This is the accepted version of the journal article: Hernández, Enrique; Anduiza Perea, Eva; Rico, Guillem. «Affective polarization and the salience of elections». Electoral Studies, Vol. 69, Núm. 102203 (February 2021). DOI 10.1016/j.electstud.2020.102203 This version is available at https://ddd.uab.cat/record/238023 under the terms of the license Affective polarization and the salience of elections Enrique Hernández, Eva Anduiza and Guillem Rico Department of Political Science Universitat Autònoma de Barcelona Published at Electoral Studies DOI: 10.1016/j.electstud.2020.102203 Abstract In this paper we analyze the effects of election salience on affective polarization. Campaigns and elections epitomize the moment of maximum political conflict, information spread, mobilization, and activation of political identities and predispositions. We therefore expect that affective polarization will be higher just after an election has taken place. By the same token, as elections lose salience, affective polarization will diminish. We analyze this question using CSES data from 99 post-electoral surveys conducted in 42 countries between 1996 and 2016. Our identification strategy exploits variation in the timing of survey interviews with respect to the election day as an exogenous measure of election salience. The empirical findings indicate that as elections lose salience affective polarization declines. The paper further contributes to the debate on the origins of affective polarization by exploring two mechanisms that may account for this relationship: changes in ideological polarization and in the intensity of party identification. Both are relevant mediators, with ideological polarization seemingly playing a more important role. Keywords: Affective polarization; Elections; Ideological polarization; Partisanship; Salience; Attitudes; Acknowledgments: This research has received financial support from the Spanish Ministry of Science, Innovation, and Universities through research grant CSO2017-83086-R. We are grateful to the editor and anonymous reviewers of Electoral Studies and participants at the 2019 APSA conference for their helpful comments and suggestions. Corresponding author: Enrique Hernández, Department of Political Science, Universitat Autònoma de Barcelona, Edifici B, Campus de Bellaterra, Bellaterra 08193, Spain. Email: enrique.hernandez@uab.cat Introduction Affective polarization has become one of the main concerns for the health and quality of contemporary democracies. Feelings of dislike towards political parties other than the preferred one (i.e. affective polarization) appear to be acutely present in most countries (Reiljan, 2019). Attempts to explain apparently unprecedented increases in inter-party hostility in some countries, such as the US, consider this a consequence of complex social processes such as the expansion of social media (Barberá, 2015), increasing media incivility (Mutz, 2015), negative campaigning (Iyengar, Sood, & Lelkes, 2012) or growing inequality (Gidron, Adams, & Horne, 2019). In this paper, we take a different stand and explore the short-term temporal dynamics that drive affective polarization. We argue that affective polarization is partially the consequence of electoral dynamics that are inherent to democracy. Because elections heighten and increase the visibility of political conflict and activate partisan identities (Michelitch & Utych, 2018), we expect that election salience will lead to higher levels of affective polarization. We analyze this question drawing on data from 99 post-electoral surveys conducted in 42 countries between 1996 and 2016 through the Comparative Study of Electoral Systems (CSES). Our identification strategy exploits variation in the timing of survey interviews with respect to the election as an exogenous measure of election salience (for a similar approach see Singh & Thornton, 2019). The empirical findings indicate that as elections lose salience, affective polarization declines. After an election has taken place the degree of interparty animosity progressively decreases, at least up to nine months after citizens casted their votes. Our analysis also explores the role of two mechanisms that have been related to increasing levels of affective polarization: stronger partisan identities and ideological polarization. We find that the relationship between election salience and 2 affective polarization is mediated by both mechanisms, but changes in ideological polarization seem to slightly more relevant than changes in party identification in order to explain the effects of election salience on affective polarization. These findings provide valuable comparative evidence about the mechanisms that drive changes in affective polarization in advanced democracies, at least in the short-term. Therefore, our analysis contributes to the understanding of the nature of affective polarization, which is, in part, a byproduct of the electoral dynamics that are inherent to democracies. These dynamics activate strong partisan identities and make political options be perceived as ideologically differentiated, which ultimately lead to higher levels of polarization. Importantly, at the same time, our findings indicate that levels of affective polarization can also be significantly reduced in short span of time after an election has taken place. These results, showing how individuals depolarize after elections, are aligned with recent findings about how citizens become more polarized during election campaigns (Hansen & Kosiara-Pedersen 2017). However, it is still an open question whether the process of post-electoral depolarization that we uncover in this paper allows to compensate for the increases in polarization that we might observe during election campaigns. Theory Election salience and affective polarization Increasing affective polarization is considered to be one of the main challenges faced by contemporary democracies (Dryzek et al., 2019; Gidron et al., 2019; McCoy & Somer, 2019). Citizens seem to be increasingly prone to profoundly dislike the parties they do not identify with, which puts a strain in the pluralistic values of respect for diversity that democracies are expected to protect and promote. 3 The question has been researched extensively in the US, where the increasing trend of affective polarization is well documented (Iyengar et al., 2012; Lelkes, 2016; Westwood, Iyengar, Malhotra, Lelkes, & Levendusky, 2018). In a broader comparative perspective, though, it is harder to find evidence of such an increasing trend (but see Boxell et al. 2020 and Reiljan, 2019). Efforts outside the US concentrate on the conceptualization and measurement of polarization in multiparty systems (Lauka, McCoy, & Firat, 2018; Reiljan, 2019; Rodden, 2019; Wagner, 2019), and on the analysis of case studies (see the special issue of American Behavioral Scientist, 2018, Vol. 62, No. 1). Only a few works have analyzed and attempted to explain affective polarization adopting a broad comparative perspective (Gidron et al., 2019; Reiljan, 2019), and in those cases the cross-national dimension seems to be privileged over the longitudinal one. The evidence available is still limited to provide a compelling account of whether today we live in societies with higher levels of affective polarization than ever before. However, beyond a potential monotonic increase in polarization, there are other facets of the temporal dynamics of affective polarization that should be considered. Whether or not affective polarization varies along time in response to specific contextual situations gives us a sense of the extent to which it may respond to structural factors and hence become itself a stable characteristic of some party systems, or rather, whether it fluctuates in response to specific short-term political situations and events. With this dynamic perspective in mind, affective polarization may be expected to increase at different moments in time. Specifically, we expect affective polarization to be higher at those times when political conflict is particularly salient, the richness of the information environment peaks, and political predispositions are being activated by political entrepreneurs. A context that is clearly characterized by these features are election 4 campaigns. In addition, the use of negative campaign strategies, which should reinforce political conflict during campaigns, seems to be expanding (Nai, 2018). Social media and online content also exacerbate party identities by increasing the emotional load of political messages and partisan appeals on days close to election day (Aragõn, Kappler, Kaltenbrunner, Laniado, & Volkovich, 2013), and by reinforcing online interactions among co-partisans (Gruzd, 2014). After such intense political moments (i.e. when the election campaign has finished), affective polarization is expected to decline. As elections lose salience after the campaign, individuals’ animosity towards other parties should decline. In most democracies, elections epitomize the moment of maximum political conflict, information spread, mobilization, and activation of political identities and predispositions. Previous studies have shown how elections and campaigns lead to higher levels of partisanship (Grant, Mockabee, & Monson, 2010; Michelitch & Utych, 2018; Singh & Thornton, 2019) and political information (Ferrín, Fraile, & García-Albacete, 2019; Freedman, Franz, & Goldstein, 2004). However, there remains a paucity of evidence on the effect of the salience of elections on affective polarization. Although recent research is indicative in this direction (see Hansen & Kosiara-Pedersen 2017 study showing an increase in Danes polarization before elections), it is not yet clear whether and to what extent elections promote affective polarization. This is an important question because, while party attachments or political information are considered as indicators of a healthy democratic political system, affective polarization tends to be considered as highly problematic. The key question is whether we can have the desirable consequences of elections (citizens with well-informed and clear political preferences) without the less desirable ones (citizens that have both strong positive in-group affects and strong negative out-group 5 affects). We propose that affective polarization is a predictable outcome of election salience, as campaigns and elections provide the necessary ingredients for citizens to develop not only informed policy preferences, but also greater political animosity towards parties and candidates of the opposite camp. We hence expect to find a positive effect of election salience on the levels of affective polarization (H1). The connection between election salience and affective polarization confronts us with a paradoxical situation. The very institution that lies at the core of liberal democracy (competitive free elections) becomes also a source of a strain for democratic politics by promoting affective polarization. This is why it is important to understand which are the mechanisms that might drive the expected increase in affective polarization as a result of elections and election salience. Mechanisms: partisan identities and ideological polarization By focusing on the consequences of the salience of elections on affective polarization we are able to elaborate on the micro-level mechanisms that may be at work and mediate the effects of election salience on affective polarization. The burgeoning literature on affective polarization offers two main explanations for the apparent rise in inter-party animosity (see Lelkes, 2019; Westwood et al., 2018). While some argue that affective polarization can be linked to the increasing relevance of partisanship as a social identity (Iyengar et al., 2012), others contend that polarization is the consequence of rising political extremism and ideological divides between parties and elites (Rogowski & Sutherland, 2016). Affective polarization may, first, be the result of stronger partisan identities, which are, in turn, expected to strengthen as a result of election salience (Michelitch & Utych, 6 2018; Singh & Thornton, 2019). This theory of affective polarization considers partisanship as an essential social identity that should drive changes in affective polarization. This explanation for affective polarization is related to the social identity theory of partisanship (Huddy, Mason, & Aarøe, 2015), that considers partisanship as a long standing expressive identity rather than as an attitude based on the performance or issue positions of parties. Iyengar et al. (2012) shifted the focus from policy or ideological polarization to affective polarization precisely emphasizing the importance of partisanship as a social identity that requires both the definition of an in-group and an out-group. The process of social and partisan sorting that has taken place in the US has paved the way to the definition of clear in-groups/out-groups that eventually generate higher levels of inter-party animosity (Mason, 2018). Increasing social homogeneity among the voters of parties, and a greater alignment of partisan and social identities, lead to more emotionally reactive and affectively polarized voters (Mason, 2016). The US case, therefore, suggests that the key explanatory element behind increasing affective polarization is related to an increasing relevance of party identification considered as a social identity (Abramowitz & Webster, 2016, 2018; Abramowitz & McCoy, 2019; Klar, 2018). This process of partisan sorting is quite peculiar to the US, but we do not know if the same social identity mechanism can be generalized to other countries that, rather than sorting, are currently experiencing processes of partisan dealignment, realignment or detachment (Kitschelt & Rehm, 2015). In any case, in as far as the social identity theory is applicable from a comparative perspective (Bankert, Huddy, & Rosema, 2017), we would expect that partisan identities would be activated as elections become salient. In fact, Michelitch & Utych (2018) comparative study has shown how partisanship fluctuates during the election cycle, with citizens attachment to their preferred party being stronger 7 just before and after elections (see also Singh & Thornton, 2019). Therefore, party identification is expected to mediate the relationship between election salience and affective polarization (H2). That is, election salience might exacerbate affective polarization by strengthening the attachment individuals’ feel towards their in-group (their preferred party). An alternative, but not necessarily incompatible, explanation of the origins of affective polarization locates the focus of attention on ideological differences. From this perspective, polarization would arise from the perception of increasing ideological differences between political parties. In the US, the extent to which ideological mass polarization has increased remains a matter of controversy (Fiorina & Abrams, 2008). From a comparative perspective most studies have focused on analyzing cross-country variation (Dalton, 2006), and findings about the longitudinal trends in ideological polarization are mixed (Dalton, 2008; Schmitt & Freire, 2012). But even if a clear upward polarization trend has not been identified, survey experiments have found that ideological polarization partially accounts for increasing levels of affective polarization in the US (Rogowski & Sutherland, 2016). In fact, ideological differences between candidates seem to be more important for affective polarization than partisanship (Lelkes, 2019). Beyond the US case, the emergence and success of new (and populist) parties with ideologically extreme positions (Kriesi & Pappas, 2015) would lead us to think that the ideological polarization brought into the system by these parties can be an important trigger of affective polarization (Mudde & Kaltwaser, 2018: 16-17). We argue that perceptions about the ideological differences between parties become more visible as the salience of elections increases. During election campaigns citizens are more likely to learn about the “brand” that parties represent and where those stand on 8 different issues (Arceneaux, 2006; Lupu 2013: 8). These heightened perceptions about the ideological differences between parties might, therefore, account for increasing levels of affective polarization during election time. Under an assumption of vote maximization, parties often have an incentive to emphasize their ideological identities and programmatic differences in order to attract voters during campaigns. Moreover, even if parties do not directly campaign on their policy discrepancies, for most citizens these should become more apparent than ever during the pre-election and campaign periods (Arceneaux, 2006). We therefore expect that ideological polarization will mediate the relationship between election salience and affective polarization (H3). By increasing ideological polarization, election salience is expected to increase affective polarization.1 Data and identification strategy To test these hypotheses, we draw on data from the CSES Integrated Module Dataset (IMD), complemented with data from CSES modules 1-4. The IMD includes data from 174 post-electoral surveys fielded in 59 countries. In our analyses, we rely on data for 99 of these post-electoral surveys. We exclude those post-election studies conducted in non- democratic countries,2 those for which there is no data for at least one of the variables considered in our analyses, and those that present no variation in the timing of survey interviews.3 The resulting dataset includes data for parliamentary/legislative and 1 Note that, although our measure of ideological polarization more closely captures the symbolic aspect of the left-right dimension, the argument applies to substantive ideology as well as to ideology as an identity marker (Malka & Lelkes 2010; Vegetti & Širinić 2019). 2 Countries with a value lower than 6 in the Polity IV index at the time (year) of the election. 3 In 11 election studies there is no variation in our key independent variable, i.e. the timing of interviews after election day. Apparently, all survey interviews were conducted during one single day in these studies. 9 presidential elections held between 1996 and 2016 in 42 countries.4 Figure A1 in the Appendix provides further information about the elections and countries considered in this paper. To measure affective polarization at the individual-level we rely on the measures developed by Wagner (2019), which are based on the 0-10 like-dislike scores each respondents assigns to each of the parties considered in the CSES. This indicator has the advantage of being widely available for a large number of elections and of being a suitable measure of this type of polarization in multiparty contexts. While this indicator takes parties as objects, and hence does not distinguish whether this like-dislike attitudes relate to party elites or voters, Druckman and Levendusky (2019) find that affective polarization seems to reflect mostly perceptions of party elites. Wagner proposes two different approaches to operationalize affective polarization: a measure based on the spread of like-dislike scores for all the parties (spread measure), and a measure based on the distance from the most liked party to all other parties in the party system (distance measure). In this paper, we privilege the former. The spread measure is more applicable to multiparty systems, which are more common in our sample (Wagner, 2019). In any case, we also replicate all of our analyses using the alternative distance measure (see Appendix C for further information about the operationalization of the distance measure and the results obtained through this alternative operationalization). 4 The average Polity IV score of the countries included in the sample is 9.38. In 81.82 percent of the elections analyzed voting was not compulsory, in 9.09 percent it was compulsory but it was not sanctioned, and in 9.09 percent it was compulsory and enforced. The electoral formula in these elections was: plurality in 13 percent, majority in 66 percent, and proportional in 21 percent. 10 The spread measure is operationalized as the weighted average party like-dislike difference relative to each respondent’s average party like-dislike score. Following Wagner’s (2019) notation, the equation for parties ! and voter " is as follows: % $%&'()! = +∑&#'( -# ./"01!# − 444444 /301$ 5 (1) where /"01!# is the like-dislike score assigned to a given party !, and -# is the vote share for each party (normalized to range between 0 and 1).5 444444$ is also weighted so that it reflects party size. The /301 The mean of affect /301 444444$ parameter is therefore calculated as follows: & 444444$ = 9 /301 .-# ∗ /"01!# 5 (2) #'( The resulting measure has a theoretical range from 0 to 10, with higher values indicating higher levels of affective polarization. In our sample the spread index ranges between 0 and 7.6, and the average and standard deviation are 2.4 and 1.1, respectively (see Wagner (2019) for further details about the calculation of this measure). The two countries with the lowest average levels of affective polarization in our sample are Taiwan and the Netherlands, while the two countries with the highest levels are Hungary and Romania. To operationalize the salience of elections we exploit individual-level variation in the timing of survey interviews after election day (see Singh & Thornton (2019) for an application of this identification strategy to the study of party identification). One could directly operationalize the salience of elections across countries and elections by measuring 5 The vote share of each party in the lower house is taken from the election of the CSES survey. However, the IMD does not provide information on the vote shares of parties in the lower house. We have, therefore, incorporated this data to the IMD drawing on data from CSES modules 1-4. Country-elections for which data on the vote shares of parties in the lower house is not available in the corresponding CSES module are excluded from the analyses. This implies that CSES post-electoral surveys fielded after a presidential election that was not held concurrently with a legislative/parliamentary election are excluded. 11 the degree of competitiveness of a given election, the level of partisan conflict during the campaign, or the campaign tone. However, these measures are likely to be endogenous to the levels of affective polarization in each country and election. In this paper, instead, we use the number of days that passed between the election day and the day when each respondent was interviewed as an exogenous measure of the salience of elections.6 We assume that as time passes since the day citizens went to the polls, elections and electoral competition will become less salient (Singh & Thornton, 2019). As time goes by, people will lose interest, the media will devote less time to politics, and the levels of partisan conflict will decrease. To illustrate this point, Figure A2 in the Appendix shows how the relative number of Google searches about the main national political parties and candidates sharply decrease after election day in multiple countries. It seems clear that right after election day most citizens lose interest in political competition. While the decline in Google searches is quite pronounced on the day just after the election, we expect that, many times, elections and partisan competition will lose salience in a more gradual way. Many times, elections are followed by transition periods that last a few months (e.g. in the US) or coalition bargaining that might take even longer (e.g. in Belgium). Therefore, during this post-election period the media will still pay some significant attention to political competition, although this media attention, as well as people’s interest on it, should be less intense than during the campaign period. In any case, the decline in election salience might be more pronounced soon after the election than months later (Singh & Thornton, 2019). This identification strategy hinges on the assumption that the timing of survey interviews is independent from potential outcomes of affective polarization. That is, that 6 In countries with two election rounds we measure the number of days between the second-round election day and the day when each respondent was interviewed. 12 those interviewed early and late during the survey fieldwork are comparable, so that any differences we might find in levels of affective polarization can be attributed to variation in election salience. For this to be the case, assignment of the day at which each respondent is interviewed should be as good as random. However, in most surveys this is an assumption that is unlikely to hold unless one makes further adjustments (Muñoz, Falcó-Gimeno, & Hernández, 2020). First, as shown in Figure A3 in the Appendix, the timing of survey interviews after elections varies considerably across countries and elections. While in some election studies all respondents are interviewed a few days after election day, in some other studies most respondents are interviewed months after the election. Since this variation in the starting date and the duration of the survey fieldwork across studies might be related to contextual differences that could, in turn, influence affective polarization, we conduct all our analyses through hierarchical models that include country-election random-intercepts for each CSES post-electoral survey. Second, some respondents are more elusive to interviewers and require greater time and effort to be reached for a survey interview (Muñoz et al., 2020). This might generate differences in the characteristics of those interviewed early and late during the fieldwork. To check for this possibility, we fit a series of hierarchical models that regress the time at which each respondent was interviewed on covariates known to affect the reachability of respondents.7 The models, summarized in Figure 1, indicate that there are some differences between those interviewed early and late during the fieldwork.8 For example, those who are 7 All these models are estimated through mixed-effects linear models with country-election random intercepts. We estimate one independent model for each of the covariates. 8 All figures in the main text and appendixes were generated using the plottig Stata schemes (Bischof, 2017) 13 older and out of the labor market are more likely to be interviewed earlier during the fieldwork. While the differences in the timing of interviews are small (e.g. those who are retired are interviewed, on average, just 2 days earlier than those who are employed) in all our models we control for all these respondents’ characteristics. It is much more plausible to assume that the timing of interviews is orthogonal from potential outcomes once we account for country-election differences and the characteristics of respondents related to their reachability. Figure 1: Respondents’ characteristics and timing of survey interviews (95% confidence intervals) Age Lowest income group (Ref:) Second lowest income group Income Middle income group Second highest income group Highest income group Gender Male (Ref:) Female No education (Ref:) Education Primary or lower secondary Higher secondary Post-secondary University Labor market status Employed (Ref:) Unemployed Student Retired Housework Permanently disabled Others Rural-Urban Village (Ref:) Small town Suburbs of large city Large city -2 -1 0 1 Survey interview timing (days) To analyze the mechanisms linking election salience and affective polarization we rely on measures of party attachment and ideological polarization. Party attachment is operationalized through a direct question that asks respondents how close they feel to the party they identify with. The variable ranges from 0 to 3, with 0 indicating that the 14 respondent does not identify with any party and 3 indicating that the respondent feels very close to that party. Measures of partisan identity that better tap into identity importance and sense of group belonging (see for instance Huddy et al 2015) would have been better if available. Our indicator is likely to lead to poorer predictions (Bankert, Huddy and Rosema, 2017), which in any case would run against our hypothesis. If our mediation hypothesis holds for our measure of party attachments, it should hold as well for measures of party identity. The measure of ideological polarization is based on each respondent placement of each of the parties included in the CSES on the 0-10 left-right scale. Like Wagner (2019), we estimate individual-level ideological polarization following the operationalization of 444444$ with affective polarization. That is, in equation (1) we substitute /"01!# and /301 !<=">"3