1. The Meaning of Left-Right Ideology

As stated, the principal objective of this article is to analyze mass-level political polarization in Latin America in ideological terms. An essential initial step in assessing ideological polarization involves defining the concept of ideology within the framework of the left-right political spectrum. In the following paragraphs we provide a brief discussion of the definition of ideology we base our arguments and analyses on.

Although there is evidence suggesting that left-right ideology is context-dependent and means different things in different countries (Seligson Reference Seligson2007; Zechmeister Reference Zechmeister, Ryan, Matthew and Elizabeth2015), much recent research on the topic—especially in scholarship focusing on Latin American politics—has drawn on the ideas of Norberto Bobbio (Reference Bobbio1996). The central element of Bobbio’s distinction between left and right is their respective attitudes toward egalitarianism. According to Bobbio, the “irreducible, inescapable core of the (left-right) dichotomy” lies in the different attitudes towards the ideal of equality: The left is characterized by a push toward greater equality that extends beyond purely economic factors such as wealth and income to include social and political spheres. In contrast, the right tends to accept or justify social inequality as either a natural condition or a necessity for social order. In this sense, while the left tends to see inequality as a problem to be solved, the right tends to accept it as a natural part of society.

Luna and Rovira Kaltwasser (Reference Luna and Rovira Kaltwasser2014) followed Bobbio to define left and right in order to guide the incipient studies of the latter ideological stance in Latin America. For them, the right is the political force that considers social inequalities as natural or as the legitimate product of market operations, while the left considers inequalities as illegitimate and seeks to ameliorate them through state intervention. Levitsky and Roberts (Reference Levitsky, Roberts, Levitsky and Kenneth2011) coincide that the left refers to “political actors who seek, as a central programmatic objective, to reduce social and economic inequalities” (Levitsky and Roberts Reference Levitsky, Roberts, Levitsky and Kenneth2011, 5; emphasis in original). While historically the left in Latin America has focused on class differences, these authors argue that “many contemporary Left parties have broadened this focus to include inequalities rooted in gender, race, or ethnicity” (Levitsky and Roberts Reference Levitsky, Roberts, Levitsky and Kenneth2011, 5).

While most—if not all—definitions of left-right ideology stress the importance of the divergent stances toward equality in economic terms (and thus highlight the discussions on the role of the state in mitigating economic inequalities), it is becoming increasingly common to find scholars acknowledging additional dimensions, particularly belonging to sociocultural aspects (see for example Ostiguy Reference Ostiguy, Taggart and Ostiguy2017; Bornschier Reference Bornschier2010; Feldman and Johnston Reference Feldman and Johnston2014). Scholars such as Kitschelt (Reference Kitschelt1994) and Mudde (Reference Mudde2007) have provided insightful analyses on the impact of cultural issues, including immigration, law enforcement, and moral values, on ideological alignment. Jost, Federico, and Napier (Reference Jost, Federico and Napier2009) have contributed to this discourse by advocating for a multidimensional understanding of left-right ideology, which encapsulates the conventional socioeconomic axis and a rising sociocultural axis. In his study of the relationship between populism and polarization, Roberts (Reference Roberts2022) claims that the economic and cultural axes are both analytically distinct and spatially orthogonal and describes the cultural axis as having “a cosmopolitan/universalist upper pole and a lower pole where ethnic, religious and/or nationalist particularisms are located” (Roberts Reference Roberts2022, 686). This expanded analytical framework offers a more comprehensive reflection of the complexities inherent in contemporary ideological identities. As will be seen in the next sections, we have taken into account the multidimensionality of left-right ideology when selecting the variables we use to test the consistency of left-right ideological self-placement in the Latin American region.^{Footnote 3}

2. Ideological Restructuring in Twenty-first-Century Latin America

Polarization necessarily involves a shift in ideological preferences, whether at the level of parties or at the level of citizens (masses). A typical polarization process occurs when there is a centrifugal movement from the center (or “middle”) of a distribution toward the extremes, driving groups or individuals increasingly distant from each other (Dalton Reference Dalton2008; Pew Research Center 2014; Fiorina, Abrams, and Pope Reference Fiorina, Abrams and Pope2008). Most of the scholarly attention focusing on Latin America has been devoted to analyzing ideological shifts at the elite level, primarily because of the “left turn” in the region’s politics that began at the turn of the century. Ideological shifts were clearly taking place at the elite level: as Levitsky and Roberts note: “By 2009, nearly two-thirds of Latin Americans lived under some form of left-leaning national government…. never before had so many countries in the region entrusted the affairs of state to leaders associated with the political Left” (Levitsky and Roberts Reference Levitsky, Roberts, Levitsky and Kenneth2011, 1).

But have there been important changes in the ideological self-identification of Latin American citizens? Roberts (Reference Roberts2014) argues that the electoral arena in Latin America has been restructured along the left-right ideological axis in the twenty-first century. According to him, “electoral competition in the postadjustment period associated with the ‘left turn’ helped to realign (or restructure) many party systems programmatically, often on the basis of new partisan alternatives” (Roberts Reference Roberts2013, 1428). If the electoral arena has indeed been restructured in this way, do we also see it in public opinion?

Several analysts have argued that there has been no significant ideological restructuring at the citizen level in Latin America. Arnold and Samuels (Reference Arnold, Samuels, Levitsky, Kenneth and Baltimore2011), in analyzing whether the ideological shifts found at the elite level during the “left turn” in Latin America were reflected in similar shifts in mass public opinion, found no significant mass shift to the left in the majority of Latin American countries. Based on data from the Latinobarómetro surveys between 1996 and 2008, these authors note the lack of attitudinal change among Latin American citizens during this period. The lack of evidence of ideological change among Latin American voters led them to describe the left turn as a “relatively shallow electoral phenomenon rather than a major realignment of citizens’ ideological beliefs” (Arnold and Samuels Reference Arnold, Samuels, Levitsky, Kenneth and Baltimore2011, 32).

Using data from the AmericasBarometer surveys, Mitchell Seligson (Reference Seligson2007) found that the average ideological “self-placement” of Latin American citizens on a left-right scale (1–10) in 2006 was 5.77, or slightly right-biased (the arithmetic midpoint of the scale is 5.5). He also notes a slight “shift to the left” happening between 2004 and 2006. Seligson observes that the trend is region-wide, “but the magnitude of the shift is small and the center of gravity remains somewhat to the right” (Seligson Reference Seligson2007, 84).

Along the same line as Arnold and Samuels (Reference Arnold, Samuels, Levitsky, Kenneth and Baltimore2011), other authors have argued against a polarization or left-right restructuring of politics at the mass level in Latin America. Examining electoral support for the left in the period of the “left turn,” Murillo, Oliveros, and Vaishnav (Reference Murillo, Oliveros and Vaishnav2010) conclude that it was primarily a function of retrospective economic voting rather than an ideological shift of the population. Their findings build on “recent public opinion evidence, which does not find any region-wide move to the left. Even in countries that did witness a leftward shift, there is no evidence of a dramatic ideological realignment” (Murillo, Oliveros, and Vaishnav Reference Murillo, Oliveros and Vaishnav2010, 106).

We update the analysis of Arnold and Samuels (Reference Arnold, Samuels, Levitsky, Kenneth and Baltimore2011), Seligson (Reference Seligson2007), and Murillo et al. (Reference Murillo, Oliveros and Vaishnav2010), which examined Latin American public opinion up to 2008 (in the best case). We use data from AmericasBarometer surveys starting in 2006 and continuing through 2019.^{Footnote 4} Our results are interesting in three ways: first, we find that the average regional ideological self-placement is 5.36 in 2016/17 and 5.39 in 2018/19. This shows that there has been a region-wide shift to the left since 2006; in Seligson’s terms, the ideological center of gravity of Latin Americans is now somewhat to the left. Second, we find that since 2010 there has been a sharp and continuous increase in the percentage of citizens who identify with the left, as well as a moderate increase in those who identify with the right. Third, there has been a very significant decline in the number of people who did not respond to the question on ideological self-placement. All these findings lead us to believe that between 2010 and 2019 there has been an important ideological restructuring of the Latin American public, which could also be accompanied by an increase in the levels of ideological polarization.

3. Polarization: Theory and Metrics

Although polarization is an important phenomenon in the social sciences, there is no consensus on how to conceptualize and assess it. According to Kubin and Sikorski, polarization is “not clearly defined, nor consistently measured” (2021, 188). In this section, we present our concept of polarization, show that existing measures do not fit it, and propose a new measure of our own.

Fiorina et al. offer a straightforward definition of polarization as a “movement from the center toward the extremes” where the middle should “lose to both extremes” (2008, 557; emphasis in original). This definition has several implications. First, polarization is a phenomenon that occurs in variables that have at least three values: two opposing extremes and a center. In the case of left-right ideological scales, this means an “extreme left” group, a “moderate” group, and an “extreme right” group.

Second, Fiorina et al. (Reference Fiorina, Abrams and Pope2008) suggest that polarization is bimodal in nature, a function not only of the total size of the extreme groups combined, but also of how similar their sizes are. Likewise, Bramson et al. refer to bimodality when considering “size parity” as a possible understanding of polarization: groups are “more polarized in this sense if the different clusters of beliefs are held by equal numbers of people” (2017, 127). Thus, a polarization index should be higher when the two extreme groups are of equal size. A good measure of ideological polarization would have a minimum value of zero if one or both extreme groups contained no individuals—there is no polarization without the presence of individuals in both extreme groups. In the case of ideological polarization, the maximum value, or a score of 100, should be obtained when exactly 50% of the individuals belong to the “extreme left” and 50% belong to the “extreme right.”

Fiorina et al. also seem to argue that polarization can only grow if both extreme groups grow at the same time. This last position, however, is not unanimous in the literature. Esteban and Ray (Reference Esteban and Ray1994) conceptualize polarization as a function of how many people feel antagonized by how many other people. By this definition, polarization increases when one extreme group grows while the other extreme group remains constant, and a large increase in one group can more than offset a small decrease in the other. Similarly, Hacker and Pierson argue that polarization in America is growing asymmetrically, driven by a one-sided and sharp “retreat from moderation on the right of the spectrum” (2015, 59). We agree with Esteban and Ray’s theoretical argument and Hacker and Pierson’s empirical findings, and thus we believe that an ideal metric should be sensitive to asymmetrical increases in polarization.

What metrics have been proposed in the literature and how do they conform to these ideal attributes? Some have measured polarization using statistically inspired concepts such as variance and standard deviation (Dalton Reference Dalton2008; DiMaggio, Evans, and Bryson Reference DiMaggio, Evans and Bryson1996; Fiorina, Abrams, and Pope Reference Fiorina, Abrams and Pope2008; Taylor and Herman Reference Taylor and Herman1971) or excess kurtosis (DiMaggio, Evans, and Bryson Reference DiMaggio, Evans and Bryson1996). To give an example of how similar these metrics are to statistical concepts, Dalton’s (Reference Dalton2008) polarization index, often used to measure the ideological polarization of parties, is the standard deviation linearly transformed to produce a number between 0 and 10. Similarly, Taylor and Herman’s (Reference Taylor and Herman1971) use variance weighted by the number of seats to measure ideological disagreement in a party system.

Standard deviations and variance measure how spread out a distribution is, while excess kurtosis measures how much the tails of a distribution deviate from what one would expect from a normally distributed variable. While these metrics are appropriate for these tasks, they lack a substantive understanding of which parts of a distribution matter most (Koudenburg, Kiers, and Kashima Reference Koudenburg, Kiers and Kashima2021, 2–3). As shown in Appendix 6, these metrics produce positive levels of polarization even when everyone is a moderate or when an entire half of the ideological distribution is empty, ignoring Fiorina’s advice that polarization requires the presence of both extremes. A similar problem is found in Taylor and Herman’s (Reference Taylor and Herman1971) ordinal disagreement metric.

Abramowitz and Saunders (Reference Abramowitz and Saunders2008, 544) suggest assessing levels of polarization based on the sum of the percentages of people at the poles of the scale. (i.e., the “extremes”). In their work, they sum the percentages of people who consistently hold “liberal” and “conservative” positions. This approach to measuring polarization has been criticized because it does not take into account the assumption of bimodality that polarization entails. For example, suppose that all individuals in an ideological distribution self-identify as “far left”; according to Abramowitz and Saunders’ metric, this would imply maximum polarization, which is inconsistent with Fiorina et al. (Reference Fiorina, Abrams and Pope2008).

One could also conceive of a metric of ideological polarization based on the size of the smallest of the two extreme groups, although we found no such example in the literature. While this metric would be fully compatible with Fiorina’s bimodality assumption, it would fail Hacker and Pierson’s (Reference Hacker, Pierson and Persily2015) requirement to be sensitive to asymmetric polarization. After all, the Min(x,y) function is not sensitive to variations in the largest of x and y.

Moving to metrics based on welfare concepts, Esteban and Ray (Reference Esteban and Ray1994) proposed measuring polarization as the sum of all antagonisms or total alienation. Their polarization metric requires a multilevel categorical variable with ordinal properties, such as income group (their example), left-right ideology, or a political opinion. Individuals within the same group are considered to feel at peace with other individuals in the same group but feel alienated from all individuals in other groups. This feeling of alienation is more intense toward individuals who are further away on the variable of interest. Their measure of polarization is the sum of all the alienation that every individual feels toward every other individual.

Recently, Koudenburg, Kiers, and Kashima (Reference Koudenburg, Kiers and Kashima2021) noticed that the Esteban-Ray index assumed cardinal properties in the variable of interest. It assumed that an individual who self-identified as having an ideology of 4 (center-left) on a scale of 1 to 10 would feel equally alienated by an individual with an ideology of 2 and by another individual with an ideology of 6. They decided to empirically estimate the degree of alienation that groups feel toward each other by analyzing how experts rate opinion distributions as more or less polarized.

Like standard deviation and excess kurtosis, these welfare-based measures are not entirely compatible with the idea that there can be no polarization without both extremes. If the entire population consists of moderates and extreme leftists, these metrics will measure some positive polarization. In addition, Koudenburg, Kiers, and Kashima’s (Reference Koudenburg, Kiers and Kashima2021) estimate of alienation is difficult to replicate because it relies on expert surveys to calculate alienation weights.

Based on the shortcomings of existing polarization metrics, we decided to define a new polarization index. It should be noted, however, that our empirical results in this article do not depend on our specific polarization index. In fact, we have measured polarization in Latin America using other metrics: standard deviation, Esteban and Ray’s Effective Antagonism (Reference Esteban and Ray1994), Abramowitz and Saunders’ sum of extremes (Reference Abramowitz and Saunders2008), Taylor and Herman’s (Reference Taylor and Herman1971) ordinal disagreement and variance metrics. The results for the Latin American region using these metrics are presented in Appendix 4.

4. Definition: Polarization Index

The Ideological Polarization Index (IPI) we propose is defined as:

$$IPI\left( {L,R} )\right = 2\sqrt L\sqrt R$$

Where L ∈[0, 1] is the proportion of individuals who self-identify as having an extreme left ideology, and R ∈[0, 1] is the proportion of individuals who self-identify as having an extreme right ideology. The function can also be expressed in a way that includes moderates:

$$IPI = 2\sqrt {1 - R- M} \sqrt {1 - L- M} $$

Where M ∈[0, 1] is the proportion of individuals who self-identify as moderates and is equal to M = 1 – L – R. Surveys vary on how they collect ideology from interviewers, and so a transformation is required to fit existing data into this three-category variable. In the Data and Measurement section, we explain how we operationalize our measure using AmericasBarometer data.

Proposition A below confirms that our polarization index is compatible with the theoretical requirements mentioned above. In addition to being straightforward to compute and interpret, it respects bimodality, asymmetric polarization, and has a well-defined range. It is also quasi-sensitive to bimodal distance.

1. 1. IPI is sensitive to asymmetrical polarization.

2. 2. IPI is bimodal: It increases as the extremes get more balanced, for any level of M.

3. 3. IPI conforms to the assumption that there is no polarization without both extremes.

4. 4. IPI has a well-defined range, from 0 to 1.

Our index measures polarization as a number between 0 and 1. In most of this article, we present it as a score, ranging from 0 to 100.

5. Data and Measurement

Our aim is to compare the degrees of ideological polarization between countries in Latin America and their changes within countries over the last two decades. We use data from the AmericasBarometer surveys, conducted by the Latin American Public Opinion Project (LAPOP). The AmericasBarometers are nationally representative surveys that have been conducted every two years since the early 2000s in most North, Central, and South American countries, as well as in a significant number of countries in the Caribbean region.^{Footnote 6}

To measure ideological polarization within countries, one must start from assessing ideological self-placement of individuals. We use the item of the AmericasBarometer surveys that asks about respondents’ self-placement on the left-right ideological scale. The AmericasBarometer surveys have included the following question for tapping ideological self-placement since the early 2000s:

As mentioned in previous sections, our polarization index requires separating respondents into three groups: the “extreme left,” the “moderates,” and the “extreme right.” We consider an individual to have an extreme ideological position if he places himself in one of the two positions located at the ends of each tail of the original ideological scale. Accordingly, we construct a new three-category variable for ideological self-placement, in which responses of “1” and “2” are considered “extreme left,” scores of “9” and “10” are considered “extreme right,” and everything in between (individuals with scores ranging between “3” and “8”) are considered “moderates.”

7. Discussion: Trends in Ideological Polarization in Latin America

The results so far show that in Latin America as a whole, ideological polarization declined from 2006 to 2010, increased steadily from 2010 to 2016/17, and then declined somewhat from 2016/17 to 2018. This section shows that a similar trend can be found in most countries in the region and almost all countries in South America. Moreover, polarization has increased in all but three countries over the period. More than an interesting curiosity, we believe that these common trends can be linked to other processes in the region, motivating further research.

In 13 of the 19 countries studied (68%), ideological polarization at the mass level evolved in a V-shape, characterized by (1) an initial decrease until 2008 or 2010, when it reached a minimum, and (2) a constant or almost constant increase thereafter, reaching a peak in one of the last two survey waves. The countries that followed this pattern are Argentina, Bolivia, Brazil, Colombia, Ecuador, Dominican Republic, Haiti, Honduras, Panama, Paraguay, Peru, Uruguay, and Venezuela. In Colombia, polarization fell from 27.7 in 2006 to 25.4 in 2010, and then increased almost steadily until 2019, when it peaked at 31.5. In Brazil, the valley of polarization was reached in 2008, when Luiz Inácio Lula da Silva was still a very popular president, and then increased in an irregular manner until 2019. In Haiti, the changes in ideological polarization have been particularly dramatic: a decline from 24.1 in 2006 to 10.7 in 2010, one of the lowest scores in the region, followed by a rapid increase, reaching a maximum of 61.6 in 2017, one year before the beginning of a major economic and political crisis. This V-shaped trend is particularly prevalent in South America, where all countries except Chile follow it.

The correlated movement of ideological polarization levels in most countries in Latin America, and especially in South America, suggests a potential common cause. Perhaps the V-shaped trend is related to the left wave, the global economic crisis of 2008–9, the commodity boom, the diffusion of norms across borders, or increasing fragmentation in Latin American multi-party systems. These causes could have a non-linear effect; left-wing presidencies seem to have caused an initial decrease in polarization, but later an increase. They may also interact in complex ways. In any case, it is left to future research to establish these relationships, as it is beyond the scope of this article.

Another common trend we observe is that ideological polarization at the mass level increased almost everywhere in Latin America from the beginning to the end of the period studied. Of the 19 countries, 16 experienced an increase in polarization, and 9 experienced a double-digit increase. This finding contrasts with the literature, in which older studies found no change in mass-level polarization (Arnold and Samuels Reference Arnold, Samuels, Levitsky, Kenneth and Baltimore2011; Murillo et al. Reference Murillo, Oliveros and Vaishnav2010; Seligson Reference Seligson2007), while more recent studies found only a slight increase (Murillo Reference Murillo2022). Table 2 classifies the countries studied according to whether they show a V-shaped evolution of ideological polarization and whether they experienced an increase in their level of polarization from 2006 to 2019. As can be seen, most countries (12 out of 19) fall into the upper left quadrant: V-shaped evolution and an increase in polarization, the same pattern experienced by Latin America as a region.

Table 2. Trends in Mass-Level Polarization in Latin America

8. Does Ideological Self-placement (and Polarization) Matter in Latin America?

It has been argued that in Latin America “policy preferences play a secondary role to issues of partisan identity or feelings of my team versus yours” (Murillo Reference Murillo2022). Therefore, we felt it necessary to examine whether the policy preferences of Latin American leftists and rightists are consistent with theoretical predictions.

Following Abramowitz and Saunders (Reference Abramowitz and Saunders2008), we examine the relationship between political interest, self-reported ideology, and actual policy preferences to see whether voters’ policy preferences are consistent with their ideology. In the AmericasBarometer data, political interest is measured by variable pol1, with four categories: “a lot,” “some,” “little,” “none.” We will focus on comparing the opinions of extreme right and left wingers, not moderates. Table 3 shows that polarized individuals—whether on the left or the right—are much more likely than moderates to show “a lot” of interest in politics. About 16% of far-left and far-right individuals say they are very interested in politics, compared with only 9.3% of moderates.

Table 3. Political Interest by Polarized Groups

As noted earlier in this article, the literature sees egalitarianism as the core issue separating left and right ideologies. Left ideology supports public policies that improve economic equality, while right ideology views these inequalities as natural and not a priority for public efforts. In addition, sociocultural aspects have been added as a second dimension to the ideological dichotomy, with the left usually seen as cosmopolitan, while the right is tied to traditional values and regionalism. To test the alignment of the Latin American left and right in terms of their ideologies, we selected two variables from LAPOP: ROS4 and D5, which capture support for policies that reduce inequality and support for the right of homosexuals to run for office, respectively. In addition to capturing important political predictions, ROS4 and D5 have the best sample coverage among the ideologically relevant variables, having been included in more than 100 country-wave surveys.

Table 4 shows the percentage of people in each group who strongly support policies to reduce inequality (response 7 to ROS4). It shows that among those who are very or somewhat interested in politics, left wingers are more supportive of these policies than right-wingers, which is consistent with ideological theory. However, among those who are not interested in politics, opinions are the opposite of what the theory expects. Considering all individuals at the ideological extremes, regardless of political interest, leftists are only slightly more supportive of redistributive policies than rightists.

Table 4. Percentages of Individuals Who Strongly Support Redistribution, by Ideology and Political Interest

Significance levels: *** 0.1%, ** 1%, * 5%, + 10%.

Next, we examine ideological consistency using D5, a variable that measures “support for the right of homosexuals to run for office” on a scale of 1 to 10. Table 5 shows the percentage of people who support this right (answers 6 to 10 on D5) by ideology and political interest.

Table 5. Percentages of Individuals Who Support Homosexuals’ Right to Run for Office, by Ideology and Political Interest

Significance levels: *** 0.1%, ** 1%, * 5%, + 10%.

The survey shows that, for the most part, people surveyed by LAPOP have preferences that are consistent with what the theory suggests. Among those who are interested in politics, most left wingers support the right of homosexuals to run for office, while only a minority of right-wingers do. As before, however, this ideological consistency diminishes and disappears as political interest declines. Overall, left wingers are 3.2% more likely than right-wingers to support this right, a difference that is statistically significant.^{Footnote 9}

What we take from these results is that some of the polarization we see in Latin America is consistent with theoretical predictions about ideological preferences, and some is not. Ideological self-placement seems to be well aligned with theory among the most politically engaged, but not so much among other groups.