Fragmented foes: affective polarization in the multiparty context of the Netherlands

Affective polarization, or antipathy between the supporters of opposing political camps, is documented to be on the rise in the United States and elsewhere. At the same time, there are limits to our understanding of this phenomenon in multiparty contexts. How do citizens draw the line between 'ingroups' and 'outgroups' in fragmented contexts with multiple parties? Answering this question has been ha mpered by a relative lack of data on citizens’ views towards compatriots with opposing political views outside the US. This paper is based on novel data collection, measuring citizens’ evaluations of various political and non -political outgroups among a representative sample of 1,071 Dutch citizens. These data allow to study the extent and configuration of affective polarization in the highly fragmented context of the Netherlands. The analysis shows that respondents report more dislike towards political outgroups than towards almost all non-political outgroups. Rather than disliking all out-partisans equally, evaluations grow gradually colder as ideological distance towards a group increases. Affective polarization is especially strong between those who disagree on cultural issues, and between those who support and oppose the populist radical right. The article discusses how these findings enhance our understanding of affective polarization in multiparty systems.


Introduction
Affective polarization -or antipathy between citizens with opposing political views -has been identified as a major development affecting democracies (Iyengar et al. 2018;Reiljan 2019).Affective polarization can erode citizens' willingness to engage with opposing political views and the willingness to accept each other's democratic claims (Hetherington & Rudolph 2015;Strickler 2017).Taken to its extremes, affective polarization can spur dehumanization (Tappin & McKay 2019;Martherus et al. 2019) and lower the bar for political violence (Kalmoe & Mason 2018).While a functioning democracy can cope with heated disagreement, excessive affective polarization can erode the norms that underpin peaceful democracies (McCoy et al. 2018;Ziblatt & Levitsky 2018).
While research has vastly increased our understanding of affective polarization (see Iyengar et al. 2018), there is still a relative scarcity of research on affective polarization in multiparty systems.Westwood et al. (2015) successfully operationalized the concept in Spain, Belgium and the UK using trust games, showing that affective polarization in these contexts overtakes mutual hostility across linguistic or regional lines.Wagner (2017), Gidron et al. (2019b), and Reiljan (2019) studied affective polarization comparatively, showing the concept to travel well across borders.While these studies confirm that systems with more than two parties often experience just as much affective polarization as the US (or even more), there remain empirical and theoretical challenges to our understanding of AP in multiparty systems.This paper aims to address these.
Empirically, the study of affective polarization in multiparty systems has so far been hampered by a lack of measurements of dislike towards fellow citizens, which lies at the center of the concept.Virtually all studies of affective polarization outside of the US (Wagner 2017;Gidron et al. 2019;Reiljan 2019;cf. Westwood et al. 2015) has had to rely on respondents' evaluations of parties, usually sympathy towards parties on a like-dislike scale.While sympathy towards political parties is obviously empirically related to evaluations of these parties' supporters, the two are conceptually and empirically different (see Druckman 2017) and driven by partly different mechanisms (e.g.Mason 2015;2016).To explore the ways in which politics is driving a wedge between citizens, horizontal measures are needed of citizens' feelings towards each other.This paper reports the results of data collection of exactly such items: feeling thermometers and social distance towards various political and non-political outgroup members.These were fielded among a Dutch representative sample (N = 1,071).This allows to study the extent and configuration of affective polarization in a context that is less likely to experience strong affective polarization, given its fragmentation, low levels of partisanship, and an historically consensus-oriented political culture.
On the basis of this novel data, this paper contributes to an ongoing theoretical discussion regarding the question how affective polarization operates in multiparty systems, given that these do present a less clear-cut ingroup and outgroup (Wagner 2019).In such fragmented landscapes, positive affect is not likely to be confined to the supporters of just one party; nor do citizens necessarily dislike all those supporting other parties equally.This raises the question whether citizens divide the world in political in-and outgroups based on purely partisan cues, based on distinct blocks, or along some continuum.After all, political identities can also be informed by ideological positions (such as being a 'Left-wingers' or 'Right-wingers'; see Devine 2015) and more specific issue positions (see Hobolt et al. 2020).Of these, some issues -in particular, cultural issues -have been argued to be more affectively divisive than others (Gidron et al. 2019c).To test these assertions, I included measures of affect towards not only partisan outgroups (such as Green voters or Populist Radical Right voters), but also ideological outgroups ('Left-wingers' and 'Rightwingers') and economic and cultural issue outgroups (e.g.'those supporting [opposing] taking in more refugees').
In response to the methodological and theoretical issues identified above, this article investigates the extent and configuration of affective polarization in the multiparty context of the Netherlands by answering four interlocking questions.First, do citizens make a distinction in their evaluations between parties and their supporters?I show that respondents' evaluations of partisans correlate with their views of the respective parties, but far from perfectly so.Moreover, the residual affective polarization can be systematically predicted by variables that should theoretically foster it: ideological polarization and partisanship.Second, is such antipathy towards political opponents similar in magnitude to dislike along non-political lines?I show that, even in the less likely case of the Netherlands, respondents report systematically more negative affect towards political outgroups than non-political outgroups defined by region, urbanity, education, and ethnicity.Third, how do citizens define political in-and outgroups in a fragmented system?In line with Wagner (2019), I find that citizens do not show exclusive positive affect towards supporters of one party while feeling negative towards all others; instead, such feelings get gradually cooler as ideological distance grows.Fourth, do some issues or parties create more affective polarization than others?I find more affective distance, net of all other factors, between citizens who think differently about cultural (rather than economic) issues, as well as between supporters of populist radical right parties and the rest.
As I will discuss in the Conclusion section, these findings help to shed light on the complex patterns of affective polarization in multiparty systems.

Theory
Below, it is first discussed why, in order to understand affective polarization, it is important to explicitly study how citizens evaluate each other across political camps.I then raise the question whether such antipathy between citizens with a different political outlook approaches, or even exceeds, dislike along 'classic' non-political divisions, even in a less-likely case such as the Netherlands.In a third section, I move to the main feature that differentiates affective polarization in multiparty systems from the US context: the question who ingroups and outgroups are, given the fragmented system.I finish by discussing why it is likely that not all issues and parties are equally affectively polarizing.Each of these sections is accompanied by a research question.
Although these are embedded in theoretical expectations, I opted not to formulate them as formal hypotheses because some ask for a more explorative analysis.

Do citizens evaluate partisans differently than parties?
Citizens' political preferences can constitute a social identity, which is "that part of an individual's self-concept which derives from his knowledge of his membership of a group […] together with the value and emotional significance attached to the membership" (Tajfel 1979).In other words, we tend to see like-minded people as 'one of us'.A salient social identity fosters favoritism towards ingroup members and -depending on intergroup dynamics -a negative bias towards the outgroup (Brewer 1999).The latter phenomenon -bias towards citizens who are perceived as outgroups based on their political views -is the core of the concept of affective polarization.1 Indeed, the term affective polarization was coined by Iyengar et al. (2012) to describe such horizontal evaluations: "the tendency of people identifying as Republicans or Democrats to view opposing partisans negatively and copartisans positively" (Iyengar and Westwood 2014: 691).
Outside the US, Westwood et al. (2015) report trust games to gauge behavior towards a number of partisan groups in Belgium, the UK and Spain.Helbling and Jungkunz (2019) use vignettes and trust games to study social distance towards several partisan groups compared to non-partisan outgroups in Austria and Germany.Both studies show that outpartisans are generally judged more negatively than other outgroups.At the same time, neither of the two studies sets out to study affect towards the full range of partisan identity groups available, which can be quite extensive in multiparty contexts.
As a result, most of our comparative insights on the configuration of affective polarization are based on so-called sympathy ('like-dislike') scores towards political parties (Reiljan 2019;Gidron et al. 2019a;Wagner 2019).While this has yielded important insights, such as the fact that in multiparty systems affective polarization is often not directed equally towards all partisan outgroups (Wagner 2019), it is important to scrutinize the dependent variable used in these studies.As Druckman and Levendusky (2019) show, when asked about 'parties' in the abstract, respondents tend to think of the elite actors -i.e., the politicians and organizations.While party sympathy measures have the advantage of being widely available across countries and allows to obtain retroactive measures, there are also downsides to using party evaluations to study affect between citizens.
First of all, many of the normative concerns regarding affective polarization stem from the way it might impact interaction between citizens.Affective polarization makes citizens less likely to engage with those with opposing views, less responsive to information from the other side, more likely to discriminate against each other in non-political domains, and more inclined towards political intolerance or even violence (Strickler 2017;Hetherington & Rudolph 2015;McConnell et al. 2018, Tappin & McKay 2019;Martherus et al. 2019;Kalmoe & Mason 2018;McCoy et al. 2018).Many of these outcomes concern the way citizen relate to each other, rather than merely governing their interactions with parties or candidates as elite actors.If we want to know how politics divides societies, we need measures of such horizontal evaluations.
Secondly, there are theoretical reasons (as well as empirical evidence) to expect that citizens' evaluations of the two types of objects will often diverge.To be sure, citizens' evaluations of fellow citizens, when defined by political allegiance, are obviously related to their views of the respective party.If citizens dislike a particular party, they are ceteris paribus more likely to also dislike the citizens who support that party.This is confirmed empirically by Iyengar et al. (2012), who note that sympathy scores towards parties and partisans correlate at r = 0.69.At the same time, this correlation is far from perfect.Druckman and Levendusky (2019) experimentally varied the object of sympathy from Democratic or Republican 'voters' to 'candidates and elected officials' and found different patterns.The fact that citizens judge elites and voters differently is plausible, because there are factors that foster dislike towards political outgroups without being fully mediated by their views of parties.Social Identity Theory (SIT) tells us that antipathy towards outgroups is not a given, nor of constant intent towards all outgroups.Rather, this depends, among others, on the relative salience of the political identity (Hogg 2003), as well as the existence or absence of cross-cutting identities (Mason 2016) and the mental representation people have of outgroup members (Roccas and Brewer 2002).As a result, even if two citizens equally dislike a particular party, it is possible that one of them is much more negative about its supporters than the other.
In short, given its roots in social identities, it is plausible that affective polarization between citizens develops at least partly independently from the way citizens evaluate political elites.At the very least, the extent to which affective polarization can be meaningfully assessed using party evaluations is an empirical question that requires horizontal measures as well as vertical ones.If affective polarization is shaped by partly different mechanisms than evaluations of parties, the correlation between the two should not be perfect, and the divergence between the two measures should be systematically related to ideological polarization, partisanship, and political interestfactors that make it more likely that a political identity is salient to somebody.2RQ1: do citizens make a distinction in their evaluations between parties and their supporters?

Does politics trump non-political divisions?
Druckman and Levendusky (2019) find that citizens are systematically (and substantially) more negative towards parties than towards fellow citizens supporting these parties.As a result, party evaluations are less suited to compare the relative extent of affective polarization vis-à-vis other social divisions.By contrast, measures that directly capture views towards fellow citizens do allow for some comparison between political and non-political outgroups.Evidence in the United States suggests that politics is now more divisive than other social divisions such as class, religion, and perhaps even race (Iyengar and Westwood 2014).Westwood et al. (2015) show 'partyism' overtakes antipathy across longstanding linguistic or regional divisions in Britain, Spain and the UK.They argue that "the intensely competitive nature of democratic representation encourages parties to demonstrate overt hostility toward their opponents -hostility that is un-tempered by the social norms of respect and tolerance that regulate competition between most social groups" (idem: 334).Helbling and Jungkunz (2019) show this is also the case in Germany and Austria using vignettes and trust games.
The Netherlands can be considered a less likely case to find strong affective polarization.At the latest elections, 13 parties obtained representation in parliament, the largest of them with merely ~22% of the votes and votes (Kiesraad 2017).The landscape is thus extremely fragmentated.Furthermore, Dutch voters have relatively weak, although not absent, partisan identities (Bankert et al. 2017).The need for coalitions of usually three or four parties (and the absence of two clear alternating governing 'blocks') underlines that an enemy today might be a coalition partner in the future.This is generally reflective of the Dutch political culture, which has historically been regarded as highly consensus-oriented.It is therefore perhaps unsurprising that Reiljan (2019) shows that the Netherlands are the least affectively polarized country based on party evaluations.This makes the Dutch case uniquely interesting to benchmark the divisiveness of politics by comparing it to other social divisions.
Of course, making such comparisons is in practice affected by social desirability: as noted in the quote above by Westwood et al. (2015), it is often deemed less societally appropriate to dislike people based on non-political divisions than outgroups.At the same time, in the Dutch context regional, ethnic and religious divisions have in recent years been heavily politicized, too, which begs the question whether the social norms are still that much more prohibitive than they are towards political outgroups.Furthermore, the fact that social norms do not preclude prejudice towards political opponents to the same extent as it does across other divisions might be of substantive, not only methodological, interest, as it can actually explain why affective polarization can emerge quickly (idem: 336).
In short, it is relevant to compare the divisiveness of politics compared to 'classic' nonpolitical divisions in society.Studies in other contexts have found that politics trumps other divisions.To find out whether this is also visible in the less likely context of the Netherlands, I included measures of respondents' views of outgroups in terms of religion, ethnicity, region, urbanity, and education.RQ2: is antipathy towards political opponents similar in magnitude to, or does it even overtake, dislike towards non-political lines?

Who constitutes the outgroup in a fragmented landscape?
As noted before, affective polarization in the United States has been studied mostly as "the tendency of people identifying as Republicans or Democrats to view opposing partisans negatively and copartisans positively" (Iyengar and Westwood 2014: 691).While there exist differences in degree of Democratic or Republican identification, as well as a substantial share of Independents, this 'supply side' does create a relatively straightforward opposition that resonates with large swaths of the population (Iyengar et al. 2018).By contrast, the higher levels of fragmentation in multiparty systems matter for how citizens relate to any particular party and its supporters.Citizens in multiparty systems often do have meaningful partisan identities (Bankert et al. 2017;Greene 2004).Still, partisanship does not define citizens politically to the same extent as they do in the United States.Citizens often feel attached to more than one party (Garry 2007).It is also unlikely that citizens dislike all 'outpartisans' to the same extent.This raises the question how citizens define boundaries between ingroups and outgroups, if not by dichotomous inpartyoutparty lines (Wagner 2019).While most Green party supporters will experience affective distance towards a Populist Radical Right voter, their feelings towards Social Democratic supporters are probably much milder or even positively friendly.
The latter example suggests that an underlying ideological dimension matters for affective polarization between two given citizens.Citizens might identify themselves, broadly, as 'left-wing' or 'right-wing' and hence see people who see the world likewise as 'people like me' (Malka & Lelkes 2010;Devine 2015;Lelkes 2019).This might lead citizens to assess their fellow citizens as belonging in either of two blocks (Wagner 2019), which would still preserves a relatively dichotomous ingroup-outgroup distinction.However, it is also likely that citizens are capable of locating parties as being closer or further away from them ideologically (Dahlberg 2009).In that case it would follow that affective distance is correlated, in a more gradual way, to ideological distance.
In short, it is an open question is therefore whether citizens dislike all outpartisans equally or perhaps divide them into two 'blocks'.Alternatively, because political identities draw from multiple sources -such as party preferences, ideology, and/or issue positions -it is plausible that they gradually dislike fellow citizens to the extent that these appear more distant in terms of ideology or issues.RQ3: how do citizens define political in-and outgroups in a fragmented system?Do some issues and parties create more negative affect than others?The role of ideology brings up an additional complication.The political space citizens need to navigate is not only fragmented but also multidimensional, including (at least) an economic and cultural dimension.3Some parties are similar on economic issues and distant on cultural issues; for others the reverse is true.It has been argued that distance on cultural issues fosters more affective polarization than distance on economic issues (Gidron et al. 2019c;although cf. Iyengar et al. (2012): 442).This rest on the assertion, frequently observed in the public debate, that 'culture wars' are somehow more divisive, recently often under the banner of 'identity politics' (see e.g.Fukuyama 2018).Indeed, cultural (or in the US context, 'social') issues have been shown to align more closely with deep moral convictions or at least intuitions (Johnson and Wronski 2015).While economic issues, too, often reflect very deeply held convictions, it is possibly easier to defend a middle ground as morally acceptable.For these reasons it could be that citizens are especially loathing of fellow citizens who think differently about for instance immigration, national identity, or gender roles. 4n the other hand, the seeming heatedness of the cultural debate might simply reflect the controversial role of populist radical right (PRR) parties.These parties mobilize almost exclusively on cultural issues, but their ideology also has a populist component, which tends to be present among their supporters too (Rooduijn et al. 2016).It emphasizes the moral failings of all elites and the need to put into practice the 'general will' of the 'true' people (Mudde 2007).This moralization of politics likely affects both populist supporters and their opponents, realigning and crystallizing the 'mainstream' into an anti-populist camp opposing the populist camp (Moffit 2017), and so both groups can be expected to look upon each other disfavorably.Furthermore, PRR parties' host ideology of nativism is seen by many to overstep the boundaries of social and legal norms regarding prejudice (Blinder et al. 2010), and as a result these parties -and likely by extension their voters -tend to be 'stigmatized' by large swaths of the population (Harteveld et al. 2019).Indeed, Gidron et al. (2019b) find evidence for such 'radical right exceptionalism' in affective polarization.
In short, some issues might be more divisive then others; in particular, people might dislike citizens with opposing views on cultural issues more than those with opposing views on economic issues.However, if so, this might be confounded by the antipathy between populist radical right voters and all others.To disentangle these, the data collection reported here included not only ingroups and outgroups on partisan basis, but also towards ingroups and outgroups along general ideological (Left and Right) and issue (welfare, immigrants, and gender roles) lines.
RQ4: do some issues and parties create more affective polarization than others?

Design Data
The measures developed for this paper were fielded as part of the Dutch Longitudinal Internet Studies for the Social sciences (LISS) panel5 in August 2019.The LISS panel is drawn from a true probability sample of households (see www.lissdata.nl).Of the ~7,000 participants, 1,245 were randomly selected to fill in the affective polarization battery, and of these, 1,071 (86%) completed the questionnaire.

Measures
The new questionnaire consisted of vote choice, a feeling thermometer battery, and a social distance battery.The remaining variables could be obtained from the LISS panel's Politics and Values module which was fielded between January and March of the same year.There is thus a time lag of about half a year between these items, which might add noise to the correlations and, if so, make them somewhat conservative.Descriptive statistics of all items can be found in Appendix A (all appendices are in the Supporting Information).

Vote choice
The respondents were asked what they would vote if 'elections were held today'.They were presented with all parties that obtained representation in parliament; alternatively, they could indicate they voted for an 'other' party, did not vote, or voted blanc.This is taken to signify the respondents' current political allegiance.

Feeling thermometer
Respondents were asked express their feelings towards various groups on a continuous scale from 0 ("cold and negative"), through 50 ("neither warm nor cold"), to 100 ("positive and warm").6This 'feeling thermometer' is widely used in research in the United States and has been shown to correlate to more elaborate measures of outgroup affect (see Iyengar et al. 2018).The main advantage of this measure is that is allows, with relatively short batteries, to establish and compare affective ingroup-outgroup distance based on several identities.Respondents judged the following groups.The 27 items were presented in a randomized order.

Party (8 items)
"People who vote for …" The eight main parties were listed: conservative liberal VVD, populist radical right PVV, Christian democratic CDA, social liberal D66, Green GroenLinks, radical left SP, social democratic PvdA, populist radical right FvD.Ideology (2 items) "Left-wing people" and "Right-wing people"7 Issue (6 items) Refugees: "People who want to take in more refugees" and "People who want to take in fewer refugees" Welfare: "People who want to lower the general benefits" and "People who want to raise the general benefits" Gender roles: "People with a traditional views of the role of women" and "People with feminist views of the role of women" Nonpolitical (11 items) Education: "Lower educated people" and "People who studied in university" Ethnicity: "People with a Moroccan immigration background" and "People without an immigration background" Religion: "Christians", "Atheists", and "Muslims" Region: "People from the Randstad" and "People from outside the Randstad" Urbanity: "People who live in a back city" and "People who live in the countryside"

Social distance
Social distance measures give a more substantive understanding of the social impact of negative affect towards people with a different worldview.A social distance item (as first proposed by Bogardus [1926]) asks the respondent how he or she would react to an outgroup member becoming e.g. a neighbor, friend or family member.Given limited space, I relied on a single indicator that is very similar to the one used by Iyengar et al. (2012) in the US and the UK, stating: "Imagine a son or daughter of yours would get married.How would you feel about the following situations?".While this question is hypothetical to respondents without children, it does force them to envision intimate outgroup interaction.This battery was restricted to three items about outgroups derived from the thermometer question above.
• Party outgroup: "If she or he would marry somebody who votes [least liked party]" • Welfare issue outgroup: "If she or he would marry somebody who wants to [lower / raise] general benefits" (based on least favored position) • Refugees issue outgroup: "If she or he would marry somebody who wants to take in [more / less] refugees" (based on least favored position) The answer scale ranged from 1 (I would be very happy about that) through 4 ("I wouldn't mind") to 7 ("I would have a very hard time with that").Klar et al. (2018) raise the critique that this item also captures dislike of politics in general.While this is plausible, the negative effect should be similar across the items, and thus it would still be possible to compare between them.Druckman and Levendusky (2019) show that the social distance item actually correlates relatively poorly with thermometers.They suggest that an unwillingness to interact with others is better thought of as a possible behavioral outcome of negative affect.Because such outcomes are nevertheless of interest to gauge substantive implications, I include the measure while acknowledging these caveats.

Party sympathy (replication/validation)
The LISS panel includes a yearly battery that records respondents' sympathy towards all relevant parties on a 0 (dislike) to 10 (like) scale.This standard battery allows to compare respondents' sympathy towards parties to their views of these parties' supporters.

Economic and cultural distance between parties
To (indirectly) assess how ideologically apart supporters of different parties are, I rely on the ideological distance between the parties according to the Chapel Hill Expert Survey (CHES), a regular expert survey on party positions.The distances between two parties on economic issues is calculated based on the 'Economic left-right position' indicator.The distance on cultural issues is based on the indicator 'Position on immigration'.The scores were first standardized to allow for a better comparison between the economic and cultural dimension.

Other variables
As proxies for the salience of politics for respondents, I rely on ideological polarization as measured through a Left-Right position on a 0-10 scale (classified as 0-2 = Far Left; 3-4 = Center Left; 5 = Center; 6-7 = Center Right; 8-10 = Far Right); whether the respondent has a Party identity (1 if the respondent indicates to be either 'attracted to' or 'an adherent of' a party; 0 if neither); and Political interest (being Low, Middle, or High).

Results
The results section proceeds in four steps.I first test the relation between support for parties and their respective partisans.Do respondents provide the same answers to both?And to the extent that they do not, can the difference be explained by factors that should foster affective polarization independently (RQ 1)?I then move to a comparison between different identity dimensions, both political and non-political (RQ2).Finally, I move to a more detailed analysis explaining feelings towards partisan outgroups (to answer RQ3 and RQ4).

Do citizens evaluate partisans differently than parties?
Table 2 reports correlations between various indicators that all refer to the same party or its partisans.a "Strongly dislike" (0) to "Strongly like" (10) b "Cold and negative" (0) to "Warm and positive" (100) c "I would be very happy" to "I would have a very hard time" if a son or daughter would marry a supporter of this party Note: all correlations are significant at at least the 5% level.

Source: LISS
The correlation between respondents' sympathy towards a party (on a like-dislike scale) and their evaluation of the same party's supporters (on a feeling thermometer) is r=.66.This correlation is remarkably similar to the r=.69 found in the United States for the same association (Iyengar et al. 2012).This correlation confirms that evaluations of the two are related, but not identical.Interestingly, social distance towards an outpartisan does not correlate strongly with sympathy towards the respective party (r = -.20); but neither does it correlate very strongly to thermometer scores towards the same partisans (r = -.33).8This mirrors the low correlation with social distance found by Druckman and Levendusky (2019), who conclude that social distance measures likely capture a possible behavioral consequence of particularly high affective polarization.I therefore focus on the feeling thermometer in the main analysis and report in footnotes whether findings replicate using the social distance measure.
In short, evaluations of parties and partisans do not correlate perfectly.Still, this might merely reflect noise (perhaps due to unfamiliar scales, or the lag between the two items).In that case their divergence should not be systematic.If it reflects the fact that different mechanisms produce variation in affective polarization between citizens compared to party evaluations, then we should find more systematic divergences.To find out, I predict the feeling thermometer towards each party9 by the indicators mentioned in the theory section that can be expected theoretically to influence affective distance towards party supporters, controlled for party sympathy.Is the remainder theoretically explainable?Figure 1 shows the predicted affective rating based on this model (of which the full table is reported in Appendix B, which in addition controls for education, age and gender).Higher scores indicate warmer feelings.The variable Evaluated party denotes the partisan outgroup the respondent rated on the thermometer scale; the other variables are individual-level characteristics of the respondent her-or himself.It shows that variation in evaluations of fellow citizens of various partisan backgrounds, net of party sympathy, is not random, but rather systematically related to multiple factors.It depends, most clearly, on the partisan group that is being evaluated.Supporters of populist radical right parties (and to some extent towards to populist radical left SP) are systematically more disliked than would be predicted based on people's sympathy towards those parties.Indeed, the literature suggests these groups of voters are strongly held accountable for their political preferences (Harteveld et al. 2019).Supporters of centrist parties, by contrast, are evaluated more positively than would be predicted by sympathy towards those parties.This means citizens might dislike these parties but not necessarily judge the citizens supporting these parties.10Residual affective polarization is furthermore significantly predicted by some of the variables that should reflect the salience of politics for an individual.The variable Left-Right shows residual affect is higher among those who are ideologically more polarized, although mostly so in the case of the far left (p = 0.044).It is also predicted by Party identity: people who are not attached to any particular party are less negative towards fellow citizens than those who do have a partisan identity (though only significantly so at the 10% level; p = 0.077).Third, it is merely descriptively higher, but not significantly so, among those who are highly interested in politics (p = 0.111). 11ll in all, even in this conservative model controlling for party sympathy, several patterns emerge that suggests that Dutch citizens are relatively judging of political opponents to the extent that politics is more important to them, and to the extent that their opponents are more politically extreme.This suggests that affective polarization as a 'horizontal' evaluation is derived from party different mechanisms, and thus relevant to study on its own.Below we continue to do so.

Does politics trump non-political divisions?
Citizens judge each other based on their worldview, but how much so?To explore whether affective distance towards political outgroups is comparable in extent to that towards non-political outgroups, Figure 2 maps the average thermometer rating towards ingroups and outgroups on the various political and non-political identities.In each case, the ingroup score is assumed based on the highest score of a particular identity dimension (for instance, the most liked of all listed party supporters); the outgroup score from the lowest score handed out within the same dimension. 12In case there are more than two groups in a dimension (that is, the 8 parties and 3 religious identities), the figure in addition shows the mean score assigned to all non-ingroup positions.The identity dimensions are presented in an ascending order of the lowest score assigned to the outgroup.

FIGURE 2 THERMOMETER SCORES BY IDENTITY DIMENSION
Note: with 95% confidence intervals.

Source: LISS
The first striking feature of Figure 2 is that almost all political identity dimensions are associated with greater affective distance than non-political identities.The exception is religion, which creates greater affective distance towards the outgroup than Left-Right ideology.The affective force of religion is due to the category Muslim, which receives very strongly negative affect of citizens on the far right (20 o , compared to 50 o on the center left and 40 o on the center right).This does not invalidate the importance of negative affect along religious lines, but rather highlights that the order of importance differs between groups.As Appendix D shows, among those who are at least moderately interested in politics, political divisions do trump all non-political ones, including religion.
In short, Dutch citizens report more negative affect towards political outgroups than towards almost all other studied outgroups.As discussed in the theory section, this pattern could reflect social desirability: it is deemed less societally appropriate to dislike people based on these latter characteristics compared to political outgroups.Still, the predominance of political divisions over others replicates earlier studies in the US (Iyengar et al.) and Europe (Westwood et al. 2015;Jungkunz & Helbling 2019).Furthermore, it is an open question whether social desirability really applies as strongly to categories such as region and religion, which have recently come with heated debate.As a case in point, strong dislike towards a religious outgroup is openly declared by a large part of the sample.I therefore tentatively conclude that political identities are across the board as divisive as many other socio-demographic ones even in the Netherlands, which is classified in Reiljan (2019) as the least affectively polarized society.This is a relevant assertion, because prior studies relying on party sympathy scores could not make such comparisons.Furthermore, looking forward to R3 ("do some issues and parties create more affective polarization than others?"), Figure 2 suggests that cultural issues -refugees and gender roles -are indeed more affectively divisive than economic ones (welfare).At the same time the differences should not be overstated either: the differences in average distance between refugees and welfare are minor. 13I therefore study this potential 'culture wars effect' also in the analysis of the party dyads further below.

Who constitutes the outgroup in a fragmented landscape?
We can now turn to the question how citizens distinguish ingroups and outgroups in a fragmented multiparty system like the Netherlands.To find out, Figure 3 reports, for each ingroup party, how its supporters feel towards the supporters of each other parties.
Unsurprisingly, voters like the supporters of their own party -the scores on the diagonalmost.But they do not dislike all out-parties equally; and neither do two clear 'blocks' emerge.Rather, evaluations turn gradually cooler as partisans are ideologically more distant.For instance, left-wing parties are positive to mildly negative about the supporters of other left-wing parties, somewhat more negative about centrist voters, still more negative about right-wing voters, and the most negative about far right voters.The same is roughly true, mutatis mutandis, for right-wing voters.In short, ideological distance is a predictor of affective distance.This suggests that in multiparty systems, citizens do not think in dichotomous ingroups and outgroups, but rather make a more gradual distinction in their evaluations (Wagner 2019).
The pattern that comes closest to a dichotomous configuration is that between populist radical right (PRR) supporters and the rest.Mainstream party voters are relatively uniform in their dislike of PRR parties (PVV and FvD): although this dislike is higher on the mainstream left (~22 o ) than mainstream right (~33 o ), even the mainstream right is still more negative about PRR voters than they are towards mainstream left voters (low 40 o 's).In return, especially supporters of the oldest PRR party Party for Freedom (PVV) make relatively little distinction between voters of the mainstream left and right. 14In other words, PRR and mainstream supporters are not only relatively negative towards each other, they are also more homogeneously so, suggesting the emergence of two PRR/non-PRR 'blocks' (Moffit 2018).This 'PRR exceptionalism' will be explored in more detail below.

FIGURE 3 AVERAGE FEELING THERMOMETER RATINGS TOWARDS PARTISAN OUTGROUPS
Source: LISS A second pattern that emerges is that affective distance appears stronger along cultural than economic lines.For instance, dislike between VVD and SP, two opponents on the economic dimension, is less heated than that between the cultural opposites Forum for Democracy (FvD) and the Greens (GroenLinks).However, this might be confounded by 'PRR exceptionalism'.Because this descriptive figure cannot disentangle all possible factors, I return to a more formal analysis of these phenomena below.

Do some issues and parties create more negative affect than others?
To disentangle whether some issues or parties receive more negative affect, I predict citizens' evaluations towards each party in a regression model.To this end, I created a stacked ('long') dataset containing dyads of each respondents' evaluation of each partisan outgroup.This allows to explain thermometer scores by dummies for the respondent's inparty and dummies for the parties that are being evaluated.If some parties systematically receive or radiate more dislike, this will be captured by these dummies.In a second model, the ideological distance between the inparty and outparty on both economic and cultural issues (based on the Chapel Hill Expert survey) is added.This allows to test whether cultural issues still create more affective distance even when controlling for the party dummies (and thus possible 'PRR exceptionalism').In a third model, party sympathy is added.As in the first analysis reported above, this conservative model isolates the affect towards fellow citizens that is not entirely due to views about parties.All models contains random intercepts for respondents.Only evaluations towards parties the respondent did not vote for are included.Figure 4 shows a coefficient plot of the results (the full regression tables can be found in Appendix E).Negative (positive) numbers indicate that a variable is associated with lower (higher) scores on the feeling thermometer.
The party dummies in the first model replicate the pattern visible in Figure 4: PRR parties (especially the oldest PRR party, PVV) are both senders and (especially) recipients of uniquely negative affect.The second model adds ideological distance to test whether there exists any additional 'culture wars effect'.The two coefficients show that both economic and cultural distance is associated with a more negative evaluation, but more strongly so in the case of the latter.The third model controls for sympathy towards the party, which means that the only variation that is left is purely affective towards supporters.In this model, the residual affect is only predicted by cultural issue distance.
In short, even in the most conservative model, the data thus confirm the existence of both a 'culture wars effect' and 'PRR exceptionalism'.Affective polarization is stronger towards opponents that think differently on cultural issues, as well as towards those who are on the other side of the PRR/non-PRR divide.This would mean that if cultural issues become more salient and/or PRR parties grow, the aggregated level of affective polarization can be expected to grow too.

FIGURE 4 COEFFICIENT PLOT OF MODELS PREDICTING THERMOMETER RATINGS
Note: based on models with respondent-outparty dyads and a random intercept for respondents.

Conclusions
The emergence of the concept of affective polarization has increased our understanding of political divisions in the United States and elsewhere.It has brought to the fore that growing polarization between citizens is about more than a divergence of worldviews: to a substantial degree, hostility between political camps is the result of intergroup dynamics (Iyengar et al. 2018;Mason 2016).A growing literature has established that affective polarization occurs in countries around the world.Still, with some notable exceptions (e.g.Westwood et al. 2015;Wagner 2019;Reiljan 2019) most theorizing and measurement has concentrated on the United States.This paper set out to study affective polarization in the multiparty context of the Netherlands.Whereas almost all previous work on affective polarization outside the United States has relied on citizens' evaluation of political parties, this paper presented novel survey data from the Netherlands that asked respondents for their views towards a broad array of political outgroups defined by party preference, ideology, and issue positions, as well as non-political outgroups.These data allowed me to draw four conclusions about the extent and configuration of affective polarization in this highly fragmented multiparty system.
First, the data show that Dutch respondents distinguish 'vertical' evaluations of political parties from 'horizontal' evaluations of their fellow citizens: the correlation between the two is only moderately strong.Importantly, the divergence between the two can be explained along theoretical lines.Centrist parties are often disliked while their voters are not (or less so), while supporters of the controversial populist radical right (PRR) are judged even more harshly than would be expected based on the evaluation of the respective party.Affective distance is systematically higher, even net of party sympathy, among respondents for whom politics is probably more salient for their identity: those who are ideologically polarized and who have a party identity.In short, if we want to understand whether and how politics divides citizens, it is relevant to measure 'horizontal', intercitizen, as well as 'vertical', citizen-elite, evaluations.
Second, respondents generally reported more antipathy towards outgroups along political lines than outgroups along the non-political lines of education, ethnicity, urbanity, or region.(The one exception was the religious group of Muslims, which was judged more unfavorably than ideological opponents by a sizeable share of the sample.)To be sure, this primacy of politics should be confirmed using (e.g.implicit association) measures that are less sensitive to social desirability.However, as Westwood et al. (2015) note, the absence of strong social norms that preclude prejudice towards political opponents is actually a potential explanation why such animus could grow undisturbed in many contexts.It is therefore at least very suggestive that a primacy of political divisions also appears in the Netherlands, which is highly fragmented, has a historically consensus-oriented political culture, and was reported the least affective polarized (based on party evaluations) in Reiljan (2019).
Third, the respondents did not rigidly distinguish between one partisan ingroup and all other voters, and nor did two 'blocks' emerge (Wagner 2019).Rather, they tend to dislike fellow citizens more to the extent that these are more ideologically distant.This finding is not incompatible with the growing consensus in the literature that on the aggregate level ideological and affective polarization are only weakly, if at all, correlated (Lelkes 2019).While in any given context citizens will dislike those (objectively or subjectively) furthest away more, there are also other mechanisms at work that impose a given ideological distance with much more antipathy in some instances than others.
Fourth, the analysis provided evidence for two of such factors: a 'culture wars effect' and 'PRR exceptionalism'.First of all, in contrast to Iyengar et al. (2012) but in line with Gidron et al. (2019c), cultural issues tend to create more antipathy than economic issues.Respondents give somewhat lower thermometer scores to people who disagree with them on the issues of refugees or gender roles than on economic issues.Moreover, when predicting the level of affect between the supporters of two parties, cultural distance (according to the Chapel Hill Expert Survey) was associated with much more affective distance than were economic issues.Secondly, I found clear evidence for populist radical right exceptionalism.Consistent with the party line (Mudde 2007), populist radical right supporters have a strong and relatively homogeneous dislike for the supporters of mainstream parties, even controlling for ideological distance and party sympathy.This resonates with the moralizing and homogenizing 'anti-mainstream' core of populist ideology.In return, PRR supporters were even more disliked by mainstream voters.This provides evidence that for many citizens, these are parties 'one does not vote for' (Harteveld et al. 2019).As a result, populist radical right parties foster a 'double boost' of antipathy to the system by being both the object and subject of unique antipathy, in addition to swelling the salience of cultural issues.In time, the populist moralization of politics might even spill over into the relations between other voters, charging politics as a whole.This remains an open question that calls for longitudinal data.
Several topics remained outside of the scope of this study.For instance, there are many more factors on the individual level that shape affective polarization, including information and media environment (Hutchens et al. 2019) and social sorting (Mason 2016).Furthermore, the relative importance of politics vis-à-vis other divisions in society, as well as the relative divisiveness of some issues over others, will likely differ between groups in society.Still, the findings in this paper invite me to present two recommendations.First, the study of affective polarization in multiparty systems can benefit from theorizing further about the gradual distinctions citizens make towards various political outgroups, rather than assuming affective polarization to be configured in a relatively dichotomous sense.Second, there is a need for more theoretical and empirical work that link micro mechanisms to macro patterns (Gidron et al. 2019b).How can individual-level patterns, such as the role of ideological distance, the primacy of cultural issues, and PRR exceptionalism, account for some of the increases and decreases in affective polarization over time and between countries?This can shed more light on the way affective polarization is affecting twoparty and multiparty systems alike.

FIGURE 1
FIGURE 1 PREDICTED AFFECT TOWARDS PARTISAN GROUPS BASED ON A MODEL CONTROLLING FOR PARTY SYMPATHY

TABLE 2 CORRELATIONS
BETWEEN ITEMS REFERRING TO THE SAME PARTY