporting parties. On the other hand, and in line with other findings in the literature that stress the scope of partisanship achieved by the PT, Rousseff retained captive voters and largely benefited from a positive evaluation of Lula’s government (Reference SamuelsSamuels 2006; Reference Samuels, Zucco, Levitsky, Loxton, Van Dyck and DomínguezSamuels and Zucco 2016). These results are consistent with other studies suggesting a valence advantage for the incumbent that allowed Rousseff to adopt more centrist strategies (Reference Cho and EndersbyCho and Endersby 2003; Reference GrosecloseGroseclose 2001; Reference Ansolabehere and SnyderAnsolabehere and Snyder 2000). Both government performance and vote in the previous election became crucial in 2010, as the left was evaluated by its achievements and failures in office, rather than its programmatic stances. In sum, the electoral success of the PT was largely driven by left-wing programmatic motivations in 2002 but not in 2010.
Conclusions
Based on spatial models of voting, this article contributes to accounting for the conditions that favored the electoral success of the left in Latin America over the last two decades. Unlike other accounts based on structural factors like the economy or the social costs of economic reforms, we focused on the programmatic strategies of competing parties in elections. Unlike what the median voter theorem predicts, we argued that the programmatic strategy of successful parties does not always reach the preferences of the median voter. As in most cases around the world, the median voter is a moderate one. Yet several parties in our sample are located away from the center of the ideological spectrum in order to win, at least in their first election. This result is in line with previous findings in the literature on spatial models of electoral behavior in developed democracies (Reference Adams and MerrillAdams and Merrill 1999).
Successful Latin American left-wing parties seem to have privileged a strategy of locating themselves away from the center, even facing moderate electorates. Such a strategy seeks to provide clear programmatic cues to the electorate regarding the direction that policies will follow instead of minimizing the distance between parties and the median voter. By implication, the electoral success of the left does not necessarily come from moderation but a deliberate positioning at the left of the spectrum. This presumption receives empirical ground by comparing the predictability of proximity and directional models of voting behavior across eleven elections in Latin America. In comparing models, our statistical tests show that voters vote not necessarily for the party or candidates whose location is closer to their position but for those who provide a set of clear and intense set of cues on the left-right continuum.
Arguing that left-wing parties won on left-wing programs does not reject the argument that these parties made an effort to moderate their programmatic outlook to capture at least part of a majoritarian moderate electorate. This was the case of the PT with Lula in Brazil or even Mujica in Uruguay, whose parties lost several elections competing against right-wing parties before winning their first presidential election. In general, there is an important consensus among country experts that most left-wing parties moderated their platforms to win elections. Yet, although both the PT and other left-wing parties made an effort to moderate their platforms, they never stopped being leftist parties, at least in their first electoral victory. The electoral success of the left is partly due to moderation as shown by our estimation of proximity models. At the same time, directional models across cases indicate that a significant portion of Latin American voters were predominantly induced to vote for options providing clear programmatic cues from the left.
An important implication of our statistical results is that they lead us to rethink the conventional wisdom of “ideological moderation” as the unique successful electoral strategy. Our results suggest that party leaders and the grass roots have a great opportunity in picking noncentrist strategies, even facing moderate electorates. If this is the case, the implications of this work exceed our contribution to accounting for the electoral success of the left, by helping future contributions to elaborate general arguments around the electoral success of programmatic strategies in Latin America. Arguably, the electoral success of noncentrist candidates in the region may help to account for not only the wave of left-wing governments but also a possible right-wing turn in the near future.
Additional File
The additional file for this article can be found as follows:
† Online Appendix. DOI: https://doi.org/10.25222/larr.466.s1, see pdf.
Open access
