Electoral volatility and political polarization in developing democracies: Evidence from Latin America,1993–2016

Introduction

The spread of democracy in the developing world has been one of the most interesting phenomena in the past 40 years. ¹ Around 40% of the world’s population, more people than even before, live in countries that hold free and fair elections. But the impetus for democratic progress has lost momentum, and numerous scholars and pundits have been quick to blame the increasing level of polarization for it. These claims are in line with extant research in political science and economics, linking polarization with a series of problems for democracy including breakdown (Huntington 1968; Linz 1978), the inability of governments to implement redistributive economic policies (Alberto et al. 1999; Rodrik 1999) or policy reforms (Mainwaring and Shugart 1997; Shugart and Carey 1992), and regime stability (Levitsky and Roberts 2011; Weyland et al. 2010). While the aforementioned studies provide important insights about the consequences of polarization, they offer no systematic explanation of its origins.

This is surprising because polarization has been on the rise since the early 2000s and, as illustrated by Figure 1, shows substantial variation across developing democracies. The substantial growth yet underlying variation of this phenomenon raises the following question: what explains the varying levels of polarization in developing democracies? In the realm of the literature on parties and party systems, some scholars have suggested that structural factors or socioeconomic processes affect voter alignments and de-alignments that trigger programmatic politics between clear ideological stances (Roberts 2013). Formal and empirical models of spatial competition use a similar heuristic to predict centrifugal and centripetal party competition (Adams et al. 2005; Downs 1957; Hinich and Munger 1994; Rabinowitz and Macdonald 1989; Schofield and Sened 2005) where the preferences of voters are exogenous. Further, scholars like Cox (1990) and Sartori (1976) posit that electoral rules provide incentives or constraints to seek a polarizing strategy, and recent empirical evidence shows how these factors affect polarization (Dalton 2021).

OPEN IN VIEWER

Figure 1.

Mean of party polarization across Latin American countries, 1992–2018.

The aforementioned studies are very insightful, yet they have a few drawbacks. To begin with, socioeconomic and spatial theories assume that voters’ preferences are exogenous, paying little attention to the possibility that position taken by party elites affects the level of polarization in the system. This is problematic for at least two reasons. First, extant literature on parties and party systems suggests that parties, as opposed to public opinion, are responsible for driving ideological polarization (Abramowitz and Saunders 2008; Adams et al. 2005; Adams et al. 2012; Iversen 1994; Mainwaring and Torcal 2003). Przeworski and Sprague’s (1986) influential piece argues that the preferences of party elites shape the preferences of voters by choosing their location in the policy space, and more recently Bullock (2011) and Sohlberg (2017) find that the positions taken by party elites influence voters’ preferences and ultimately behavior.

Second, many of the predictions of socioeconomic and spatial theories are not supported by empirical evidence when tested at the comparative level. For instance, there is no systematic evidence supporting the classic Downsian or proximity model (Adams et al. 2005; Iversen 1994; Zur 2021). Further, there is little empirical evidence linking socioeconomic factors with ideological polarization beyond advanced democracies (see, e.g., Dalton 2008, 2021; Knutsen 1998). Moreover, the effect predicted by institutional explanations of polarization does not hold when we observe evidence from the United States and several Latin American countries. That is, bipartisan systems are not necessarily less polarized than multiparty systems (Dalton 2021).

While our argument builds on supply-side or elite-driven explanations of polarization (Przeworski and Sprague 1986), we depart from them in two main ways. First, we particularly emphasize the role of the electoral setting in which parties compete for votes, which often is unstable in developing democracies. Specifically, we argue that parties use polarization to provide voters with unequivocal signals about their location in the policy space in contexts of high electoral volatility. As electoral volatility increases the risks of electoral survival, polarization becomes an efficient strategy to mobilize voters to the polls with clear programmatic cues. We therefore hypothesize that higher levels of electoral volatility lead to higher levels of ideological polarization, and particularly in cases where newcomers get into the electoral game. Given that party system change in such democracies is associated with the emergence of new political parties or fusions and fissions or existing parties, we further analyze the effects of party replacement—as opposed to stable party—volatility on polarization, and hypothesize that higher levels of replacements volatility engender polarization.

We test our main claim focusing on Latin America. Two reasons explain this choice. First, extant literature on democratization argues that polarization is one of the main features of Latin American politics in the long run. As such, polarization has had a negative impact on regime stability, particularly in contexts of democratic breakdown and transitions from authoritarian regimes. Second, other literature also has focused on the causes and consequences of electoral volatility and its impact on parties and party systems, where Latin America shows chronic problems of party survival and institutionalization (Cohen et al. 2018; Mainwaring et al. 2016: Powel and Tucker 2014; Roberts and Wibbels 1999).

The paper has implications for the literature on party behavior, party systems, and democratization. First, this research shows that parties in highly volatile systems have incentives to use polarization as a mobilization tool. As such, the results presented here not only allow us to better understand the behavior of parties and elites in developing democracies, but also to explain why parties and politicians make decisions that often seem at odds with the dominant prediction of the median voter theorem (Moraes and Luján 2020; Zur 2021). Second, this research sheds light on the determinants of a phenomenon that has not only increased over the past two decades in most developing democracies, but also has created instability and democratic reversals in some of those countries. In light of the new political developments in several advanced democracies where polarization has increased sharply, this manuscript also has implications for students of polarization in advanced democracies.

The paper proceeds as follows. We first present our theoretical argument, the two main hypotheses, and some anecdotal evidence. We then present our statistical methodology, data, variables, and empirical results. Next, we present the robustness tests. We conclude by discussing key implications from our findings and provide avenues for future research.

Theoretical argument and testable hypotheses

Extant research on party competition has been largely influenced by models derived from the median voter theorem (Downs 1957), which suggests that vote seeking parties should avoid positional risks by moving away from the location of the median voter (usually a moderate located at the center). Other formal theories of party polarization suggest that parties may ponder other policy and non-policy considerations such as partisanship and campaign resources when deciding whether they want to pursue a polarizing strategy or not (Schofield and Sened 2005). These studies are insightful, but they do not explain how the stakes involved in different elections may influence the party´s decision to move away from the center.

Empirical models, in contrast, have explained why parties may adopt a non-centrist strategy. ² For instance, Adams et al. (2005) suggest that parties can perfectly follow an off the center strategy of polarization without sacrificing a vote-maximizing strategy. In this case, partisan elites will prefer a location between the median voter and the position of core constituency voters of their own party. Directional models also offer an alternative (but policy-oriented) account for the location of parties away from a moderate median voter. As Rabinowitz and Macdonald (1989) put it, parties capture the attention of voters by providing clear policy signals to the electorate, which is moved by a primary orientation towards one the side of the spectrum and will be particularly attracted by those parties who provide the most intense signal (on that particular side chosen by the voter).

Away from pure policy factors determining the location of parties in the policy space, Schofield and Sened (2005) explain the parties’ decision to move away from the center using valence considerations. Specifically, they argue that non-convergent (or polarizing) policy choice is a rational vote-maximizing decision made by party leaders in response to the motivations of party activists. Thus, empirical and directional models of party competition allow us to better understand some of the factors that explain why parties abandon centrist strategies. Yet none of them considers how uncertainty—and the potential risks (i.e., party survival)—associated with different electoral settings may affect the parties’ decision to abandon the center and move towards a polarizing strategy to mobilize voters to the polls.

Empirically, patterns of party competition tend to be stable in developed democracies, where parties and party systems usually have deep roots in society. But the picture is different in developing democracies. Indeed, a sizable amount of literature has observed that weak institutions, a common feature of political systems in such states, are largely derived from political instability in its different forms (see, e.g., Lupu and Riedl 2013). Additional evidence suggests that party systems in new and developing democracies are far from being stable. Mainwaring et al. (2016), for instance, document that the mean electoral volatility in developing democracies is quite high compared to developed states and varies substantially across cases, ranging from a low 6.8 in Honduras to a high 60.3 in Ukraine. These findings are not trivial because extant research suggests that electoral volatility not only influences the nature of political competition, but also raises the stakes of the electoral game both in the short and the long run (Pacek et al. 2009).

In this context, electoral volatility may just be an electoral expression of a larger setting of uncertainty and instability in which party competition usually operates. In a number of developing democracies, parties frequently change their policy positions and even make explicit mandate violations once they reach office (Stokes 2001). Thus, it is plausible that electoral volatility may have an impact on the positioning and strategies of parties that compete in elections. In this fluid environment, party labels are likely to mean little to most voters who usually lack long-term commitments with determined parties and constantly change allegiances (Lupu 2016; Madrid 2005; Zielinski et al. 2005). Voter de-alignment, in turn, creates important changes in the preferences of the electorate, generating a great dispersion of the vote (Moser, 2001) and, consequently, a substantial increase in the level of competitiveness in elections (Lachat 2008), making it more difficult for elites to predict the results.

In addition, electoral volatility affects the incentives of politicians to perform electoral coordination, dissolve extant parties or even fuse with others to create a new brand (Cox 1997; Tavits 2006). When volatility is high, parties also lose the ability to predict the likelihood of success of competing parties with whom they can coalesce. Inability to coordinate may lead to an over-supply of candidates or parties and increase the chances of overlapping with others on the same policy position. Thus, by locating themselves in the same ideological position of their competitors, parties risk losing the votes even of their core supporters, increasing the probability of defeat.

A dynamic electorate, highly competitive elections, and inability to predict the prospects of rivals introduce a high degree of uncertainty and risk into the party´s decision over electoral competition. This is exacerbated by the fact that most parties in fluid party systems generally have anemic organizations and limited resources to mobilize voters to the polls (Hicken 2009; Mainwaring 1999; Mainwaring and Torcal 2003). Frequently, elites in volatile party systems are often confronted with the possibility of facing important—if not determinant—defeats at the polls that can be irreversible for their electoral survival. Therefore, the scope of electoral stability (or instability) at the party system level is a crucial contextual factor that affects the calculations of parties of whether sticking to their policy positions close to the median voter but without clear policy or programmatic cues, or mobilizing voters on a polarizing basis with clear policy stances.

The fear of losing votes drives parties to make electorally risky decisions to mobilize with a polarizing strategy that differentiate them from political rivals. When volatility is low, by contrast, elites face less challenges to predict the likelihood of winning elections and the long-term prospects of their parties. As the certainty about their electoral and long-term prospects increases, parties have incentives to be risk averse and stay put in their current programmatic or ideological positions. Therefore, polarization will be less in systems with low levels of electoral volatility. The preceding discussion thus leads to the following testable hypothesis:

H1

Higher levels of electoral volatility lead to higher levels of ideological polarization.

Thus far, electoral volatility has been treated as a systemic trait. While this level of analysis provides a general picture for most party systems, it comes with a potential drawback: the presumption that electoral volatility takes place among stable parties. Given that numerous party systems are subject to an intense flux due to the emergence of brand new parties, as well as fusions and fissions of existing parties, systemic volatility can gloss over two completely different phenomena. The systemic or aggregate measure using Pedersen’s (1979) index equates the supply and demand for political parties. That is, the aggregate measure does not discriminate between changes in the preferences of voters (as the demand side) from changes in the supply of political parties (Schmidt 2018). Therefore, aggregate measures of electoral volatility provide a good overall picture but perhaps incomplete, depending on the party system.

Recently, several scholars have analyzed different forms of conceiving electoral volatility from the demand and supply side of parties. The contributions of Cohen et al. (2018), Mainwaring et al. (2016), and Powel and Tucker (2014), for instance, capture the variation of volatility among established and newly competing parties, coming from completely new parties or fusions and fissions of existing party organizations. These two indicators give more leverage to students of party systems in regions featured by high levels of electoral volatility. Of course, the mere appearance or even the threat of new emerging parties may cause the fusion or coordination of new coalitions and alliances of established parties to compete against newcomers. This potential endogeneity problem may be a limitation to disentangle the difference in both types of volatility measures. Yet, discriminating between stable party volatility and party replacement volatility constitutes a notorious advance in the literature on party competition, especially in fluid party systems.

This raises an important question: how do stable and party replacement volatility affect polarization? Theoretically, it is possible that new vote seeking parties will maximize their chances to succeed or survive by getting close to the median voter location. Yet, this expectation colludes with empirical models and studies, which suggest that newcomers provide clear ideological signals away from the center or moderate positions where the median voter usually locates (Lupu 2016; Roberts 2013; Seawright 2012; Schofield and Sened 2005; Singer 2016; Zur 2021). Furthermore, anecdotal evidence from Latin America suggests that some of the most successful “newcomers” in the area have achieved substantial electoral success by locating away from the center (Moraes and Luján 2020). Among other examples, these cases include Alberto Fujimori in Perú, Hugo Chavez in Venezuela, Evo Morales in Bolivia, Rafael Correa in Ecuador, Jair Bolsonaro in Brazil, Nayib Bukele in El Salvador, and Andrés Manuel López Obrador in Mexico. ³ This leads us to state the following hypothesis:

H2

The higher the level of replacement volatility—as opposed to stable volatility—the higher the level of polarization.

Data, methods, and dependent variable

To test our predictions, we compile a panel dataset that contains all Legislative elections that took place in 18 Latin American countries from 1993 to 2016. Our sample is thus comprehensive as it includes all the independent states in Latin America for which data to operationalize the dependent and main independent variables are available during the aforementioned period. In all, our database contains 125 country-election observations. We thus conduct a large-N analysis, which in turn can make our claims much more generalizable if the main hypotheses are indeed supported by the statistical evidence. Further, the data presents substantial and temporal variation in both the main independent and dependent variables, which allows us to test hypotheses 1 and 2 more efficiently.

Given the structure of our dataset described above, we test our main hypotheses via ordinary least squares (OLS) regression models with country fixed effects. ⁵ The augmented Dickey-Fuller test reveals that our data is stationary (p-value = 0.018). ⁶ Therefore, accounting for a lagged dependent variable is not necessary, given that we do not find evidence of serial correlation in our data. Out of an abundance of caution, we conduct a number of robustness tests that include estimation of the main models with alternatives estimation techniques, inclusion of additional control variables, and a battery of tests (see Robustness Tests section below).

Admittedly, the operationalization and measurement of party polarization at the aggregate level across space and time is not an easy task. Researchers have traditionally relied on either the Dalton (2008) or the Taylor and Herman (1971) indexes of political polarization to study this phenomenon (see, e.g., Dalton 2021; Singer 2016). They are insightful but suffer from at least two important drawbacks. First, both indexes aggregate individual responses obtained from cross-sectional surveys. While individuals’ responses contain clear (ordinal) values, none of them account properly for the systematic variation in judgments across individuals or group of individuals who answer those questions. More substantively, such responses (often from elite surveys) say very little about the messages that are delivered to the electorate. That is, data coming from legislators’ policy or ideological positions may depict a polarized setting, while actual policies implemented by governments or parties in office do not fit into that configuration. For instance, Chile and Uruguay are cases with moderately high levels of polarization according to Dalton´s index, but the policies implemented by all governments since the democratic restoration have been clearly moderate in both countries (Pribble 2013).

To circumvent this potential problem, we follow Pemstein et al. (2020) who develop a measure of party polarization for the V-DEM project using responses from county-experts to the following question: Is society polarized into antagonistic, political camps? Specifically, Pemstein et al. (2020) refer to the extent to which political differences affect relationships beyond the political arena. Responses to the aforementioned questions go from 0 (Not at all. Supporters of opposing political camps generally interact in a friendly manner) to 4 (Yes, to a large extent. Supporters of opposing political camps generally interact in a hostile manner). V-Dem gathers data from expert surveys in each country. It is important to note, however, that the V-DEM project aggregates experts’ opinions with a measurement model, which produces an interval-level point estimate of the latent trait that typically vary from -5 to 5 (continuous variable). To allow for a more intuitive interpretation of the results, our main dependent variable—party polarization—is re-coded from 0 to 5.

A careful look at relevant data indicates that our dependent variable has substantial temporal and cross-sectional variation. Figure 1, for example, illustrates that there is a heterogeneous group of cases above the mean, from post-conflict societies such as El Salvador and Nicaragua, where government and opposition parties have been deeply divided since the early nineties, to cases such as Chile or Uruguay that feature robust democratic traditions, where government and opposition parties are able to share a set of common policy preferences and values. This variation is even more significant in cases below the mean like Mexico.

Results

The hypotheses presented in the theoretical section suggest that: (i) higher levels of electoral volatility are likely to increase the level of aggregate party polarization, and (ii) higher levels of electoral replacement volatility increase party polarization as well. Figure 2 plots the estimated coefficients obtained via OLS with country fixed effects. Specifically, Model 1 in Figure 2 (blue lines) presents the estimated coefficients of a fully specified model where party polarization is the dependent variable and total electoral volatility is the main independent variable. ⁸ It also includes the following covariates: log of GDPPC, growth, unemployment, enp, and district magnitude. The results indicate that total electoral volatility has a positive and statistically significant effect on the main dependent variable—political polarization—at the 5% confidence level despite the relatively small sample size. This suggests that, on average and holding everything else constant, higher levels of electoral volatility provide parties and their leaders with incentives to polarize, which ultimately engenders higher levels of party polarization. This statistically corroborates Hypothesis 1.

OPEN IN VIEWER

Figure 2.

Effects of covariates on party polarization.

We also assess the plausibility of Hypothesis 2. Specifically, Model 2 in Figure 2 presents the estimated coefficients of a fully specified model where the main independent variables are replacement and stable volatility. We find that both replacement volatility and stable volatility have positive and statistically significant coefficients at the 5% confidence level, which provide strong statistical support for hypothesis 2. As expected, the size of the coefficients reveals that replacement volatility has a larger effect than stable volatility on the main dependent variable—political polarization.

The effects of total electoral volatility, replacement volatility, and stable volatility are also substantially large and thus politically relevant. Figure 3 uses results from Model 1 to simulate the expected values of political polarization across different levels of total electoral volatility. A one standard deviation increase in total electoral volatility produces an increase of roughly 8% in political polarization. Figure 4 plots the predicted values of political polarization at different levels of (i) replacement volatility, and (ii) stable volatility. The effect of replacement volatility is substantial as one standard deviation in its value creates roughly a 10% increase in political polarization. Regarding stable volatility, Figure 4 reveals that its effect on political polarization is also substantial, but smaller in magnitude when compared to the effect of replacement volatility. Specifically, a one standard deviation increase in stable volatility increases political polarization by 3%. These results provide substantive support for Hypothesis 2.

OPEN IN VIEWER

Figure 3.

Predicted value of party polarization at different levels of total electoral volatility.

OPEN IN VIEWER

Figure 4.

Predicted value of party polarization at different levels of replacement and stable party volatility.

We find varied effects of the economic and other political control variables. For example, the results suggest that polarization is higher in countries with lower levels of GDP per capita, indicated by the negative and statistically significant at the 5% confidence level coefficient on gdppc(log). We also find that mean district magnitude has a positive and statistically significant coefficient at the 5% level. This result is consistent with Matakos et al. (2016), suggesting that higher levels of disproportionality engender political polarization, but inconsistent with Dalton’s (2021) results for advanced democracies. In contrast, the coefficients on unemployment and effective number of parties (enp) are negative but lack statistical significance at the 5% confidence level while growth is positive, but not statically significant.

Robustness and diagnostic tests

Although we are confident about the accuracy of the results presented in the previous section, we further check their validity by conducting a series of robustness and diagnostic checks. First, out of an abundance of caution, we assessed whether the results described above remained robust when the models were estimated with random rather than fixed effects. The results presented in Models A1 and A2 in the Supplemental Appendix support our main hypotheses. Specifically, the coefficients on total electoral volatility (Model A1), replacement volatility and stable volatility (Model A2) remain positive and statistically significant (at 0.05). Therefore, the results remain robust even when the models are estimated with random effects. Further, careful examination of our data reveals the presence of heteroskedasticity. ⁹ While this issue is unlikely to bias the covariate coefficients, we estimated an OLS model with country and year fixed effects (see Models A3 and A4 in the Supplemental Appendix). We do not find evidence of biased coefficients in our main models.

We also assess the robustness of our results by introducing a number of variables into our models that might, according with previous research, influence the level of political polarization. Specifically, we introduce the variable Gini (Model A1) as extant research has suggested that higher levels of income inequality increase political polarization (Winkler 2019; McCarty et al. 2006). To test claims that link the type of electoral system with polarization (Cox 1997), we include a dummy variable that is coded as 1 for PR electoral systems and 0 otherwise. Finally, we substitute the variable unemployment for the natural log of the inflation rate (labeled as inflation). Including these additional control variables do not substantially or significantly change our main results.

Conclusion

Numerous countries in the developing world have made important democratic progress in the past 40 years, but the levels of political polarization have also increased substantially. The rapid increase in polarization has motivated scholars and pundits to blame this phenomenon for the democratic slowdown in many developing democracies. We argue here that the level of electoral volatility in the system provides parties and candidates with incentives to polarize—or not. The uncertainty and much higher risk of losing elections associated with volatility provide parties and candidates with incentives to use polarization as a mobilization strategy.

This paper has important theoretical and practical implications. First, as mentioned earlier, extant studies on the determinants of political polarization have provided useful explanations of the demand side of such phenomenon. The theoretical argument in this paper, however, develops a supply-side explanation of such phenomenon. As such, it allows us to better understand why parties seek polarization as an electoral strategy even when this strategy seems to be at odds with the predictions of rational choice models. Rather than rivaling with demand-side explanations of polarization, this paper complements them and extends our understanding of parties and elites’ behavior in new and developing democracies.

Our theoretical argument and empirical results have implications for the literature studying the effects of party systems on political outcomes. Specifically, we argued that higher levels of electoral volatility introduce an important degree of uncertainty among competing parties. This, in turn, influences their choice of seeking polarization as mobilizational tool to drive voters to the polls. Indeed, polarization increases turnout across Latin America (see, e.g., Béjar et al. 2020). Therefore, even though the argument is elite-driven, our research is linked to other literatures on mass political behavior and democracy.

More specifically, scholars have long argued that systems with high levels of volatility are destabilizing for party systems and democracy (Mainwaring and Zoco 2007). Our argument (and empirical tests) extends beyond the effects of total electoral volatility on political polarization, analyzing how different types of volatility—replacement party and stable party electoral volatility—influence political polarization, allowing us to better understand whether and how the entry of new political parties into the system affects the level of polarization.

Our empirical results may initially lead us to believe that by increasing political polarization, high levels of electoral volatility may endanger the stability and survival of democracy. But, our argument and results suggest that polarization can have positive consequences too. In particular, we believe that a certain level of polarization can help the process of programmatic structuring of parties and party systems, where agents do not have clear programmatic differences (see APSA 1950). In many cases, the lack of substantive alternatives may obscure the selection of parties among citizens and complicate the process whereby voters have to evaluate parties’ performance. Thus, at the party system level, polarization may tend to stabilize party systems by providing ideological cues to voters, especially in contexts with marked electoral volatility. Developing and testing this argument is beyond the scope of this paper, but worth exploring.

Whether polarization is good or bad for democracies is an important empirical question with huge normative implications. Polarization may help party systems to become more programmatic, with more voters willing to capture the policy preferences of competing parties during elections. Yet, this optimistic picture also collides with an important caveat. We do not know the limits upon which polarization hurts party systems and democracies alike. For students of Latin American politics, increasing levels of polarization could have been a reasonable sign to expect the advance of programmatic politics in the region. Yet, cases like Venezuela and more recently Nicaragua show that polarization can hurt both party systems and democracies at the same time. Similarly and perhaps paradoxically, who would have imagined among those who applauded the 1955 APSA report on political parties that polarization would cause the observable consequences for US politics during recent years? Future research could identify such thresholds on polarization to capture positive and negative effects for party systems and democracies.

This paper can also be extended in other directions. First, the results provide strong statistical support for our main hypothesis at the aggregate level. However, it would be worthwhile to explore whether the main prediction of the theory presented here hold at the individual level. In this regard, scholars could use experimental techniques to test the hypotheses presented here. Second, scholars interested in explaining the causes of polarization should also device strategies that allow us to extend the cross-sectional and cross-temporal scope of the data presented here. While we are, to the best of our knowledge, the first to disentangle the effects of different types of volatility on polarization, future scholarship should theorize in more detail and empirically test the aforementioned relationship using available data from new and developing democracies.

Supplemental Material