Introduction
Over the last two decades, the study of polarization has gained increased interest among scholars. A large number of studies have been published attracting different disciplines and perspectives, from social psychology and communication to sociology and political science. In particular, many researchers expressed concerns regarding the impact of polarization on undermining democracy, an emerging debate that has been observed mainly in Europe and the United States (McCoy and Somer Reference McCoy and Somer2019; Torcal and Comellas Reference Torcal and Comellas2022; Voelkel et al. Reference Voelkel, Chu, Stagnaro, Mernyk, Redekopp, Pink and Willer2023; Carlin and Love Reference Carlin and Love2018).
During the last ten years in particular, a large body of literature focused on AP, which studies the trend of people’s resentment towards political adversaries. These studies mostly hypothesize that the driver of polarization is the increased distrust and dislike of political opponents vis-à-vis the intense preference of their own partisans. Many studies describe and analyze the widening gap between in-party preference and contempt for out-parties. These analyses are mostly based on surveys that measure such differences (Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019; Mason Reference Mason2015; Wagner Reference Wagner2021; Gidron et al. Reference Gidron, Adams and Horne2020; Druckman and Levendusky Reference Druckman and Levendusky2019; Reiljan Reference Reiljan2020).
To describe the characteristics of AP and identify its determinants, several studies focused on the shifting patterns of political affiliation (Mason Reference Mason2015; Lelkes Reference Lelkes2021), the impact of partisanship and affects on social and political identity (Iyengar et al. Reference Iyengar, Sood and Lelkes2012; Comellas and Torcal Reference Comellas and Torcal2023), the role played by social networks and the growing effect of bubbles and echo chambers (Waisbord Reference Waisbord2020; Garret el al. 2019; Levandusky Reference Levendusky2013), how social drivers such as the loss of status, inequality, changes in labor markets, and many others socioeconomic variables have yielded higher polarization (Gidron et al. Reference Gidron, Adams and Horne2020; Torcal and Comellas Reference Torcal and Comellas2022; Barber and McCarty Reference Barber, McCarty and Persily2015). This emerging literature, particularly over the last decade, emphasizes the importance of social and in-group identity as drivers of attitudes and political behavior that has produced intense division, and a reduction of interpersonal interactions (Mason Reference Mason2018; Hartveld Reference Harteveld2021). Studies have documented the growing negative attitude and evaluations of out-group partisan which denote fractures in the social fabric and defies the foundations of democracies in Europe and the United States (Hartveld et al. Reference Harteveld, Mendoza and Rooduijn2022; Somer et al. Reference Somer, McCoy and Luke2021).
Despite clear signals of similar polarization processes described in many Latin American countries, few empirical studies assess and measure the intensity of AP in the region (Segovia Reference Segovia2022; Kessler and Vomaro Reference Kessler and Vomaro2021; Castro Cornejo Reference Castro Cornejo2023). And there are no well-grounded attempts to compare AP across countries of Latin America, as well as between nations of the region and countries in Europe and North America. Are Latin American countries more, less, or equally polarized as other Western nations? Is the intensity of hostility toward political opponents similar or different among Latin American countries?
In this article, we first describe the level of AP in six countries of the region (Argentina, Brazil, Chile, Costa Rica, Mexico, and Peru). Using data from representative surveys conducted in these countries after elections between 2000 and 2018, we measure the degree of dislike and disapproval of opponents reported by respondents. We replicate three different methodologies applied for many Western nations to compare results with these countries, as well as to analyze the evolution of AP in two nations. We also study several traits of in-party liking and out-party dislike, while also providing a glimpse into the effect of support for democracy from different publics. Finally, we provide initial evidence to ascertain whether AP is a general trait found in the majority of citizens or whether AP is encapsulated among intense minorities of voters.
This research note proceeds as follows. The next section will discuss the method and the data. The results section presents the results followed by discussion in the next section. The last section will briefly describe the research agenda of Latin America AP.
Data and Method
We measure AP using the traditional “thermometer” widely tested in the literature. Based on data from surveys administered several weeks after general elections in a given country, a random sample of more than 1,000 voters was interviewed on a wide variety of topics. Footnote 1 The data is assembled through the Comparative Study of Electoral Systems (CSES) Footnote 2,Footnote 3
For this study, we relied on questions that ask respondents to assess how much they like the party they favor against how much they dislike other parties. Respondents were asked to rate on a scale from 0 to 10 (this is the thermometer scale) how positive their feeling is toward their party of preference as well as towards other parties. Higher numbers denote warmer feelings. Footnote 4
Conceptually, measuring AP involves assessing individuals’ affinity towards political parties, which can present challenges in several Latin American countries. Unlike many Western European nations with a strong tradition of political party identification and affinity, some countries in the region demonstrate very loose party affiliations. Measures of AP that rely on such association with partisanship should be interpreted with care in some instances, a topic to be addressed in the discussion section.
For this research note, we tested AP using three different methodologies that attempt to capture the level of AP in multiparty systems, a feature shared by most countries in Latin America. Since the standard estimation method has been developed for the United States, it has been designed for a competitive two-party system and it is much more straightforward. However, in the last few years adjustments have been advanced to measure AP when the party of choice is one among many parties. Moreover, in many multiparty systems, individuals may feel an affinity towards two or more parties and may switch their vote among a variety of large and small parties.
Three methods have been advanced for multiparty systems, one by Reiljan Reference Reiljan2020 (the difference method) and two by Wagner Reference Wagner2021 (the spread, and the distance methods). We briefly describe each one before presenting the results.
The Difference Method
The standard way to estimate AP is to capture the difference in feelings individuals have towards political parties. The difference between the average in-party score (between 0 and 10) against the average out-party score is the affective polarization index (API). Higher scores indicate more polarization.
This method widely used in the literature on the United States relies on a two-party system, so warm feelings towards Republicans and Democrats can be easily interpreted. Footnote 5 Using difference rather than plain warmth towards one party is a better measure since scales of 0 to 10 might be interpreted differently by each individual. Thus, the scores can be biased by subject perceptions and scales. However, each person, presumably, uses the same scale of warmth and coldness for in-party and out-parties, and therefore the difference is a much more accurate measure than the raw score for each party (Lelkes and Westwood Reference Lelkes and Westwood2017).
Most Latin American countries have multiparty systems as in Europe. We, therefore, adjust feelings to the actual share of the vote received by all relevant parties. Footnote 6 Let us assume that in a given country Party A received 50% of the vote, Party B 40%, and PC 10%. The scores are then adjusted by the proportion of votes received. For example, if subject i voted for Party A and scored a feeling of 9 towards Party A, 1 for Party B, and 5 for Party C, subject i score will assign a larger weight to the difference between Party A and B (9−1=8) and a smaller weight to the difference between Party A and C (9−5=4). According to this method, the computing of AP for partisans of a given party is the average feelings of the preferred party (in-party) concerning their feelings towards other parties (out-parties). This is given by:
Affective polarization of Party A is
$$A{P_n} = \sum\limits_{m = A}^n {\left[ {\;\left( {Lik{e_n} - \;Lik{e_m}} \right)*\left( {\;{{Vote\;Shar{e_m}} \over {1 - \;Vote\;Shar{e_n}}}} \right)} \right]} \;\;with\;m\; \ne n$$
Where vote share, is each party share of votes in the first presidential round, like is the score assigned to each party n by individual i. Finally, AP is the affective polarization of each party n.
Affective polarization for partisans of different parties weighted by their vote share are added to obtain the total AP for a given country
$$API = \;\mathop \sum \limits_{n = 1}^N \left( {A{P_n}*\;Vote\;Shar{e_n}} \right)$$
Once the average in-party liking and out-party dislike are computed, the total AP for a given country and year is added up. These scores can compare API between countries of Latin America and other Western nations. Any score below 3 and above 5 will be considered a very extreme case. Most scores will range between 3.5 and 4.5; therefore small differences will depict clear differences.
The Spread Method
A second conceptualization of AP focuses on the total dispersion of individual preferences rather than the above favorability difference between in-party and other out-parties (Reiljan Reference Reiljan2020). Instead of measuring the difference of scores each individual assigns to their preferred party (in-party) vis-à-vis the non-preferred party (out-party), Wagner (Reference Wagner2021) claims that in multiparty systems it makes more sense to capture individual feelings towards several parties that may be close ideologically but had not been the first preferred party. Footnote 7 Conceptually, this rationale holds true because, as demonstrated by Torcal and Comellas (2023), citizens tend to polarize more along ideological lines rather than on specific issues. Likewise, according to Bantel (Reference Bantel2023), in multiparty systems, the basis of AP is often more rooted in ideological camps than in specific political parties.
The spread method seeks to capture the dispersion of all individual feelings, their range, and their distance from a mean. The higher the score and the greater dispersion of feelings and the affects how affectively polarized a nation is. The assessment seeks first to estimate the individual score of spread (some publics will have feelings of warmth towards most parties and therefore the score will be low, while others will have warmer feelings towards selected parties and very cold feelings towards other parties, and in this case, the spread score will be high). Then the individual scores are added to generate an average score. Formally, this is given by:
$$Spread\;Unweighte{d_i} = \;\sqrt {{{\sum\nolimits_{p = 1}^p {{{\left( {lik{e_{ip}} - \;l{{\overline {ike} }_i}} \right)}^2}} \;\;} \over {{n_p}}}} $$
Where p is the party, i is the individual respondent, and the
$lik{e_{ip}}$
is the like-dislike score assigned to each party p by the person “i”, and n is the total number of parties. A weighted version makes adjustments for the quantity of votes received by a party and is estimated by
$$Spread\;Weighte{d_i} = \;\sqrt {\mathop \sum \limits_{n = 1}^N Vote\;Shar{e_p}*\;{{\left( {lik{e_{\;ip}} - \;l{{\overline {ike} }_i}} \right)}^2}} $$
Where
$Vote{\rm{\;}}Shar{e_p}$
is the proportion of votes received each party in the last election, the mean affect should itself be weighted by party size and is calculated as
$$l{\overline {ike} _i} = \;\mathop \sum \limits_{p = 1}^p \left( {Vote\;Shar{e_p}*lik{e_{ip}}} \right)\;$$
While difference estimation includes only respondents that identify or clearly state that they like a given party (approximately half of the respondents in each sample) the spread estimation includes all respondents that provide scores for each party (80%–90% of respondents in the samples). The Wagner method seems to be more sensitive to coalitions of many parties, which are more frequent in Latin America.
The Distance Method
In addition to the total spread, individuals may differ in the intensity of liking a preferred party against other parties. This approach seeks to capture such intensity, where a low score indicates low AP. This method measures the average affective distance of other parties from one’s most liked party. It is more sensitive to the dislike of out-parties, while the positive affects of one party are the key element. A higher average dislike of other parties yields higher coefficients. Footnote 8 The estimation of mean-distance unweighted AP is:
Where n is the number of parties not including the most-liked party,
$lik{e_{max,i}}$
is the score assigned to the most-liked party. Similarly, a weighted version can be estimated by,
$$Distance\;Weighte{d_i} = \;\sqrt {\mathop \sum \limits_{n = 1}^N Vote\;Shar{e_n}*\;{{\left( {lik{e_{in}} - \;lik{e_{max,i}}} \right)}^2}} $$
As mentioned, both spread and distance estimation produce unweighted and weighted estimations. Unweighted results use the survey sample as given, while weighted results adjust for the share (%) of votes each party has received in the election. Footnote 9 In this article we will mostly use the weighted estimation because it takes into consideration the social distributions of affects and disaffects among all voters, reducing the effects of sampling bias.
In summary, quoting Wagner: “The mean-distance measure differs from the spread-of-scores measure in that the former captures how much an individual on average dislikes other parties compared to his or her favored party, while the latter simply captures the extent to which affect is spread out across the various options.”
Results
We analyze results in two subsections: The first one presents AP scores for the ten cases under study using several estimations. We also present the distance (Reiljan) method with a slight modification, which includes in some cases the entire sample. Footnote 10 Instead of using the filter question of whether respondents like a given party, in some instances we use the question of whom the respondent has voted for. This will allow us to analyze two important publics: those who do not identify with a single party but favor candidates of a party in a given election, and to separately analyze whether these non-party identifiers score differently than those who openly favor one party. The scores of these two subgroups are also replicated for the spread and distance methods. We will show that this may yield interesting results that have not yet been properly studied. Footnote 11
The second subsection uses aggregate descriptive statistics to provide initial evidence to respond to four questions: Who drives AP? Is AP growing? How does API in the region compare with polarization in the world? And, do some sociodemographic profiles of individuals explain AP in the region?
AP Scores
Party Likeability
In-party favorable scores vary across parties and countries. In some cases, high differences are driven by strong warm feelings towards the preferred party, and in other cases, high differences may be driven by strong dislike towards out-parties. Similar scores might result from favoring strongly the in-party or holding strong unfavorable views of the out-party. Therefore, our first step is to measure the in-party favorability scores for major parties in each country. Footnote 12
Figure 1A summarizes the average likeability of voters and party identifiers for each country and election wave (in-party), while Figure 1B presents the average dislike scores of voters and party identifiers towards other parties (out-parties). For the entire, the mean likeability is 7.33 (s.d. 1.37. Min 2.50–Max 9.09) and party dislike mean is 2.59 (s.d 1.01. Min 0.44 Max 4.56), indicating an expected significant gap. Footnote 13 We use a bootstrap of 1,000 iteration to construct confidence intervals for each individual score of Figures 1A, 1B, and all APIs that use the Difference method. Footnote 14
Figures 2A and 2B display scores for our ten case studies for in-party favorable view scores and out-party dislike scores of voters for the large parties, which represent between 70% and 80% of the electorate in most cases. Footnote 15 For example, FPV voters in Argentina in 2015 had on average a 7.74 likeability for their party (on a scale from 0 to 10, where 10 is very favorable and 0 is very unfavorable). The same FPV voters had a combined average likeability towards the other parties of 1.62. The difference for each party or coalition towards their political opponents is quite large (usually more than 5).
In-Party Liking and Out-Party Dislike.
In-Party Liking and Out-Party Dislike for Each Party Wave.
Initial descriptives show several features on the likes and dislikes of parties and coalitions in the region. First, the gap between in-party and out-party favorability is significant in most country years. Second, there are at least two clear groups of country/elections, one that depicts strong likeability for the in-party (Mexico and to some extent Peru) and one that has low scores of likeability towards out-parties (Brazil, Chile, and Costa Rica) with Argentina 2015 probably in the middle.
Third, as a signal of deep polarization, wider gaps are noticeable between the two larger parties in each country’s election. The mean difference of like-dislike between the two largest parties shows that voters’ affect towards one of the larger parties significantly varies from the low affect to the other large out-party. Larger scores initially denote greater polarization between the two groups. Brazil 2018 and Mexico 2018 depict the higher polarization between the two largest parties, while other cases also show high differences (around 5) that explain considerable animosity between supporters of the two main parties or coalitions.
We argue that it is crucial to first outline the levels of favorability and dislike towards given parties because raw API scores do not inherently distinguish whether they stem from strong positive sentiments towards in-parties or profound negative sentiments towards out-parties. This distinction is particularly critical for most countries in Latin America. As will be observed, some countries, such as Chile, exhibit low API scores because they have very low in-party likeability yet strong out-party dislike. Footnote 16
In short, the first indication of high polarization could be found between the favorability scores of voters of the largest parties. By this measure, Brazil 2018 and Mexico are highly affectively polarized. As will be shown in the next section, the presence of smaller parties or coalitions moderates the level of polarization.
Affective Polarization Indexes
In this section, we analyze API scores for ten case studies using three methods.
Difference
For each country/year Figure 3 presents the total scores of API using the difference method. As mentioned, it will be very rare to find API values higher than 5 or below 3. For example, in a study that includes 81 countries/years (Gidron et al. Reference Gidron, Adams and Horne2020) the API mean is 4.38 with a minimum of 2.69 and a maximum of 5.7. As these indexes show, there is some variance denoting differences between countries.
AP Index (Difference).
The three lower scores are Brazil 2014, Costa Rica, and Chile showing that at the time of the elections, AP in these countries was very low for a variety of reasons. Surprisingly, the score for Peru is one of the highest along with those of Argentina in 2015, Brazil in 2018, and Mexico, particularly in 2018. Using Costa Rica as a base, Brazil 2018 is 5% higher, Argentina 10% higher, Peru 30% higher, and Mexico 2018 has a 70% higher API than Cosa Rica. Initial results signal a wide variance in these ten Latin American APIs (see more API scores in the Appendix, Table A1).
Spread
The spread index measures the total dispersion of AP in a given country/year. This is a convenient method to capture feelings toward multiple parties, which is the case in most Latin American democracies. The previous difference method treats all parties that are not the favorable one as out-parties, which in multiparty systems skew to some extent the polarization landscape since many people with affinities to several political parties, switch allegiance between them in different elections. Spread scores are very sensitive to the likeness of small parties. Larger scores signal higher polarization. Using the same CSES data, Figure 4A shows results for the entire sample, including individuals with missing values towards one or multiple parties. In the note to this figure, we present the percentage of respondents that have not voted in the recent election (presumably less politically motivated) which will illustrate, as we will see below, significant variance among the different samples. The statistical annex presents results for smaller samples of individuals who voted in the recent election. In addition, Figure 3 depicts a graphic representation of weighted spread and distance scores as well as difference scores.
These spread index scores show that most countries in Latin America do not depict, on average, a very large dispersion of animosity towards political parties. Spread scores tend to be lower because the presence of multiple small parties tames the overall cold feelings towards not preferred parties. Also, once the sample is not restricted to voters of the largest and opposite parties, the overall dispersion of likeability is smaller.
The following sections will demonstrate that the restricted samples consisting solely of voters who report initial preference for political parties, exhibit slightly higher polarization spreads compared to the entire survey sample, which also includes non-voters. Also, when populations are weighted, the total spread of polarization is usually larger
Distance
The distance method is conceptually closer to the difference method. It captures the average difference between the in-party liking towards other out-party feelings. Among other things, it differs from the difference method because the former includes a larger share of respondents and some sensitivity toward smaller parties. Figure 4B shows the results for the entire sample of respondents.
These scores indicate that AP fluctuates among the examples under study. Once again Chile (2017) has a low score (the sample includes a very low share of voters who show a strong affinity to their preferred party, resulting in a general resentment towards all parties). Peru, on the other hand, has one of the highest distance scores as well as some of Mexico’s elections. Surprisingly, Brazil has relatively low scores. In the following sections, we provide a plausible explanation of this score of Brazil.
Summary of Spread, Distance, and Difference Scores
Spread, distance, and difference are sensitive to different weights of voters’ share, rates of participation, and the size of the parties or coalition (measured as a percentage of votes received). Figure 3 displays a visual presentation that allows us to draw some initial conclusions. First, as expected, difference and distance scores are always higher than spread and they reflect more neatly levels of polarization. Second, despite using similar data, each method yields relatively different scores, yet they keep the same order and distribution. Footnote 17 Third, as mentioned, Peru, Mexico 2018, and Brazil 2018 show the higher levels of polarization whereas Chile is among the lowest (because in-like party scores are very low showing a general animosity towards all traditional political parties. This might have caused the October 2019 mass protests). Fourth, Brazil 2014 was not very polarized because a large share of the population did not score high in party likes. This changed in 2018.
In sum, although they use the same data, the three methods yield slightly different results, yet very consistent. In some countries and certain elections, the level of polarization has been noticeable, however excluding a few exceptions, API in most cases is moderate and very sensitive to the type of sample used to estimate. Since API is very much determined by the strength of difference in the valuations individuals give to their liked party and the level of animosity they have toward the non-favorable party, narrow differences can emerge from strong favorable ratings individuals assign to most parties (Costa Rica, Brazil 2014, Mexico 2012) or strong dislike of most parties (Chile).
Characteristics of Latin American AP
Who Drives AP?
A critical question for social analysts is to ascertain whether AP is a widespread social phenomenon or just a staple of “intense minorities” who drive up the perception of high social polarization. Mean scores of large samples provide only a partial perspective on aggregate sentiments, yet the CSES survey allows for further examination of this issue. Respondents were asked (v3018_3) whether they feel close to a particular political party. Footnote 18 Approximately half of respondents report that they feel warmth to one of the parties, while the other half did vote in the previous election but did not report close attachment to any party. Assuming that these last voters are less politically involved in the parties’ programs, ideological identification, or membership, they could be thought of as less party politically involved. We compare scores between this group of citizens (we will call them low political identification group or LPI) and those who report warmth towards a particular party (High political identification group or HPI). Footnote 19 In this section, we provide initial evidence that supports the hypothesis that AP is concentrated in a selected group of citizens.
Figures 5A and 5B depict spread and distance results for HPI and LPI across ten case studies. These scores show that intense AP is higher among those who report identification with a particular party, and they drive up the AP scores in each country by a significant margin. The differences are generally known. It is the larger spread found among different subgroups. In 9 out of the 10 case studies these differences are significant. For example, in Brazil, the HPIs’ distance score doubles those of LPIs, implying a strong polarization among the former. Similarly, this is the case of Chile and to some extent Mexico in 2018. The LPI mean score for most countries is 50%–75% lower than those who reported preference for a given party. Not only is difference among these groups noticeable, but it has also been found in all ten case studies.
AP Spread and Distance Scores.
Spread Scores for High and Low Politically Motivated Citizens. Distance Scores for High and Low Politically Motivated Citizens.
This finding offers an explanation towards the mixed results in Brazil, where the raw spread and distance scores are among the lowest. However, this is primarily due to the fact that the sample in this country includes more than 70% of people who report not having a preference for a party. Such a low proportion of HPI in the Brazil sample dwarfs the polarization score. Similarly, as in the case of Chile, the results in Figures 5A and 5B demonstrate that the difference between LPI and HPI are among the largest in these two countries.
This evidence is rather compelling. Individuals who identify with a party appear to be more mobilized and report stronger warmth towards their own liked party as well as stronger resentment towards nonpreferred and opposing parties. The strong difference between the two subgroups (party and non-party identifiers) allows us to hypothesize that AP is driven by mobilized subgroups who intensely favor in-parties and despise out-parties. We hypothesize that the larger the population with HPI within a country, the higher the AP score will be. Future research will delve into these findings.
Is AP Growing?
AP appears to be rising in the region. For this section we only analyze countries where multiple surveys were conducted and questionnaires were similar. Waves 4 and 5 of CSES have data for multiple surveys only for Brazil and Mexico. In these two nations, the observed polarization trend is clearly upward. Table 1 presents the difference and distance API scores of party identifiers for these two nations. In the case of Brazil, data is only available for elections held in 2014 and 2018. Footnote 20 For Mexico, these surveys (with some minor modifications) were held since 1997 yet we only present results for presidential elections (2000, 2006, 2012, and 2018).
Evolution of API (Average Scores among Respondents Who Reported Party Identification)
Age Groups API Distance Weighted
Despite very few observations and a smaller share of people who identify with political parties, it appears that AP is deepening in the region, although we lack the data to conclusively prove this pattern. Footnote 21 In Mexico, the score of the difference method increased by two full points, from 3.92 to 5.92. Only the results for the 2012 election show a moderating trend. Footnote 22 In the case of Brazil, scores also increased albeit more moderately compared to Mexico yet in a shorter period (four years). New waves of CSES data for recent years will allow us to test the upward trend of the polarization hypothesis.
Where does Latin America’s Polarization Stand?
The use of similar data and method allows for a comparative perspective of polarization among the publics of other Western countries and the available cases in Latin America. Figure 6 signals where Latin American API cases stand and seeks to respond to the question of whether Latin American citizens are more, less, or similarly affectively polarized as other Western nations?
Latin American countries, at least for the CSES 4 and 5 survey waves of previous decade, are not highly affectively polarized as compared to other European nations. These results show that a large share of Latin American nations’ API score is higher than the Netherlands, but in the same range as Iceland, Finland, Germany, Switzerland, or Denmark, and much lower than for example Spain, Greece, and Portugal, and other eastern European nations. Yet a few observations (Mexico and Peru) have above-average API scores. In short, AP in most countries of Latin America, at least during the 2010s decade, had AP levels similar to some mid-European nations.
A Glimpse into Demographics
Initial analysis of data on personal attributes shows that for the most part gender and age do not explain different levels of AP (see Table 2, and Table A2 in the appendix). We have produced scores stratified by gender for all cases (see Figure 7) and in most of them, the aggregate level of API by gender does not show a statistically significant difference. Yet, in the majority of them, scores of females are slightly higher probably denoting marginal higher AP. This suggests, as in many cases in Europe (Torcal and Comellas Reference Torcal and Comellas2022), that among polarized groups, females appear to depict slightly more affinity with in-parties and animosity toward other parties. On average, the API spread score for males is 2.37, while for females is 2.41. For the ten case studies, however, four cases (including both in Brazil) have males slightly more polarized than females. Although AP is more present among females, the differences in the vast majority of cases are not statistically significant.
API for European and Latin American Countries: Difference Method.
API Scores by Gender (Respondents Who Voted in First-round Elections).
In terms of age, we inquire whether AP is more intense at given age categories and whether API correlates with respondents’ age. Here we find a trend of higher API with older age that holds for most cases (with the only exception being Brazil in 2018). There is however an interesting finding worth exploring. The higher API aggregate scores are in the 41–60 age group declining slightly in eight of our ten case studies among individuals over 60. In short, the data indicates that it appears to be a moderate positive association between individual valuations of in-party and out-party and the age of respondents. In Latin America, older voters tend to be more polarized than younger voters.
Discussion
In this first article, we analyze survey data to provide a quantitative magnitude of AP in the region. Far from introducing conclusive and causal explanations, we present measurements on the extent and depth of political animosity in six Latin American countries over ten election cycles.
The polarization map in the region is rather complex. We rely on representative surveys and develop three measurements that capture different dimensions of AP. Despite some inconsistencies, the data for the previous decade indicates that Latin American publics are on average moderately polarized. The difference between their favored party and the non-favored party is not as large as in many countries in Europe and elsewhere, yet the polarization is considerable in most cases.
This article makes a very important inroad by showing that mean scores are very sensitive to the different methods that estimate and score API. They depend on the type of samples they rely on, and whether different weights are used. This indirectly proves that there is a wide variance in political sentiments within each country that needs to be studied. Affective polarization of individuals in all countries is not uniform nor homogenous but somewhat very diverse.
Some results are very intriguing. Peru stands out as a very polarized country and its scores are among the highest in the region. On the other hand, Chile’s API scores are very low, due to the general hostility towards all political parties, which produces very low scores for in-party favorability. Brazil 2014 also has an unexpectedly low score which is probably due to a small proportion of individuals who identify with a political party, as well as considerable sympathy towards the Marina Silva candidacy as a large third-party option.
However, as previously mentioned, the interpretation of scores should be contextualized because they are derived from individual association with partisanship that varies significantly across the region. Indeed, for countries such as Peru that is included in our sample along with Ecuador and most central American nations that are not, lack strong traditions of political party association and affiliation. Moreover, political parties in these countries tend to be weak supporters of political stability (Levitzky and Zabaleta Reference Levitzky and Zabaleta2018; Roberts Reference Roberts2002, Reference Roberts2015; Mainwaring Reference Mainwaring2018). Conversely, countries such as Argentina, Mexico, and Chile have strong and long established political parties, where citizens might have stronger identification or resentfulness towards them.
Other measures of polarization should be pursued to complement those presented here. To the extent that AP is tied to partisanship it’s crucial to determine whether high or low scores stem from liking or disliking specific parties. This aspect has been addressed by presenting results of political likeability in the first section. Additionally, it’s important to consider whether the strength of political parties may influence scores. For example, despite the strength of Chile’s parties, citizens may have expressed high levels of animosity and low levels of affinity towards all parties, possibly in anticipation of the 2018 crisis, which could have impacted the overall scores. Future multivariable analysis should control for the effect of partisanship in each country to assess whether political traditions marginally affect API scores.
In this last section, we highlight four important contributions this article makes:
First, and perhaps the most significant finding, is that in Latin America AP is significantly more intense among a reduced fraction of the population. Although further research will estimate the size of these groups, it is very clear that non-party identifiers are not as polarized as intense minorities who identify with a party, and they are politically motivated.
Second, although there are very few observations, the data indicates that AP in the region is rising. In the two countries that have multiple waves, the upward trend is very clear.
Third, API in the region does not appear to be very high compared to other parts of the world. Although we lack data for recent polarized elections (particularly those for Brazil 2022, Chile 2021, or Peru 2021) the data for the previous decade show a rising but moderate polarization.
Finally, the first demographic variables such as gender and age do not appear to decisively drive AP. There is very weak evidence that gender accounts for some variability in scores, and age appears to have a moderate effect that positively correlates with AP.
The Road Ahead
Affective polarization is a social construct that has received increasing attention over the last decade. Despite a growing interest in Latin American polarization, there are very few empirical studies on the range and depth of animosity towards political adversaries in the region. Despite different narratives found in many countries over the wedge for animosity and resentfulness towards political opponents, few attempts have been made to systematically measure and draw conclusions over AP and its effect on the quality of democracies (recent contributions such as Segovia (Reference Segovia2023) for Chile, and Castro Cornejo (Reference Castro Cornejo2023) for Mexico, have started to fill the gap). This research note invites at least four research agendas for the coming years that are relevant to Latin American politics and society.
First, while the causality between the erosion of democracy and the rising trend of AP remains to be determined, initial analysis reveals a strong association between these social and political phenomena, as both indicators are increasing, suggesting a positive correlation. Somer et al. (Reference Somer, McCoy and Luke2021) find a strong correlation between democratic erosion and high levels of polarization using global V-Dem data. Although further research is needed to explore this association in Latin America, our data on AP and support for democracy from LAPOP and Latinobarómetro also supports the hypothesis of a strong positive correlation. It is crucial for future research to assess the impact of each variable in the region.
Second, the emerging literature on AP in Europe studies the effect of institutional features and their impact on AP. Special interest focuses on parliamentary and presidential regimes, voting systems, and party strengths. Latin America has a variety of different regimes, including political parties and coalitions, compared to Europe. A research agenda should study how these institutional variables affect the level and intensity of AP.
A third research agenda should inquire into the array of determinants of AP. The findings of this article signal that there is a wide variance in individuals’ AP levels. Research initiatives should inquire whether Latin American nations have widely polarized societies or vociferous minority segments of society that drive up polarization while attempting to measure the extent and intensity of these groups.
Finally, research should attempt to study in depth the determinants of AP. Using individual rather than aggregate data, studies should focus on an array of sociodemographic, institutional, emotional, communicational, and many other variables that might explain the causes of AP. Identifying these variables may provide valuable insights for a better understanding of the dissatisfaction with democracy in the region.
Competing interests
Marcelo Bergman and Pablo Fernández declare none.
Appendix
Scores for the Entire Sample and Restricted Samples of Voters Who Voted in First-round Elections
Spread Scores by Age
