INTRODUCTION
Since the 2010s, numerous studies have taken a novel approach to party polarization, concentrating on partisan feelings. At their center is the concept of affective polarization, defined by the extent to which voters have positive sentiments for their own party, while holding negative ones toward competing out-party/parties (Iyengar, Sood, and Lelkes Reference Iyengar, Sood and Lelkes2012). Although the topic has mostly been studied in the USA, recent evidence has shown that affective polarization is also present in parliamentary and multiparty systems, sometimes to an even greater extent (Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020; Reiljan Reference Reiljan2020; Wagner Reference Wagner2021).
At the same time, a growing body of literature has documented the personalization of modern politics, as party leaders and candidates have become increasingly central for processes of political competition (Garzia, Ferreira da Silva and De Angelis Reference Garzia, da Silva and Angelis2021; Lobo and Curtice Reference Lobo and Curtice2014; Rahat and Kenig Reference Rahat and Kenig2018). While candidates have always been prominent in presidential regimes, a trend toward presidentialization has also been detected in parliamentary democracies (Poguntke and Webb Reference Poguntke and Webb2007). Empirical manifestations of this can be found in the enhanced relevance of candidates in the media (Langer Reference Langer2007), internal party organization, and voting decision processes (Ferreira da Silva Reference Ferreira da Silva2021). For many voters, leaders have now become the main reference through which they relate to politics (Garzia, Ferreira da Silva, and De Angelis Reference Garzia, da Silva and Angelis2021).
Studies on affective polarization and personalization of politics have thus far remained distinct from one another, although the two concepts appear to be partially connected. Recent contributions reveal that both positive and negative attitudes toward leaders appear in the United States and in several European democracies (Garzia and Ferreira da Silva Reference Garzia and da Silva2021; Webster and Abramowitz Reference Webster and Abramowitz2017), suggesting the presence of affective polarization over leaders, that is, “leader polarization” (Bordignon Reference Bordignon2020). However, the existent literature on affective polarization has defined the concept almost exclusively in terms of party affect or feelings toward out-group members, and little is known about voters’ polarized feelings toward political leaders, especially outside of the United States. This is somewhat surprising, considering that several developments that have been linked to higher affective polarization—such as increasingly negative campaigning practices and the rise of social media—prime hostility mostly at the candidate level (Auter and Fine Reference Auter and Fine2016; Druckman et al. Reference Druckman, Klar, Krupnikov, Levendusky and Ryan2021; Pruysers and Cross Reference Pruysers and Cross2016). Moreover, several studies have found that affective polarization among mass publics originates at the elite level (Banda and Cluverius Reference Banda and Cluverius2018; Rogowski and Sutherland Reference Rogowski and Sutherland2016).Footnote 1 Therefore, improving our knowledge about leader affective polarization (LAP) could also help us to better understand the origins of party affective polarization (PAP).
The present study addresses this gap by contrasting political parties and leaders as objects of affective polarization in contemporary democracies. Our first goal is to offer a broad cross-national comparison of LAP and PAP. Data from the Comparative Study of Electoral Systems (CSES) project allows us to study 40 democracies over a time span of more than two decades, to investigate the extent to which leaders and parties can be objects of affective polarization (The Comparative Study of Electoral Systems 2020; 2022). Drawing on the growing (party) affective polarization literature, we apply an identical empirical index to measure both concepts, enabling their direct comparison, and determining which type of affective polarization primes over the other. Our second goal is to develop (and empirically test) a theoretical framework to explain the varying patterns of LAP/PAP across contexts. Building on classic contributions from the literature on political institutions, and particularly on the work by Arend Lijphart and Juan Linz, we argue that institutions of democracy matter. Rather majoritarian and/or presidential contexts favor more leader-based affective polarization, while the polarizing appeal of leaders is lower in multiparty parliamentary democracies. We also study the impact of factors relating to regime input and output, claiming that better government performance leaves less potential for conflict over specific leaders.
The article proceeds as follows. First, we clarify some crucial conceptual matters, and place LAP and PAP into the broader, multidimensional framework of affective polarization. Next, we offer an overview of both the literature on affective polarization and the personalization of politics, focusing on the potential points of contact between the two. Subsequently, we outline our theoretical expectations regarding the relative degrees of LAP and PAP across countries. Then, we introduce the data and operationalization strategy. Our first empirical section offers a descriptive overview of LAP vis-à-vis PAP across 40 democracies. The second empirical section presents regression models that scrutinize the difference in the predictors of LAP and PAP, to get a grip on cross-national variations. Finally, we summarize our findings and discuss their theoretical and empirical implications.
LAP AND PAP WITHIN THE MULTIDIMENSIONAL FRAMEWORK OF AFFECTIVE POLARIZATION
Although research on affective polarization has proliferated over the last decade, surprisingly little attention has been dedicated to conceptual issues. Thus, it is not unequivocally clear what the concept of affective polarization precisely does entail and what kind of distinct manifestations it has. When Iyengar, Sood, and Lelkes (Reference Iyengar, Sood and Lelkes2012) brought the topic of affective polarization into the spotlight, they lumped together several indicators that tap into distinct aspects of the concept: party and partisan thermometer scores, social distance measures (e.g., not accepting an out-party in-law) and partisan-based stereotyping. They noted that feelings toward parties and party supporters are not completely congruent, despite demonstrating a strong correlation between the two measures and concluding that “people perhaps extend their dislike of the parties to dislike of people within the parties” (415). However, most of the subsequent studies have not dealt with the potential lack of conceptual clarity.
Still, a few exceptions stand out. They emphasize the importance of distinguishing between attitudes toward parties as political objects, and behavioral outcomes outside the political sphere, for example, the perceived social distance between party supporters (Druckman and Levendusky Reference Druckman and Levendusky2019; Kingzette Reference Kingzette2021; Rudolph and Hetherington Reference Rudolph and Hetherington2021). Affective polarization has usually been measured using party feeling thermometers (Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). Most of the comparative literature has so far relied on an equivalent of the feeling thermometer, that is, the party like–dislike scale (Boxell, Gentzkow, and Shapiro Reference Boxell, Gentzkow and Shapiro2022; Garzia, Ferreira da Silva, and Maye Reference Garzia, da Silva and Maye2023; Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020; Reiljan Reference Reiljan2020; Wagner Reference Wagner2021). Implicit to these measures is the assumption that political parties, as broad, abstract, encompassing entities, are the main objects of partisan affect and disdain. Whether these feelings also manifest themselves in, for example, social interactions between partisans, is a different question. Yet, this distinction has not received sufficient attention and there is no consensus on what is, or what is not affective polarization, and which indicators are most suitable for measuring the concept.
We propose that affective polarization should be treated as a multidimensional concept. At the most general level, we can distinguish between political and social manifestations of affective polarization. The first entails feelings that are closely linked to the political sphere, whereas the second manifestation is tied to society more broadly. The social aspect of affective polarization is more in line with the original conceptualization by Iyengar, Sood, and Lelkes (Reference Iyengar, Sood and Lelkes2012), which builds on social identity and intergroup conflict theories, and perceives partisanship as a social identity. Conversely, as argued by Ruckelshaus (Reference Ruckelshaus2022), partisan identification can also remain within the political sphere, without collapsing into a broader social/cultural identity. Consequently, partisan affective polarization could also be limited to the political realm. This is what we try to tackle in our contribution, which focuses on how voters view political parties and their elites, rather than how they think of partisanship in social terms and affect toward fellow citizens. The two types of affective polarization are expected to be connected; yet, they still constitute distinct manifestations of affective polarization and a strong correlation between the two is not axiomatic (Druckman and Levendusky Reference Druckman and Levendusky2019; Knudsen Reference Knudsen2021).
In this contribution, our focus lies on the political manifestations of affective polarization. We claim that another distinction is necessary to adequately map the concept. When thinking about parties, people can perceive them as broad organizations, with their own brands and symbols, ideological standpoints, and programs. At the same time, people can also perceive parties as narrow groups of political elites, with party leaders being their most prominent personal representatives. Leaders can have their own distinct symbols and brands, styles, and views. While we can assume that feelings toward parties as organizations and party leaders are related to each other, it is by no means a given that the two are congruent. Theoretically, polarized feelings toward parties and leaders form distinct subconcepts of affective polarization, and their relationship is subject to empirical scrutiny. Thus, we propose that the political aspect of affective polarization should be divided into PAP and LAP. Figure 1 presents our theoretical discussion and maps affective polarization as a multidimensional concept.
Figure 1. A Conceptual Map of Affective Polarization
LAP AND THE PERSONALIZATION OF POLITICS
Comparative literature has by now accumulated an important amount of knowledge about the political aspect of affective polarization, but it has almost exclusively focused on PAP. Polarization of feelings toward party leaders, conversely, has not received systematic attention outside the United States. This is somewhat surprising, considering that a number of recent studies in the US context have revealed that it is mostly top politicians that people have in mind when they are asked to express their feelings about political parties (Druckman and Levendusky Reference Druckman and Levendusky2019; Kingzette Reference Kingzette2021). This suggests that the political aspect of affective polarization might first and foremost be a manifestation of attitudes toward a narrow group of political leaders rather than political parties as broader collective actors. Yet, this should be verified in a comparative perspective.
The importance of elites in shaping people’s attitudes about parties refers to the personalized nature of modern politics. Over the last decades, the prominence of individual actors in politics has increased, while at the same time, that of the parties has declined (Rahat and Kenig Reference Rahat and Kenig2018). This decline in the importance of parties is visible in the substantial drop in the levels of partisanship in many Western democracies, as less and less people identify with a specific party (Garzia, Ferreira da Silva, and De Angelis Reference Garzia, da Silva and Angelis2022; Huddy, Bankert, and Davies Reference Huddy, Bankert and Davies2018),Footnote 2 let alone become members of a party (van Biezen and Poguntke Reference van Biezen and Poguntke2014). Consequently, voters’ evaluations of party leaders have gained increasing relevance in determining vote choice, at the expense of partisanship whose magnitude is generally seen as weakening. Nowadays, voters are more willing to vote for the leader they like the most, regardless of their partisanship (Garzia, Ferreira da Silva, and De Angelis Reference Garzia, da Silva and Angelis2021).
The concepts of partisanship and personalization are predominantly studied from the perspective of a “positive” attachment that focuses on identifying with parties or liking leaders. More recently, however, and in the coattail of affective polarization research, growing attention has been assigned to negative partisanship, that is, identifying specifically “against” some party (Abramowitz and Webster Reference Abramowitz and Webster2016; Bankert Reference Bankert2021; Medeiros and Noël Reference Medeiros and Noël2014). Correspondingly, the concept of negative personalization has also developed in the literature (Garzia and Ferreira da Silva Reference Garzia, da Silva and Angelis2021; Pruysers and Cross Reference Pruysers and Cross2016). In practice, negative campaigning is now a common tactic not only in US politics but also in European democracies (Valli and Nai Reference Valli and Nai2020), and several authors have linked it to higher levels of affective polarization (Hetherington and Rudolph Reference Hetherington and Rudolph2015; Iyengar, Sood, and Lelkes Reference Iyengar, Sood and Lelkes2012; Webster and Abramowitz Reference Webster and Abramowitz2017). Moreover, some findings indicate that attack ads tend to target mostly political candidates rather than parties, and focus on the personal aspects related to the candidates, as opposed to substantive political issues (Geer Reference Geer2006; Pruysers and Cross Reference Pruysers and Cross2016). This behavior is even more present in digital media, where candidates involved in highly competitive races tend to resort more frequently to personal attacks (Auter and Fine Reference Auter and Fine2016). Thus, considering the predominantly candidate- and personality-based nature of negative campaigning and its relationship with affective polarization, it is reasonable to presume that, beyond political parties, affective polarization can also transpire to candidates themselves and that the feelings toward parties and leaders/candidates are not fully congruent.
While the simultaneous presence of positive and negative partisan feelings—that is, affective polarization—has been under growing scrutiny since the seminal article by Iyengar, Sood, and Lelkes (Reference Iyengar, Sood and Lelkes2012), much less attention has been paid to expressions of affect toward party leaders/candidates. Some recent work has done so, however. Results for the United States, unsurprisingly, indicate that feelings about presidential candidates are strongly polarized along party lines (Webster and Abramowitz Reference Webster and Abramowitz2017). Albeit scarcer, research outside of the American context also suggests the existence of LAP. Garzia and Ferreira da Silva (Reference Garzia, da Silva and Angelis2021) demonstrate that voters in 14 Western European parliamentary democracies assign significantly higher levels of affection to the leader of the party they support, while exhibiting negative feelings toward other party leaders. Moreover, they found that for vote choice, negative assessments of out-party leaders have become, in recent elections, as important as positive feelings toward in-party leaders. Studying the cases of Italy and Germany, Barisione (Reference Barisione2017, 623) detected a significant gap in in-/out-party leader evaluations, which the author argues “is both a powerful manifestation and indicator of political polarization in multiparty systems, and it may be effectively used as a leader-based measure of partisan polarization.” In an Italian case study, Bordignon (Reference Bordignon2020) also finds strong evidence of polarized feelings over party leaders. Hence, the literature offers some initial proof that LAP also surfaces in parliamentary, multiparty democracies.
Although these early works indicate the presence of LAP, we still lack some very basic knowledge about polarized opinions toward party leaders outside the United States. No broad, cross-national comparative analysis of LAP has hitherto been carried out; thus, we do not know which political systems are the most/least polarized in terms of feelings toward party leaders. Nor is it known how LAP compares to and correlates with the classic form of PAP, or how the predictors of the two differ from each other. This is the task at hand and in the following sections, we will offer a novel descriptive and analytical understanding regarding these questions.
PARTY AND LEADER AFFECTIVE POLARIZATION: THEORETICAL EXPECTATIONS
The previous discussion calls for an inquiry into the patterns and determinants of party and leader affective polarization in contemporary democracies. Foundational research in political behavior suggests a significant amount of endogeneity between party and candidate/leader evaluations (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Fiorina Reference Fiorina1981), which is virtually impossible to disentangle with classic cross-sectional survey designs, and this is also not the aim of this manuscript. We concur with the interpretation of Barisione (Reference Barisione2017, 624) that there likely is a continuous feedback between feelings toward parties and leaders, and that the two sets of attitudes reinforce each other. However, while we can expect a substantial overlap between the two measures, we also anticipate that their relative strength shall be contingent on relevant structural, political, and institutional conditions. Our research focus is laid on the factors that account for the potential differences between the two constructs, rather than looking at PAP and LAP in isolation. In this section, we outline our theoretical expectations regarding the relationship between these two manifestations of affective polarization.
First, it is important to clarify our definition of LAP and PAP. We use the classic definition of affective polarization as the divergence in affect toward the in- and out-parties (Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020; Iyengar and Westwood Reference Iyengar and Westwood2015).Footnote 3 This approach was adjusted for a multiparty context by Reiljan (Reference Reiljan2020) who conceptualized affective polarization as the average (weighted) like–dislike difference between an in-party and all relevant out-parties.Footnote 4 We apply the same definition to LAP: the higher the divergence between the evaluation of the leader of one’s preferred party and the leaders of all other parties, the higher the degree of LAP. Accordingly, levels of LAP would be higher than those of PAP if voters rate the leader of their preferred party more positively than the party itself and/or evaluate the leaders of out-parties lower than the parties.
Regarding which type of affective polarization should prevail over the other, the existing literature points in different directions. On the one hand, research in social psychology has previously found evidence of a person positivity bias, meaning that, because of individuating information, people tend to have more favorable feelings toward individual members of some out-group, compared to the group itself (Iyengar, Sood, and Lelkes Reference Iyengar, Sood and Lelkes2012; Sears Reference Sears1983). Therefore, we could expect that voters exhibit less hostility toward the leaders of out-parties, which would predict lower levels of LAP as compared to PAP. On the other hand, there is also evidence of a person negativity bias, as in several US elections, especially the more recent ones, out-party candidates were perceived more negatively than the party itself (Bolsen and Thornton Reference Bolsen and Thornton2021). Bordignon (Reference Bordignon2020) demonstrates the specific polarizing drive of certain politicians in Italy, especially of the controversial populist leaders Berlusconi and Salvini, suggesting that some leaders serve as “super-issues” in the political system and personal divisions can form around them. Similar leader-related divisions of public opinion have also been described in several South American countries, for example, Peronism/anti-Peronism in Argentina and Chavismo/anti-Chavismo in Venezuela (McCoy and Somer Reference McCoy and Somer2019; Mudde and Rovira Kaltwasser Reference Mudde and Kaltwasser2018). Our aim is to determine which factors facilitate a situation in which most intense feelings form around leaders/candidates, and which conditions rather push toward more party-based affective polarization.
HYPOTHESES
Such a first cross-country exploration of the two forms of affective polarization needs a theoretical compass, helping us navigate the murky empirical waters one inevitably has to face. Our compass is made up of three building blocks that are derived from the literature on PAP: regime input, regime output, and institutions. Regarding each block, we outline specific hypotheses concerning their relationship with PAP, LAP, and the relative strength of LAP vis-à-vis PAP (i.e., the LAP/PAP ratio).
With regime input, we mean a set of determinants that are linked to citizens and political parties. A first predictor of both LAP and PAP, and of their respective ratio, is structural in form, relating to the sheer stronghold on citizens by political parties, that is, the collective level of partisan identification in a society. In a context where political conflict is predominantly articulated by parties and where citizens identify with these parties, the playground for PAP to emerge is set. On the contrary, in contexts characterized by “parties without partisans” (Dalton and Wattenberg Reference Dalton and Wattenberg2002), where political dealignment and the hollowing-out of democracy are on the rise (Mair Reference Mair2013), PAP has a tougher stand, simply because there are less partisans around in the first place. That partisanship correlates with higher levels of PAP has been already shown by several studies (Iyengar, Sood, and Lelkes Reference Iyengar, Sood and Lelkes2012; Mason Reference Mason2015; Reiljan and Ryan Reference Reiljan and Ryan2021; Wagner Reference Wagner2021), although this relationship has not been thoroughly assessed at a system level. However, individual-level studies clearly suggest that if more voters have (strong) partisan identities, the level of PAP is elevated.
Regarding LAP, it is reasonable to expect that feelings toward party leaders are also more polarized in a society where the stronghold of parties on citizens is more extensive, considering that leaders are the most visible symbols of parties. Yet, if voters are nowadays more willing to vote for their favorite leader even if it conflicts with their partisan identity (Garzia, Ferreira da Silva, and De Angelis Reference Garzia, da Silva and Angelis2022; Reference Garzia, da Silva and Angelis2021), we could expect that leader evaluations are also more independent from partisanship. Voters who do not identify with any party, but are still participating in the electoral process, could display more polarized feelings around different leaders than parties. Conversely, those who identify with a party, are more likely to hold stronger feelings toward parties, rather than the specific individuals that are currently leading these parties. Thus, the level of partisan identification could have a weaker relationship with LAP than PAP, and in a more partisan society, PAP should prime over LAP, that is, the LAP/PAP ratio is lower. Our first set of hypotheses can therefore be derived as follows:
H 1.1. The higher the overall level of partisan identification, the higher the levels of PAP and LAP, but this relationship is stronger with PAP;
H 1.2. The higher the overall level of partisan identification, the lower the ratio between LAP and PAP.
A second regime input determinant that we focus on in this article is political in form and relates to the ideological structure of the party system. From the comparative literature, we know that higher ideological (left–right) polarization in the party system correlates with higher levels of PAP (Gidron, Adams, and Horne Reference Gidron, Adams and Horne2022; Reiljan Reference Reiljan2020; Wagner Reference Wagner2021). It is sound to assume that the same is the case for LAP. However, there are reasons to believe that the effect of ideological polarization on the latter could be somewhat weaker/inconsistent. As mentioned above, there are cases where a strong division over some political figures emerges in the political competition. Although such divisions could overlap with ideological dimensions (e.g., liberals are much more likely to loath Trump), they could also be related to personal characteristics (e.g., style, personal history) of specific leaders. As such, leaders can themselves become political issues and could, in some cases, even be seen as functional alternatives for ideologies (Bordignon Reference Bordignon2020, 4). Thus, we hypothesize that leader evaluations could be more independent from ideological considerations than feelings toward political parties. Accordingly, our second set of hypotheses can be formulated as follows:
H 2.1. The higher the ideological polarization in a political system, the higher the level of affective polarization, but this relationship is stronger with PAP;
H 2.2. The higher the ideological polarization in a political system, the lower the ratio between LAP and PAP.
Our second theoretical building block relates to regime output. The literature on satisfaction with democracy and institutional trust has established that government performance affects citizen’s attitudes toward the entire regime, as lower output associates strongly with lower satisfaction and trust (Dahlberg and Holmberg Reference Dahlberg and Holmberg2012; Reference Dahlberg and Holmberg2014; Torcal and Trechsel Reference Torcal, Trechsel, Ferrín and Kriesi2016; Wagner, Schneider, and Halla Reference Wagner, Schneider and Halla2009). However, this relationship could be more nuanced: Reiljan (Reference Reiljan2021) shows that more effective government covaries with lower levels of PAP, as government performance has a stronger effect on the attitudes toward out-parties. It appears that voters ascribe more blame for bad government performance to their political opponents, while being more permissive toward their own preferred party. Thus, we can expect a significant relationship between regime output and PAP.Footnote 5
It is reasonable to assume a similar link between regime output and LAP to exist. At the same time, we also posit that the ratio between LAP and PAP could be affected by regime output. As mentioned above, divisions about leaders have developed around several controversial figures in various countries. However, this rather appears to happen in cases of weak government performance, giving an incentive for leaders to arise and promise to “fix the system” (e.g., Second Italian Republic, South American countries). In countries with a well-functioning and impartial government sector, there should be less leverage for personal agency and the development of such leader-based divisions should be less likely. Thus, we assume that while quality of governance has a taming impact on both LAP and PAP, the effect is even stronger on the former, resulting in a lower LAP/PAP ratio.
H 3.1. The better the quality of governance, the lower the level of affective polarization, but this relationship is stronger with LAP;
H 3.2. The better the quality of governance, the lower the ratio between LAP and PAP.
Finally, the comparative affective polarization literature has shown the importance of institutions. Arend Lijphart, has famously introduced the difference between majoritarian and consensus democracies, where corresponding institutions correlate with outcomes such as representativeness, welfare, and other redistributive policies. Lijphart (Reference Lijphart1999) distinguishes between two major dimensions: the executive-parties dimension and the federal-unitary dimension. While the latter is mainly related to the type of construct of a state, the former dimension more directly deals with elements linked to electoral systems, party systems, and power concentration in the executive. These elements may help us better understand the variance in PAP and LAP across democratic regimes.
One of the central variables of Lijphart’s executive-parties dimension is party system fragmentation. More fragmented party systems, populated by numerous relevant parties, generally go together with lower levels of PAP (Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020; Reiljan Reference Reiljan2021). Two parties dividing the entire cake almost naturally must compare to—and compete with—each other. Election campaigns become ever-recurring derbies between the two major protagonists. In larger party systems, however, parties usually need to form coalition governments, which induces cooperation. Gidron, Adams, and Horne (Reference Gidron, Adams and Horne2020) have demonstrated that this also reflects in partisan feelings, as voters are significantly less hostile toward parties that have been in the same governing coalition with their preferred party, even when controlling for other relevant variables such as ideological distance between parties.
We also argue that the simple form of party competition—us versus them—that is evident in party systems with a limited number of electorally relevant parties may lead to higher visibility of the leaders and, hence, to higher levels of LAP. It is not only us versus them, but also her/him versus him/her. In the aftermath of such elections, the leader/candidate of the winning party is (almost) certain to assume the highest political office in the country. In party systems with a larger number of relevant parties, conversely, the number of different coalition scenarios is much larger and winning the elections does not guarantee that the leader of the party becomes the head of government, or even gets into government at all. Thus, regarding the ratio, we predict that the more fragmented the party system, the lower the relative weight of LAP compared to PAP.
H 4.1. The more fragmented the party system, the lower the level of affective polarization, but this relationship is stronger with LAP;
H 4.2. The more fragmented the party system, the lower the ratio between LAP and PAP.
This, however, is not the end of the “institutional story.” Another highly crucial institutional feature is the form of government. The arguably most common dichotomy of types of government distinguishes between parliamentary and presidential systems. The nature of the political competition in presidential systems should induce personalized politics to a much higher extent than parliamentary forms of government. In fact, Lijphart does not use this classic regime typology for his executives-parties dimension, mainly for theoretical reasons. Besides the effective number of parliamentary parties, the dimension encompasses “minimal winning one-party cabinets,” “electoral disproportionality,” “interest group pluralism,” and “executive dominance,” thus offering a much finer-grained institutional description than the simple distinction between presidential and parliamentary systems (Bormann Reference Bormann2010). For our purposes, the inclusion of such elements into our theoretical framework is less straightforward. For the reason of a clear connection between presidentialism and personalization, we therefore complement the “Lijphartian” approach, condensed to party system fragmentation (actually making it, in its essence, become a “Taagepera-Laaksonian” one), with a more “Linzian” one, making use of the classic juxtaposition between presidential and parliamentary systems (Linz Reference Linz1990).Footnote 6
Reiljan (Reference Reiljan2021) finds that, when controlling for a broad range of relevant variables, presidential systems exhibit lower levels of PAP than parliamentary regimes. This could be explained by the strong importance of the presidency, which could trigger parties to more intensively reach out to other parties to have a chance to gather at least 50% of the overall votes for their candidate (see Curini and Hino Reference Curini and Hino2012 for a similar discussion regarding the effect of presidentialism on ideological polarization). As presidentialism nurtures personalization more than parliamentarism, we expect that this negative correlation with PAP is much lower (or even reversed) regarding LAP. Thus, we assume that presidential systems induce affective polarization rather between leaders than parties, which should lead to a higher LAP/PAP ratio.
H 5.1. Affective polarization is lower in presidential systems, but this relationship is stronger with PAP;
H 5.2. The ratio between LAP and PAP is higher in presidential systems.
The previously outlined hypotheses are summarized in Table 1, where we list the expected directions of effects on PAP, LAP, and the LAP/PAP ratio.
Table 1. Expected Predictors of Party Affective Polarization (PAP), Leader Affective Polarization (LAP), and the LAP/PAP Ratio: Summary of Hypotheses
Note: LAP/PAP refers to the ratio between the two constructs, that is, “+” indicates that the variable is expected to associate with higher LAP vis-à-vis to PAP, whereas “−” signifies the opposite.
DATA AND METHODS
This manuscript relies on the Comparative Study of Electoral Systems (CSES) dataset, which is used to calculate both PAP and LAP. It is the only available data source that allows us to do so across time and space in such an encompassing way. We use data from CSES Modules 1 to 5, except Wave 2 (covering the years 2001–2006) which did not include the party leader like–dislike item, which is the basis of our LAP measure.Footnote 7
All the available countries where (at least partly) free competitive elections take place are incorporated in the sample.Footnote 8 To ensure valid comparisons across countries, we also set certain criteria regarding the type of election. First, we must have an estimate of party vote shares to calculate PAP and LAP (and ideological polarization). Therefore, solely presidential elections were not suitable, as then votes are assigned to candidates, not to parties. However, we included the cases (such as the United States and several South American countries) where presidential and legislative elections were held simultaneously. Secondly, the highest political office (either Prime Minister or President, depending on the regime type) had to be at stake in the elections. Thus, we did not include the elections in presidential systems where only the legislative assembly, but not the president, was elected.Footnote 9 Finally, we selected only the cases for which data were available for each variable in our regression model. Eventually, that left us with a sample of 102 elections from 40 countries, covering the period stretching from 1996 to 2019.Footnote 10
The number of elections included from different countries is uneven and varies from one to six. This allows us to capture some trends over time. However, most of the variance regarding all our main variables can be found between countries, rather than within countries over time (see Table A.4 in the Supplementary Material for an intraclass correlation analysis). Thus, it becomes apparent that the observational relationships we detect are mainly driven by cross-national rather than within-country variations. Moreover, as one country-year (e.g., Iceland 2009) represents one case in the regression models, we have to consider that not all cases are independent (e.g., Iceland 2013 is correlated to Iceland 2009). To address this problem, we use cluster-corrected robust standard errors to report the statistical significance of the coefficients in our OLS regression models and also run two alternative models to check for the robustness of our findings: first, we replicate the results with country average scores (one country = one case), and, second, we restrict the sample to the countries that are represented with more than one election and run the model with just two (most recent) elections from each of the countries, thus eliminating the problem of an uneven country sample (see Tables A.5 and A.6 in the Supplementary Material).
VARIABLE OPERATIONALIZATION
Dependent Variables
Party and Leader Affective Polarization
We rely on the classic like–dislike item to operationalize the central dependent variables of this study. The CSES dataset not only includes the party like–dislike question (measured on a 0–10 scale), but also an identically framed question regarding party leaders. This allows us to measure LAP and PAP in a directly comparable way, which is crucial for the aims of this study. As explained earlier, our study focuses on the political dimension of affective polarization. Thus, these items fit our aims well, but it should be kept in mind that they do not necessarily say much about the more social manifestations of affective polarization.
We use the Affective Polarization Index (API) developed by Reiljan (Reference Reiljan2020) to measure polarization in party and leader evaluations. API indicates the average divergence of partisan affective evaluations between in-party and out-parties, weighted by the electoral size of the parties.Footnote 11 We apply the same index for evaluations of party leaders. Thus, LAP signifies the average weighted divergence in the evaluations of in-party leader and out-party leaders. The formula of the index is the following:
$$ \mathrm{API}=\sum \limits_{n=1}^N\left[\sum \limits_{\begin{array}{c}m=1\\ {}m\ne n\end{array}}^N\left(\left({\mathrm{Like}}_n-{\mathrm{Like}}_m\right)\times \left(\frac{{\mathrm{Vote}\ \mathrm{share}}_m}{1\hskip0.35em -\hskip0.35em {\mathrm{Vote}\ \mathrm{share}}_n}\right)\right)\times {\mathrm{Vote}\ \mathrm{share}}_n\right]. $$
In this equation, “Like” signifies the like–dislike evaluation (0 to 10) toward the party/leader. “n” denotes the in-party/in-party leader and “m” refers to the out-party/out-party leader.
Like–dislike evaluations were asked for the relevant parties and their respective leaders/presidential candidates in each country, with the N of parties/leaders ranging from 2 to 9. In some cases, the number of evaluated parties and leaders did not match perfectly. For example in Switzerland, the like–dislike question was asked about six parties, but only four leaders. In such cases, we adjusted the calculation to make the parties/leaders match; for example, in the Swiss case, party API was also calculated based on four parties.
Finally, we have made an adjustment regarding how to determine the “in-party.” While most studies on affective polarization determine in-party based on partisan identification (Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020; Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019; Reiljan Reference Reiljan2020), we assign it based on vote choice. Considering that partisanship has decreased in many countries and is at very low levels in some parts of the world (e.g., Central Eastern Europe and Southeast Asia), defining partisan groups based on vote choice could give us a better comparative perspective. Also, it should be a more conservative measure of affective polarization, as voting for a party is a somewhat more lenient indicator of partisan preference than declaring partisan identity.Footnote 12
LAP/PAP Ratio
As explained in the theoretical section, our main variable of interest is the relative intensity of LAP compared to PAP. To calculate this variable, we simply divide the Leader API with the Party API. Thus, ratio values above 1 indicate that LAP is higher than PAP, whereas values below 1 signify the opposite.
Independent Variables
Partisanship (PID)
To measure PID, we make use of the CSES survey item that asks the respondents whether they are close to one political party, as compared to others. The questionnaire also includes a follow-up question to identify partisan “leaners” (Petrocik Reference Petrocik2009; Reiljan Reference Reiljan2020). We count as partisans only those who answered affirmatively already to the first PID question, thus excluding the leaners. This provides us with a more stringent measure of partisanship and avoids dropping a number of cases where the follow-up question was not asked in the survey. The value used in the models is the percentage of respondents who reported that they feel close to a party, rescaled to run from 0 to 1.
Ideological Polarization
We measure ideological left–right polarization with the party polarization index developed by Dalton (Reference Dalton2008). Dalton’s index is based on the public perceptions of left–right placements of the political parties. In the CSES questionnaire, respondents are asked to place the parties on the left–right continuum ranging from 0 (left) to 10 (right). To calculate party polarization, Dalton combines the average left–right scores assigned to each party with the vote shares of these parties from preceding elections. The index ranges from 0 (no polarization) to 10.
Government Performance
To capture regime political output, we use the Government Effectiveness indicator from the World Bank’s Worldwide Governance Indicators (WGI) dataset (The Worldwide Governance Indicators Project 2022). The Government Effectiveness indicator “…reflects perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies.” This index is a very broad measure for an entire government’s performance (Dahlberg and Holmberg Reference Dahlberg and Holmberg2012) and is thus well suited for the broad, cross-national nature of this study. The index originally ranges from −2.5 (weakest performance) to 2.5. To ease the interpretation, we have recorded it to values ranging from 0 to 5.
Party System Fragmentation
We use the classic effective number of electoral parties index developed by Laakso and Taagepera (Reference Laakso and Taagepera1979), which provides for the adjusted (by relative strength) number of parties in the party system. The formula of the index is:
$$ \mathrm{ENP}=\frac{1}{{\sum \limits}_{i=1}^n{v}_i^2}, $$
where “n” denotes the number of parties and “v” is the normalized vote share of the party.
Presidentialism
A dummy variable takes the value of 1 in case of presidential systems or semi-presidential systems where the president’s office is clearly the most dominant. We used insights from Curini and Hino (Reference Curini and Hino2012) and Shugart (Reference Shugart2005; Reference Shugart, Rhodes, Binder and Rockman2006) to identify presidential systems. As semi-presidentialism is a rather vague concept that could include very different regimes, such as Finland and Peru (Elgie Reference Elgie2007), we were strict in our criteria for selecting presidential/strong semi-presidential systems to minimize the number of questionable cases. The cases that are classified as presidential in our current sample are France (2007), Mexico (2012), Peru (2011, 2016), Taiwan (1996), Turkey (2018),Footnote 13 USA (2008, 2012, 2016), and Uruguay (2009).
Finally, we control for the year of election in the models to account for the possibility of change over time. The descriptive statistics of all the main variables are presented in Table A.2 in the Supplementary Material and bivariate correlations between variables are presented in Table A.3 in the Supplementary Material.
RESULTS
Leader Affective Polarization across the Democratic World
In this section, we present the descriptive overview of the levels of LAP and its relationship with PAP across the 40 countries in our sample. Figure 2 ranks the countries, based on their average degree of LAP. The country that exhibits the highest level of LAP is Turkey, with most Central Eastern European (CEE) countries also ranking very high. At the bottom of the list, we find the countries that are known for consensual political culture and broad-based governing coalitions: the Netherlands, Switzerland, and the Nordic countries. Also, the only Southeast Asian country in the sample—Taiwan—is among the least polarized nations in terms of feelings toward party leaders. Overall, the list that is displayed in Figure 2 looks similar to the rankings of PAP (Lauka, McCoy, and Firat Reference Lauka, McCoy and Firat2018; Reiljan Reference Reiljan2020; Wagner Reference Wagner2021), although there is also a notable difference: the United States ranks higher with regard to LAP, with its average value being well above the sample average.
Figure 2. List of Countries by Leader Affective Polarization Scores
Note: The scores indicate country averages over all the elections that are in the sample.
As expected, there is a strong overlap between LAP and PAP. Figure 3 shows that most cases remain close to the regression diagonal and the R 2 value is 0.85. There is clearly a high degree of endogeneity between party and leader ratings, which is not surprising, especially considering that in the CSES questionnaire, respondents are asked to evaluate party leaders right after they have answered the party like–dislike question. With such data, it is not possible to determine whether feelings toward parties are rather driving the leader evaluations, or vice versa, and we will not probe into this question any further in this article, as it is not the purpose of our study.
Figure 3. Leader Affective Polarization as a Function of Party Affective Polarization
Yet, the two constructs are not identical. Despite the very high correlation—which is probably inflated at the system level of analysis as smaller individual or party level differences flatten out—there is a systematic difference between PAP and LAP: on average, the latter is significantly weaker. Within our sample, the mean LAP value is more than 0.5 point lower compared to the respective PAP measure (LAP: 3.75; PAP: 4.28). Thus, on average, LAP amounts to only 87% of PAP (i.e., the LAP/PAP ratio is 0.87). The difference emanates from both sides of the affective polarization equation: voters rate their preferred party better than the leader of that party, while being slightly less hostile toward the out-party leaders as compared to the parties themselves, although the difference is larger regarding in-party/in-leader evaluations. These divergences are illustrated in Figure 4.
Figure 4. Average Weighted In-Party/In-Party Leader and Out-Party/Out-Party Leader Evaluations with 95% Confidence Intervals
Moreover, we find a significant cross-national variance in the LAP/PAP ratios. As displayed in Figure 5, there is an almost two-fold difference between the cases with the highest and lowest ratios, with the country average values ranging from 0.65 to above 1.1. However, there are only three countries (USA, Mexico, and Peru), where the average ratio is above 1. In other words: levels of LAP are lower than those of PAP in more than 90% of the countries in our sample.Footnote 14 Figure 5 also reveals significant regional disparities in the LAP/PAP ratio. All three Latin American countries in the sample rank high in terms of the ratio, with the average value being below 1 only in Uruguay. Average values among CEE, Southern European, and Oceanian, countries remain around 0.9. In Northwestern European countries, however, affective polarization clearly manifests itself predominantly in feelings toward parties, as on average, LAP amounts for only 80% of PAP.
Figure 5. List of Countries by Leader/Party Affective Polarization Ratio
While previous studies of PAP have consistently demonstrated that the degrees of affective polarization are not particularly high in the United States (in comparative terms), Figure 5 indicates that the USA is definitely an exceptional case regarding the LAP/PAP ratio. It is the only country where the ratio is significantly above 1, showing the central role of presidential candidates for affective polarization in the United States. Compared to 2008, there has been a swift upward trend regarding the LAP/PAP ratio in the United States, with the 2016 election being the case with the highest value (1.13) among the 102 elections in our sample. This is hardly a surprise, considering the extremely polarizing personality of the Republican nominee Donald Trump, although it should be noted that also Republicans were significantly more hostile toward Hillary Clinton than toward the Democratic party itself.
We do not detect any consistent time trend in our sample. Out of the 29 countries for which we have more than one data point, in 16 the LAP/PAP ratio has increased, while in 13 it has declined (see Table A.1 in the Supplementary Material). However, in most cases, the temporal variation is marginal. There are few exceptions, such as the United States and Greece, where the ratio has increased by 0.2 points. On the other hand, we also see a significant decline in some countries, for example, compared to the late 90s, the ratio has dropped by more than 0.1 points in Australia and Switzerland. Overall, our results demonstrate the continuing importance of parties vis-à-vis leaders as polarizing agents. In the vast majority of the countries in our sample, voters’ feelings are clearly more polarized over parties and this relationship has not changed significantly over the last two decades.
Predictors of LAP/PAP
The descriptive results demonstrate that regardless of the very high correlation between LAP and PAP, there are meaningful divergences between the two constructs, as indicated by the significant cross-national variation we find in the LAP/PAP ratio. We believe that this variation merits further empirical attention. Therefore, in this section, we probe into the predictors of LAP, PAP, and most importantly, the ratio between the two. Table 2 displays three models with an identical set of independent variables, but a different dependent variable (PAP, LAP, LAP/PAP ratio). The first column in the table confirms the findings of previous studies on PAP, as all the independent variables have a significant effect in the predicted direction. The second column shows that, indeed, the predictors of LAP are very similar to those regarding PAP, which had to be expected due to the very strong correlation between the two constructs. Also, the general fit of both models is almost equal, with R 2 values being at 0.64 for the PAP, and 0.63 for the LAP model. Nevertheless, the first two columns also reveal some significant differences in the effects of certain variables on the levels of PAP and LAP. These differences are reflected in the third column of Table 2, displaying several significant effects on the LAP/PAP ratio, although this model’s fit is much lower compared to other two.
Table 2. The Predictors of Party Affective Polarization (PAP), Leader Affective Polarization (LAP), and LAP/PAP Ratio (OLS Regression)
Note: Unstandardized regression coefficients, cluster-corrected robust standard errors in parentheses (Cluster = Country). ***p < 0.01, **p < 0.05, *p < 0.1.
To begin discussing the results with our first set of hypotheses pertaining to regime input variables, we find that both PID and ideological (left–right) polarization—in contrary to our hypotheses 1 and 2—have very similar effects on LAP and PAP. As predicted, and in line with previous research, countries where larger parts of the population identify with parties, and where those parties are ideologically more dispersed, the levels of PAP are significantly higher. Our findings demonstrate that exactly the same applies to LAP. These two independent variables have been at the center of the affective polarization debate in the United States, with some scholars arguing for the overarching importance of partisan identity in affective polarization and another group putting more emphasis on the ideological/policy differences (see Lelkes Reference Lelkes2021 and Orr and Huber Reference Orr and Huber2020 regarding this debate). Our results indicate that both have an independent impact on affective polarization, at least at the system level.Footnote 15 Regarding our central variable of interest—the LAP/PAP ratio—neither PID nor ideological polarization display any significant effects. Although the effect of ideological polarization is slightly stronger on PAP than LAP, the difference is too small to have any notable effect on the ratio. Thus, these two very crucial variables in the PAP literature do not have much value for the aims of this study.
Regarding regime output, the results displayed in Table 2 reveal a significant divergence in the effects of government effectiveness on PAP and LAP. As hypothesized (H3), affective polarization is generally much lower in countries with a well-functioning impartial government sector. Yet, and as predicted, this relationship is even stronger in the case of LAP, which results in a significant effect on the LAP/PAP ratio. The third column in Table 2 demonstrates that government effectiveness is the strongest variable in the model, as a one-step increase on the 0-to-5 scale corresponds with a 0.05-point increase in LAP/PAP ratio. This effect is significant both statistically and in substantive terms. A two-step increase in government effectiveness (e.g., the exact difference between Denmark and Turkey regarding this variable) brings about a slightly more than one standard deviation drop in the LAP/PAP ratio, all other variables being controlled for. The problem with the government effectiveness variable is that it correlates strongly with several other crucial indicators such as corruption and overall wealth of a country. Our broad cross-national focus does not allow us to go into much detail, but these results clearly suggest that generally more successful countries offer less options for leaders to affectively polarize more strongly than the parties.
Finally, we find significant support for our hypotheses regarding the importance of the institutional set-up on the relative strength of LAP and PAP. First, the effective number of parties has a significantly stronger taming effect on LAP than on PAP. Consequently, fragmentation relates to a lower LAP/PAP ratio. As we see from the third column in Table 2, one additional (relevant) party in the system corresponds to an almost 0.02 point increase in the ratio, controlling for other relevant variables. For example, moving from two relevant parties (e.g., in the United States) to seven (e.g., in the Netherlands) makes up an almost 10% difference in the LAP/PAP ratio. This finding aligns with our expectation (H4) that the polarizing appeal of specific personalities somewhat dissolves in highly fragmented party systems where elections are usually first followed by government coalition negotiations, not an unmediated and inevitable inauguration of the winning party’s leading candidate.
The most substantial difference between the first and the second columns in Table 2 concerns the presidential/parliamentary system variable. Although presidential systems display systematically lower levels of affective polarization, this effect is almost two times weaker (and insignificant) in our LAP model. Thus, the LAP/PAP ratio is 0.09 points higher in presidential systems, which corresponds to a highly significant effect and offers strong support for Hypothesis 5. Although in parliamentary systems, party leaders are also usually presented as the party’s leading candidates to take the prime minister’s office, the direct nature of presidential elections and the importance of that position clearly push the LAP/PAP balance in the direction of leaders.Footnote 16
Controlling for the year of election does not affect the main findings, nor does this variable have any effect on the LAP/PAP ratio. However, the first two columns in Table 2 demonstrate that there is an upward trend in the levels of both PAP and LAP, even when controlling for all other relevant variables (cf. Boxell, Gentzkow, and Shapiro Reference Boxell, Gentzkow and Shapiro2022; Garzia, Ferreira da Silva, and Maye Reference Garzia, da Silva and Maye2023; Gidron, Adams, and Horne Reference Gidron, Adams and Horne2020, who do not find such an increase in affective polarization levels, at least within the Western democracies). The question of whether and to what extent has affective polarization generally increased across the democratic world constitutes an intriguing area for future research.
Overall, our results indicate the importance of regime output and institutions in shaping the balance between LAP and PAP. In presidential systems, the LAP/PAP ratio is always very close to or even above 1, indicating the polarizing allure of presidential candidates in such regimes. However, while we do not find presidential elections where the ratio is very low, we do see parliamentary systems where the ratio is quite high, for example, Montenegro and Slovakia. Another very crucial variable—government effectiveness—could be helpful in explaining these cases, as the nonpresidential regimes that exhibit high LAP/PAP ratios tend to have rather badly performing governments. In the only nonpresidential high-performing country where LAP exceeds PAP on one occasion (New Zealand, 2008), a swift increase in the ratio coincides with a significant drop in the effective number of parties (from 4.3 to 3.1). Thus, we see that it could be a combination of different variables with independent effects that determines the relative strength of LAP to PAP.
Nevertheless, our model leaves more than half of the variation in the LAP/PAP ratio unexplained, as the R 2 remains at 0.41. Although we have tried—at least with some success—to offer structural explanations to the relative strength of LAP vis-à-vis PAP, sometimes it may simply come down to specific personalities. For example, in the previously mentioned cases of Slovakia and Montenegro, the respective political campaigns have very strongly revolved around the former long-serving prime ministers (Robert Fico in Slovakia, Milo Đukanović in Montenegro) who have developed an intense polarizing appeal. It could be that the LAP/PAP ratio takes a sudden downturn if such leaders retire. One other potential factor that we could not account for in this study is a post-electoral disappointment of some losing parties that could manifest in negative feelings especially toward the (in-party) leader. For example, the low LAP/PAP ratio in Great Britain’s 2015 elections (0.76) is partially accountable to the very poor evaluation that Labor voters assigned to their in-party leader Ed Miliband after a disappointing loss in the election—he received an average like–dislike score of 5.36, while the Labor party itself was evaluated with a score equal to 7.11 by its voters. This suggests that in-party evaluations might be more resilient toward electoral failures, whereas leaders get most of the blame. Such questions will remain to be addressed in future studies.
To address the previously mentioned issue of autocorrelation within countries, we also ran the same models as in Table 2 with country average values and with a restricted sample of two latest elections from 29 countries (see Tables A.5 and A.6 in the Supplementary Material). These replications largely confirm our main findings, although the effect of party system fragmentation misses the statistical significance threshold in the country average model. Moreover, we controlled for the possibility that the effects of our parliamentarism–presidentialism and government effectiveness measures could actually reflect different democratic/authoritarian legacies. The majority of presidential (and rather low-performing) systems in our sample are found in South America and Southeast Asia, and do not have long-standing democratic traditions. The results shown in Table A.7 in the Supplementary Material, however, do not confirm that the duration of consolidated democratic governance has a significant effect if added to our model, although it does slightly weaken the impact of government effectiveness. Finally, Table A.8 in the Supplementary Material replicates our results using the PID item instead of vote choice and shows a smaller (but still significant) effect for presidentialism. Our broad system-level approach does not allow us to exhaustively test all possible confounding variables that could drive LAP, PAP, and the LAP/PAP ratio, but these alternative specifications give us additional confidence in the robustness of our findings.
CONCLUSION
This study of affective polarization geared at leaders and parties has uncovered a number of system-level findings that relate to different strands of comparative politics literature. First, our results reveal the continuing centrality of political parties in voters’ political perceptions. Although feelings toward leaders and parties are strongly correlated with each other, polarization is significantly more intense regarding the latter. In 37 out of 40 countries that we study, the degree of PAP exceeds the level of leader affective polarization. People tend to like their own party more than the leader of the party, while being slightly more positive about the leaders of competing out-parties as compared to the out-parties themselves. While this does not necessarily refute the personalization of politics thesis, it does underline that the relevance of parties has not disappeared. Even if levels of partisan identification and party membership have dropped, parties still invoke stronger feelings among voters than leaders. This suggests that even in the context of personalized politics, there is something more about the party labels that generates affect than just the top candidate of the parties. Practical implications of this finding may crystallize in the realm of partisan campaign strategies, with parties not too rapidly putting all their money on the image of a leader. The same goes for models of party organization, crucial for leadership selection mechanisms (Poguntke, Scarrow, and Webb Reference Poguntke, Scarrow and Webb2017). Future studies may fruitfully link affective polarization toward both parties and leaders with trends in electoral campaigning and party organizational strategies.
At the same time, we show that the ratio between LAP and PAP varies substantially across countries. In some European cases, LAP makes up for less than two-thirds of PAP, while in the United States, feelings toward leaders are over 10% more polarized than party evaluations. Our multivariate analysis reveals that institutions are crucial in explaining this variance. In presidential systems, feelings toward parties and their top candidates are on par, and sometimes polarization over the latter is even higher. Also, a higher number of relevant parties dissolves the polarizing appeal of the leaders, as indicated by our finding that more fragmented party systems display lower LAP/PAP ratios. Thus, it is not surprising that the presidential two-party system of the United States is, by a large margin, at the top of the ranking, while the Swiss parliamentary consensus democracy with more than six relevant parties is at the bottom of the list. These findings indicate the importance of the classic concepts/distinctions introduced and elaborated by the likes of Lijphart, Taagepera and Laakso, and Linz in understanding the balance between LAP and PAP in contemporary democracies.
Our results also offer new insights regarding the relationship between affective polarization and factors relating to regime input/output. While the regime input variables of partisan identification and ideological polarization have the expected strong positive correlations with both LAP and PAP individually, we do not find any effect of these variables on the ratio between the two. Regime political output, measured by government effectiveness, however, affects the two constructs asymmetrically. While, as expected, countries with well-functioning governments are affectively less polarized, this effect is significantly stronger on LAP. Therefore, the LAP/PAP ratio is more biased toward leaders in cases of poor government performance. Our interpretation of this finding is that an ineffective public sector gives more chances for specific (charismatic) individuals to promise profound, overall beneficial changes and polarize the system around their own personality. Obviously, this mechanism is subject to further theoretical elaborations and empirical testing that goes way beyond the scope of this article. However, our results allow us to argue that if we want to explain the patterns of affective polarization and, especially, the relationship between LAP and PAP, then in addition to institutions, also regime performance matters.
Finally, our results call for further research on the relationship between LAP and PAP as well as its consequences in the political and social realm. Regarding the relationship between the two manifestations of affective polarization, we find that their ratio correlates with both LAP (r = 0.67) and PAP (r = 0.34) (see Table A.3 in the Supplementary Material). Thus, we can say that if the relative strength of LAP vis-à-vis PAP is higher, then affective polarization in general tends to be more intense. This is just a preliminary system-level observation, but it could be a point of departure for more detailed studies. Could it be that focusing less on leaders/top candidates in political campaigns is a potential way to ameliorate affective polarization and its pernicious consequences? Or is it more important to look for common grounds on policy issues, as ideological polarization is also strongly related to the levels of affective polarization? Answering these questions could potentially help taming intense political hostility and its adverse consequences.
Our findings open up further potential avenues for research adopting more specific research designs, building on our broader analytical framework. First, our empirical research is limited to a particular conceptualization of “affective polarization” as used in the comparative politics literature. In the United States, recent contributions rely on alternative and more refined measurements to study affective polarization, potentially leading to a growing divide between American politics and comparative politics, and in this case the different meaning the subfields are giving to the same concept of “affective polarization.” In this contribution, we have engaged with this distinction theoretically. Empirically, novel comparative data would be needed to ease this emerging divide. Second, and as we have noted, our data mainly show cross-country variations. However, future research might delve more deeply into within-country (or even within-region) comparisons across time. Third, and while our contribution is not framed to address this issue, we believe that future individual-level analyses are warranted to better understand the psychological underpinnings behind potentially differing mechanisms that lead to LAP and PAP in the first place. Fourth, and relatedly, different levels of aggregation of our key concepts may prompt answers to different types of research questions. Our article cannot make claims about the impact of different levels of LAP or PAP—nor of their relative strength—on outcomes such as out-party discrimination in social, cultural, or economic areas, nor regarding political behavior by voters and parties. To do so, would require an experimental research design that the broad, cross-national focus we adopted could not encompass.
In sum, we see that while we have shed new light on this nascent topic, important questions remain open. We hope that they can be addressed more fully in the future by, amongst others, research that will also zoom in more closely—and qualitatively—on individual cases.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055423000485.
Data Availability Statement
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/M06IRS.
Acknowledgments
We thank the participants of The European Consortium for Political Research (ECPR) 2021 Joint Sessions online panel on “Affective Polarization Around the World: Measurement, Causes and Consequences” for their feedback on an earlier version of this article. We also wish to thank the Editors and the four anonymous reviewers for their most valuable feedback.
Funding Statement
This manuscript benefited from the funding by The University of Lucerne research committee (FoKo), project “Patterns of Affective Polarization in the Democratic World: Comparing the Polarized Feelings Towards Parties and Leaders” (project number: 51369), the Swiss National Science Foundation (PCEFP1_186898), and Estonian Research Council mobility grant no. MOBJD607.
Conflict of Interests
The authors declare no ethical issues or conflicts of interest in this research.
Ethical Standards
The authors affirm this research did not involve human subjects.
Open access
