Electoral Studies xxx (xxxx) xxx

Contents lists available at ScienceDirect

Electoral Studies
journal homepage: http://www.elsevier.com/locate/electstud

Affective polarization in multiparty systems

Markus Wagner
Department of Government, University of Vienna, Rathaussstrasse 19/1/9, 1010, Vienna, Austria

A R T I C L E I N F O A B S T R A C T

Keywords: Affective polarization captures the extent to which citizens feel sympathy towards partisan in-groups and
Affective polarization antagonism towards partisan out-groups. This is comparatively easy to assess in two-party systems, but capturing
Ideological polarization the pattern of affect towards multiple parties is more complex in multiparty systems. This article first discusses
Negative partisanship
these challenges and then presents different ways of measuring individual-level affective polarization using like-
Partisanship
dislike scores, a widespread measure of party sympathy. Using data for 51 countries and 166 elections from five
modules of the Comparative Study of Electoral Systems, I then show that affective polarization adds to existing
concepts as a way of understanding political participation and democratic orientations. Studying affective po­
larization outside the US could therefore have important consequences for our understanding of citizen per­
ceptions of politics as well as citizen behaviour, but we need the appropriate measures to do so.

1. Introduction emphasised the importance of negative partisanship for understanding

voter attitudes and behaviour (Medeiros and Noël, 2014; Mayer, 2017;
Since Campbell et al. (1960), the study of partisanship has been Rose and Mishler, 1998). To capture the extent to which citizens hold
recognized as a key determinant of political attitudes and behaviour, both positive ingroup affect and negative out-group affect towards
making party identification one of the most well-established theories in parties, researchers have coined the term ‘affective polarization’ (Iyen­
research on elections and voting (Johnston, 2006). In much of this gar et al., 2012; Lelkes, 2016). In extremer cases, we can also speak of
research, partisanship has generally been considered a type of social ‘partisan prejudice’ and ‘interparty hostility’ among partisans (Lelkes,
identity (Greene, 1999, 2004; Huddy, 2001), with partisans holding a 2016; Lelkes and Westwood, 2017).
strong in-group affect towards the party they favour and identify with However, the concept of affective polarization has so far mainly been
(Brewer and Brown, 1998).1 applied to what is arguably the most straightforward case: the American
Yet, in recent work scholars have argued that to understand parti­ two-party system. Given the prominence of this concept in recent ana­
sanship and its consequences fully we also need to take into account out- lyses of American electoral and party politics, it is not surprising that
group affect, i.e. how negatively voters feel about competitors (Iyengar there have been a number of recent attempts to apply this concept to
et al., 2012; Iyengar et al., 2019). Of course, this was already recognized other countries. Important examples are Westwood et al. (2018), Ward
by Campbell et al. (1960: pp.121f.), who argued (emphasis added): ‘the and Tavits (2019), Reiljan (2019) and Gidron et al. (2018). This research
political party serves as the group toward which the individual may has shown that affective polarization is widespread outside the United
develop an identification, positive or negative, of some degree of in­ States, is not a simple by-product of ideological polarization and alters
tensity’.2 In this recent revival of interest in negative partisanship, perceptions of party competition.
Iyengar et al. (2012) and Hetherington and Rudolph (2015) show that However, apart from Reiljan (2019), these papers do not engage
the key change in partisan affect in the United States over the past 50 extensively with the conceptual issues that arise from moving to
years has been that American voters who identify with either the Re­ multiparty settings. Yet, analyzing affective polarization multiparty
publicans or the Democrats have become more negative in their atti­ systems is more demanding simply because polarization is no longer
tudes towards the other party, moving from moderate levels of sympathy between one in-party and one out-party, with polarization the affective
towards increasingly strong dislike. Other, related research has distance between the two. Applying the concept to multi-party systems

E-mail address: [email protected].

1
An alternative view of partisanship as a running tally of evaluations was put forward by Fiorina, 1981.
2
This point had also been made by other earlier research (e.g. Garry, 2007; Maggiotto and Piereson, 1977; Richardson, 1991), but had largely been ignored by the
discipline. Perhaps the increasing ideological polarization in the United States led to this renewed interest in negative out-group feelings among researchers.

https://doi.org/10.1016/j.electstud.2020.102199
Received 10 December 2019; Received in revised form 22 July 2020; Accepted 22 July 2020
0261-3794/© 2020 Published by Elsevier Ltd.

Please cite this article as: Markus Wagner, Electoral Studies, https://doi.org/10.1016/j.electstud.2020.102199
M. Wagner Electoral Studies xxx (xxxx) xxx

therefore means that we need to be more specific about our under­ Carsey and Layman, 2006; Johnston, 2006). Positive affect towards
standing of affective polarization. To address this challenges, this article one’s favoured party is the first component of affective polarization.
makes two recommendations concerning how affective polarization The second component of affective polarization is negative affect
should be conceptualized and measured in multiparty systems. towards other parties and their supporters. In terms of social identity,
First, affective polarization in multiparty settings should be defined this is the extent to which other parties and partisans are seen as ‘out-
and assessed as the extent to which politics is seen as divided into two groups’ (Greene, 1999; Brewer and Brown, 1998). Beyond the relatively
distinct camps, each of which may consist of one or more parties. This binary concept of (positive) partisanship, individuals arguably have
approach is consistent with prior evidence that multiple positive iden­ some level of positive or negative affect towards all parties they know
tification is possible and widespread in multi-party systems (Weisberg, about (Richardson, 1991).
1980; Garry, 2007). In the US context, research on affective polarization However, ‘out-group derogation’ is not an essential counterpart to
generally assumed the existence of positive in-group towards a single ‘in-group favouritism’ (Greene, 2004; Brewer and Brown, 1998). Hence,
party, but this is not appropriate for multiparty contexts, especially as already pointed out by Weisberg (1980), Richardson (1991) and
when examining the polarization of affect towards various groups. Maggiotto and Piereson (1977), individuals, including strong partisans,
Second, the size of parties matters for levels of affective polarization. can vary quite extensively in the extent to which they hold negative
Hence, a citizen’s perception of the political system is more affectively views of opposing parties and their supporters. In accordance with the
polarized if they feel negatively towards a large competitor than if they focus on attitudes towards out-groups, scholars have also argued that
dislike a minor party. For example, in Germany, negative feelings to­ there is ‘negative partisanship’ as well as positive partisanship
wards the Alternative for Germany (AfD) matter less than negative (Medeiros and Noël, 2014; Abramowitz and Webster, 2016).
feelings towards the Christian Democrats. Similarly, affective polariza­ In a two-party system such as the United States, affective polariza­
tion in Germany should increase if the AfD doubles its vote share tion is captured by the extent to which voters feel (a) positive in-group
(holding the voter’s negative feelings towards the party stable). Thus, affect towards a party and its supporters and (b) negative out-group
overall measures of affective polarization need to be weighted by party affect towards the other party and its supporters. Overall, affective po­
size. larization then captures the extent to which a voter has an ‘us-versus-
Building on this understanding of affective polarization, I then pro­ them’ perception of the party system.
pose a measurement approach for affective polarization at the individual Affective polarization is both an aggregate-level and individual-level
level and distinguish it from other approaches. It allows researchers to phenomenon. There are thus two meanings of affective polarization,
measure affective polarization for non-partisans and partisans alike, which should not be conflated: (a) each individual has a level of in- and
unlike the measure proposed in Reiljan (2019). I implement this in 51 out-group affect that may be polarized or not (individual-level affective
countries and in 166 elections using data from the Comparative Study of polarization); and (b) each political system has an average level of such
Electoral Systems. As pointed out by Druckman and Levendusky (2019) patterns of in- and out-group feelings (aggregate-level affective polari­
and Iyengar et al. (2019), such scores do not capture the full range of zation). Aggregate-level affective polarization is simply an averaged
meanings of affective polarization. However, given their widespread version of individual-level affective polarization. In general, the term
availability, like-dislike scores are a suitable place to start to engage in ‘affective polarization’ is used more to describe aggregate-level than
cross-national comparisons of affective polarization (see also Reiljan individual-level patterns (Reiljan 2019).
2019; Ward and Tavits, 2019; Gidron et al., 2018; Boxell et al., 2020). In this article, I focus more on affective polarization at the individual
The measurement of affective polarization presented here is based on level, so whether an individual’s pattern of affect towards parties is
like-dislike scores. Descriptive evidence from this data shows that af­ polarized or not. A parallel measure is the perceived level of ideological
fective polarization in the US is average in international comparison, polarization: voters differ in whether they see the parties in the system
confirming findings in Westwood et al. (2018) and Gidron et al. (2019). as ideologically similar or ideologically distant.
I then examine how affective polarization is associated with parti­
sanship, ideological polarization, democratic values and participation. 3. Affective polarization in multiparty systems
These assocations are examined for four different measurement ap­
proaches. The key finding here is that, while the overall conclusions There is little doubt that affective polarization matters outside the
remain the same, associations with other variables do differ in appre­ Unites States. Research by Westwood et al. (2018), which examines
ciable ways depending on which measurement approach we choose. levels of trust between partisans in the UK, Belgium and Spain, provides
These analyses also demonstrate that individuals’ level of affective important evidence that affective polarization is just as important in
polarization differs theoretically and empirically from partisanship and many countries outside of the US and also has a greater impact than
from ideological polarization. Moreover, affective polarization is asso­ social cleavages. Using experimental data, Helbling and Jungkunz
ciated cross-nationally with democratic beliefs and political participa­ (2019) also show that partisan identities form the basis of social divides
tion in ways that go beyond partisanship and ideological polarization. in Germany and Austria. Gidron, Adams and Horne (2019) present
Hence, in addition to presenting a conceptualization and measurement cross-country evidence that affective polarization in the US is by no
approach for affective polarization in multiparty systems, this article means unusually high (see also Boxell et al., 2020). Using CSES data,
shows how affective polarization can add to our understanding of po­ Ward and Tavits (2019) show that affective polarization shapes citizen
litical attitudes and behaviour in ways that go beyond existing concepts perceptions of political competition and party systems. Finally, Reiljan
such as partisanship and ideological polarization. (2019) shows evidence of widespread hostility to out-partisans in
Europe.
2. Partisanship and affective polarization Yet, this research tends not to reflect specifically on the challenges of
moving to a multiparty setting. Doing so raises important questions
Affective polarization builds on the research portraying partisanship concerning the conceptualization and measurement of affective polari­
as a social identity (Greene, 1999, 2004; Huddy et al., 2015; Iyengar zation. To be a useful description of how an individual relates to political
et al., 2019). Partisan ties are often generated during early adulthood parties in the party system, a conceptualization and measure of affective
and often remain largely fixed over the life course, leading to substantial polarization in multiparty systems should summarize the configuration
stability in electoral decisions (Niemi and Jennings, 1991; Shively, of feelings towards political parties (and their supporters). In doing so, it
1979; Dinas, 2014a, 2014b). Moreover, party identification does more needs to go beyond the binary measures of positive and negative
than just determine our vote choice: it helps to shape how citizens partisanship: these only measure the party that voters feel closest to and
develop attitudes and engage with politics (Campbell et al., 1960; the one that they dislike most (Medeiros and Noël, 2014; Abramowitz

2
M. Wagner Electoral Studies xxx (xxxx) xxx

and Webster, 2016; Mayer, 2017). As we will see, applying the concept two of these parties and dislikes a third. A voter’s affective polarization
of affective polarization to multiparty systems requires us to be more would arguably be more severe if the disliked party was Labour rather
specific in how we understand partisan identities. than the Liberal Democrats.
Two configurations of feelings towards parties are easy to assess by From a dynamic, over-time perspective, we would also want affec­
applying existing conceptions of affective polarization to multiparty tive polarization to vary over time depending on the size of each party.
systems. First, affective polarization is low if a citizen has the same level Take the example of a strongly disliked party that increases in size, for
of affect towards all parties and partisans. This is the case if a citizen instance a voter who strongly dislikes an emerging radical-right party. If
dislikes all parties, likes all parties, or is indifferent about all parties. this party doubles in popularity, then this dislike is more relevant for
Politically disaffected voters who feel negatively towards parties in understanding party competition and should be reflected in measures of
general are therefore not affectively polarized (Klar et al., 2018). affective polarization. Hence, if we want to capture a society’s affective
Second, affective polarization is high if a citizen likes one party and polarization, introducing weights into our measures may be helpful.
dislikes all others. This is the classic pattern expected by traditional We now turn towards operationalizing this approach with survey
accounts of partisanship as a social identity. In this view, feelings of data.
closeness towards parties are mutually exclusive as they represent
competing identities, so voters are unlikely to feel close to more than one 4. Measuring affective polarization in multiparty systems
party at once.
One option for assessing affective polarization in multiparty systems To measure affective polarization, we need to capture the extent to
would be to proceed with such a strong conceptualization of partisan­ which individuals feel positively towards one or more parties and
ship as a social identity. This would mean assuming that only one party negatively towards other parties. Ideally, this measurement would relate
can be the ‘in-group’. Affective polarization measures in multiparty to political identities and would not refer directly to ideological or socio-
systems should then capture the extent the single in-group is seen demographic aspects. So far, there is no single agreed way of measuring
positively compared to the out-group, i.e. all other parties (Reiljan affective polarization, but various approaches have been suggested. For
2019). example, affective polarization has been measured directly using atti­
However, the difficulty of measuring affective polarization in tudes towards having one’s child marry a supporter of the party and
multiparty systems arises from the fact that many citizens feel positive using questions assessing stereotypes about out-partisans (Iyengar et al.,
towards two or more parties, while disliking others. Empirical research 2012). It has also been measured indirectly using implicit attitude tests,
shows that positive affect towards parties is indeed often not confined to trust games and dicator games (Iyengar and Westwood, 2015; Westwood
one party. For instance, research on partisanship in multiparty systems et al., 2018; Iyengar et al., 2019).
shows that multiple identification appears to be possible (Weisberg, Instead of these approaches, I measure affect towards parties using
1980; Garry, 2007). Hence, people can have more than one positive the standard like-dislike question included in election surveys; for a
party identification (Garry, 2007). Hence, citizens may exhibit clear similar approach see Reiljan (2019), Gidron et al. (2019) or Ward and
polarization in terms of their affect, but without positive affect being Tavits (2019). Using the extent to which citizens like (or feel warmly
exclusive to one party. Such patterns may be quite frequent in larger towards) some parties and dislike (or feel coldly towards) other parties
party systems such as Sweden or Finland, where parties form distinct has been used to measure affective polarization by several groups of
ideological camps or blocs. researchers (Iyengar et al., 2012; Rogowski and Sutherland, 2016;
In addition, party identification is by some accounts a declining Abramowitz and Webster, 2016; Lelkes et al., 2017).
phenomenon in many countries (Dalton, 2002). However, this devel­ The key advantage of using feeling or like-dislike scores is their
opment is accompanied by the rise of parties that elicit quite strong broad availability: few other questions concerning affective reactions to
negative affect, both on the radical right (e.g. Germany, Sweden, parties have been asked systematically in surveys. These questions also
Denmark) or on the radical left (e.g. Greece, Spain). The emergence of tap into the essence of affective polarization, namely an overall positive
such parties is not necessarily accompanied by strong in-group feelings. or negative reaction to parties as a whole.
This means that affective polarization can also increase absent positive What like-dislike scores do not capture is affect towards fellow or
affect, a dynamic that is less well-captured by existing measures, at least opposing partisans rather than parties (Druckman and Levendusky,
conceptually. 2019). This is better measured using other approaches (e.g., Iyengar and
Hence, a better approach to assessing affective polarization should Westwood, 2015; Iyengar et al., 2012). Hence, using like-dislike scores
be based on the spread of positive and negative affect (see also Ward and of parties tells us less than other measures about affective polarization as
Tavits, 2019). Using such an approach, affective polarization would also a manifestation of social identities. Moreover, the general nature of
be high for the citizen who feels strong positive affect towards two like-dislike scores means that it is inherently open to individuals what
parties and negative affect towards their competitors. This more they use to provide assessments – e.g., identities, ideology or simple
expansive definition of affective polarization thus covers situations affect (Lelkes, 2016, 2019) – even if some authors argue that feeling
where citizens perceive there to be two general camps, which may each thermometers primarily measure affective responses (Marcus, 1988;
be made up of one or several parties. Marcus et al., 2000). These shortcomings should not prevent us from
A separate question is whether affective polarization should be examining affective polarization using this existing measure, especially
weighted by the size and therefore relevance of parties. Arguably, it given recent evidence that different measures of affective polarization
matters more if the liked and disliked parties are very large competitors tend to correlate strongly (Druckman and Levendusky, 2019).3
than if a voter happens to dislike a small party. This is because size and
importance are often strongly correlated: larger parties are more
4.1. Measurement approach: spread of like-dislike scores
important for the party system and government formation. The key to
understanding the affective configuration of party systems is arguably
Given the preceding discussion, I suggest that affective polarization
whether the larger parties within that system are affectively distant or
close. Curini and Hino (2012) similarly justify weighting their left-right
polarization scores. 3
Note that like-dislike scores also suffer from differential item functioning: in
Weighting is sensible from both a cross-sectional and a dynamic the words of Lelkes (2016: p.401), ‘for some, a warm glow envelops all groups;
perspective. Cross-sectionally, imagine a system with two large parties for others, the world is a much chillier place’. The measures presented below
and one smaller one, akin to the UK in the early 2000s with Labour, the avoid this problem to a certain extent by focusing on differences in scores rather
Conservatives and the Liberal Democrats. Now take a voter who likes than absolute levels.

3
M. Wagner Electoral Studies xxx (xxxx) xxx

should be measured as the spread of like-dislike scores for each asked about in the survey.
respondent. This approach recognizes that individuals can feel posi­ An example calculation of affective polarization scores for nine ideal-
tively towards more than one party. An individual with low affective typical voters is shown in Table 1. The aim of these examples is to show
polarization then has similar levels of affect for all parties, be it positive how the measures behave in different scenarios and that the measures
or negative. In contrast, an individual with high affective polarization have face validity. The scenario used is based on a simple party system
has very different levels of affect for the different parties. This measure with two large parties, parties A and B, and two small parties, parties C
of affective polarization is based directly on existing measures of and party D. The vote shares are normalized so that they sum to 100.
perceived ideological polarization, which also see polarization as the The first two voters are ideal-typical party identifiers who support
spread of the positions parties are seen as having (Dalton, 2008; Ezrow, one party and dislike all others. The first and second voters differ in the
2007). It is also how Huddy et al. (2018) implicitly assess affective po­ size of the party they identify with. For the unweighted measures, this
larization in Sweden: they look at polarization in partisan affect between does not change the result. However, the weighted measure does differ
the two large coalitions of parties. It is also how Ward and Tavits (2019) for the two voters, as the first voter shows up as having a greater level of
essentially measure affective polarization. affective polarization in the weighted version. This reflects that the first
Measuring affective polarization as the spread of like-dislike scores is voter supports a large party, so the party system is more evenly divided
simple: it is the average absolute party like-dislike difference relative to into two camps for the first voter. So, the weighted spread-of-scores
each respondent’s average party like-dislike score. More formally, it is: measures also depends on whether voter sympathies create two
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑P ( )2 similarly-sized camps. Arguably, affective polarization is indeed higher
p=1 likeip − likei if the one’s own group is confronted by an out-group of similar size.
Spreadi =
np The third and fourth voters each have one ‘in-party’ but also have
strong positive affect for another party. We can see that affective po­
p is the party, i the individual respondent and likeip the like-dislike score larization for these two cases is similar to that for voters 1 and 2. Note
assigned to each party p by individual i. that the unweighted score is higher because voters view the party system
However, it may make sense to weight polarization scores by parties’ as divided into two more equally-sized camps. The weighted score is
vote shares. In addition to the substantive justification presented above, higher for voter 3 than for voter 4 because voter 3 strongly dislikes one
there is also the more practial reason that surveys often vary in how of the two large parties, creating a more equal division of affect across
many smaller parties are included in like-dislike question batteries. By parties.
weighting smaller parties less, these inevitable distortions weigh less The next two voter profiles show what happens if the most-disliked
heavily on affective polarization scores. party is small or large. Voter 5, who dislikes only a small party, has
Hence, we can also calculate affective polarization as the weighted lower weighted affective polarization than voter 6, who dislikes a larger
average party affect difference compared to each respondent’s weighted party. Their unweighted affective polarization is naturally the same.
average party affect. The precise formula for affective polarization is Note that, with the weighted measure, voter 6 has a larger value than
adapted from Alvarez and Nagler (2004) and voter 2. This makes sense when considering that voter 6 has strong
Ezrow (2007) for party system left-right dispersion (see also Curini positive affect for the largest party and strong negative affect for the
and Hino, 2012). The Weighted Affective Polarization (WAP) equation second-largest party, so has large affective distance for the key players.
for parties j and voter i is: In contrast, voter 2 is closer to a voter who simply dislikes all players
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ P apart from a smallish party.
√∑ ( )2 The final three voter profiles show scenarios where there is less af­
Spreadi = √ vp likeip − likei
p=1
fective polarization. Voter 7 has no positive affect, voter 8 no negative
affect. Both have the same affective polarization scores as the spread of
vp is the vote share of each party, measured as a proportion with a range affect in the system is the same. These voters’ level of affective polari­
from to 0 to 1. The mean affect should itself be weighted by party size zation is also clearly lower than those for the first four voters, who
and is calculated as: exhibit clear affective polarization. Voter 9 is largely indifferent be­
tween the four parties and has very low affective polarization as a result.
∑ Overall, we can see that affective polarization using the spread-of-
P
( )
likei = vp * likeip
p=1
scores measures is high when voters have strong affect for one or
more parties. Affective polarization with the weighted measure is also
Note that affective polarization is calculated for all respondents who higher if voter sympathies and antipathies are equally divided across the
declare a level of affect for at least two parties, even if more parties were party system. Conversely, the spread-of-scores measures produces low

Table 1
Example calculation of affective polarization in a four-party system.
Voter 1 2 3 4 5 6 7 8 9

Party Positive ID, Positive ID, Positive ID, large Positive ID, Dislikes Dislikes No positive No negative Indifference
large party small party and small party large parties small party large party affect affect

Party A (40%) 10 0 10 10 10 10 5 10 4
Party B (30%) 0 0 0 9 5 0 0 5 4
Party C (15%) 0 0 0 0 0 5 0 5 3
Party D (15%) 0 10 9 0 5 5 0 5 3
Spread of scores
Unweighted 4.33 4.33 4.76 4.76 3.54 3.54 2.17 2.17 0.50
Weighted 4.90 3.57 4.85 4.41 3.49 4.15 2.45 2.45 0.46
Mean distance
Unweighted 10.00 10.00 8.19 8.19 7.07 7.07 5.00 5.00 0.82
Weighted 10.00 10.00 8.67 7.11 6.61 7.91 5.00 5.00 0.71

Weighted mean 4 1.5 5.35 6.7 6.25 5.5 2 7 3.7

like-dislike

4
M. Wagner Electoral Studies xxx (xxxx) xxx

polarization for indifference or low variation. Finally, the unweighted First, Ward and Tavits. (2019) use the simple standard deviation of
score is naturally not responsive to the size of parties that are liked or like-dislike scores in their work on how affective polarization shapes
disliked, creating a less dynamic and responsive measure. perceptions of politics in multiparty systems. This measure is closely
related to the unweighted spread-of-scores measure.
4.2. Comparison measure: mean distance from the most-liked party Second, Gidron, Adams and Horne (2018) and Reiljan (2019) both
use like-dislike for partisans only, measuring the weighted mean
As noted above, a second, more restrictive way of conceiving of af­ like-dislike score for out-parties. The measure by Reiljan (2019) also
fective polarization is as the average affective distance of other parties takes into account how much a partisan likes their own party. These
from one’s most liked party. This ‘mean-distance’ measure captures how measures are closely related to the mean-distance measure, but limit
much lower on average an individual’s affect for other parties is. Hence, measurement to self-declared partisans. Potential measurement error in
this measure assumes positive identification with one party as the key the party identification question would appear to counsel against this
aspect of affect patterns. It is worth formalizing this measure and restriction. In addition, the possibility of multiple positive identities
comparing it to the spread-of-scores measure. mean that the spread-of-scores measure may be preferable for theoret­
The mean-distance measure can be calculated as: ical reasons.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Finally, Gidron et al., (2019) consider the affective distance between
∑P ( )2
p=1 likeip − likemax,i
the main centre-left and the main centre-right party. This measure is
Distancei =
np hard to relate to the two measures presented above. However, the
measure in Gidron et al. (2019) serves the useful purpose of comparing
This time, n is the number of parties without the most-liked party. non-US systems to US levels of affective polarization.
likemax is the score assigned to the most-liked party. Again, we can also
calculate a weighted version of this measure. This would then take into 5. Data and descriptives
account that it should matter more if an individual strongly dislikes
large parties rather than a small, insignificant competitor. The dataset used is the Comparative Study of Electoral Systems. This
This is calculated as: project has included like-dislike questions in all rounds since the start of
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ P the project in 1996. Modules 1 to 4 are used, along with the advance
√∑ ( )2 release of Module 5; excluding countries with a Polity IV score below 8,
Distancei = √ vp likeip − likemax,i
p=1 this leaves us with 166 surveys and 51 countries. In the first round of the
CSES surveys, voters could place up to six parties, where applicable; in
The mean-distance measure differs from the spread-of-scores mea­ the later modules, this was increased to a total of nine possible
sure in that the former captures how much an individual on average placements.4
dislikes other parties compared to his or her favoured party, while the The precise question wording for the like-dislike question is: ‘I’d like
latter simply captures the extent to which affect is spread out across the to know what you think about each of our political parties. After I read
various options. the name of a political party, please rate it on a scale from 0 to 10, where
The example in Table 1 highlights this difference. Voters 3 and 4 0 means you strongly dislike that party and 10 means that you strongly
have lower levels of affective polarization than voters 1 and 2 using the like that party. If I come to a party you haven’t heard of or you feel you
mean-distance measure, while they have similar levels using the spread- do not know enough about, just say so. The first party is PARTY A.’ This
of-scores measure. Given that voters 3 and 4 are also strongly affectively is followed by questions about the other parties. The question text has
polarized, this example shows the limitations of using a mean-distance remained the same in all four modules. ‘Don’t know’ responses are
measure. However, it is worth noting that the two measures behave treated as missing throughout this article.
very similarly if we consider scenarios where voters favour one party, so In the following, I focus on the weighted spread-of-scores measure of
the relative patterns for voters 5 to 9 are very similar for both measures. affective polarization, but compare results to the unweighted measure
In sum, the spread-of-scores measures is superior if we want a and to the weighted and unweighed mean-distance measures.5
measure that captures the empirical reality of affect patterns in multi­ First, Fig. 1 shows the average level of affective polarization for the
party systems better than the mean-distance measure. The spread-of- countries included the CSES project. If there are several CSES surveys for
scores measure better captures opposition between blocs of parties a country, respondents are pooled for this analysis. We can see that af­
rather than between single parties. The measure is also theoretically fective polarization is particular high for less wealthy, perhaps also less
better suited to voters without a single positive party identification. For stable democracies: Montenegro, Kenia and Albania, for instance.
such voters, it does not make sense to presuppose a social identity This graph also shows that, empirically, how we measure affective
constructed around parties, but they may nevertheless hold positive and polarization is less consequential than the ideal-typical scenarios shown
negative feelings towards certain parties. The spread-of-scores measure in Table 1 would indicate. Differences in the relative ordering depending
is therefore more applicable to multiparty systems that are based more on which measure we use are small. It is consistent with this finding that
around ideologies and coalitions rather than around single parties. the bivariate correlation between the different measures is always high,
However, the mean-distance measure could be the preferred at around 0.9 or above. Appendix A.1 shows that the correlation be­
approach if researchers are more interested in exclusive partisan iden­ tween the measures is also high irrespective of party system size. Dif­
tities, so the extent to which one’s preferred party is seen as the only ferences between the two measures depend much more on the distance
‘good’ party. For this purpose, it could potentially also be restricted to
those expressing a partisan identity (Reiljan 2019). Moreover, it is worth
noting that the two measures behave similarly if voters tend to only have
4
one favoured ‘in-party’ or have no ‘in-party’ at all. As a robustness check, an analysis of Module 4 results was conducted using
only the first four like-dislike scores. The correlation between these the reduced
and the longer measure was high (r = 0.97) for the weighted spread-of-scores
4.3. Comparison with other proposed measures
measure, so analysts can feasibly rely on surveys with less extensive like-
dislike batteries.
These measures are naturally related to other attempts to use like- 5
The vote shares used to weight party affect are taken from the election of
dislike scores to assess affective polarization. Here, I will briefly the CSES survey, which means they are very close approximates of the current
distinguish how these approaches relate to the two measures discussed electoral strength of each party. Lower house vote shares are taken if available;
above. if not, upper house or presidential election vote shares are used.

5
M. Wagner Electoral Studies xxx (xxxx) xxx

Fig. 1. Mean levels of affective polarization by country.

between the most-liked party and the second-most-liked party (Appen­ affective polarization if we focus on those with a party identification.
dix A.1). Hence, partisans and non-partisans in the US differ more strongly than
This Figure also shows that the United States has average levels of elsewhere their strength of affective polarization.
affective polarization in international comparison. This is surprising,
both because the US has been the overwhelming focus of research and 6. The correlates of affective polarization
because partisanship is often argued to be less intense in many other
democracies. However, these findings are consistent with those pre­ This section’s aim is to examine how affective polarization is asso­
sented in Gidron et al. (2019) and Boxell et al. (2020). However, Ap­ ciated with four other individual-level concepts: positive partisanship,
pendix A.8 shows that the United States has a higher relative level of negative partisanship, left-right extremism and perceived left-right

6
M. Wagner Electoral Studies xxx (xxxx) xxx

party polarization. whether there is a party or candidate they would never vote for.7 Fig. 2
This section asks two questions. First, I aim to show that, while af­ shows that negative partisanship is more likely for respondents with
fective polarization is correlated with all four individual-level concepts, greater levels of affective polarization, even if the association is clearly
it goes beyond these. In other words, affective polarization captures smaller than for positive partisanship. Hence, as with positive parti­
something else beyond these concepts. Second, it is important to sanship, there is a clear link between the concepts of negative parti­
examine how these correlations depend on the measurement approach. sanship and affective polarization, but the association is not
Such variation seems likely, as the mean-distance approach is more overwhelming.
linked to partisanship and the spread-of-scores approach more to per­ Fig. 2 also allows us to compare how the weighted spread-of-scores
ceptions of party competition overall. measure compares to the other proposed measures in terms of its asso­
ciation with partisanship. Given its theoretical basis in exclusive social
identities, the link between partisanship and the mean-distance
6.1. Affective polarization and partisanship
approach should be stronger than that between partisanship and the
spread-of-scores approach. The former captures whether one party is
Partisanship is likely to increase levels of affective polarization,
liked more than others, and this pattern should be related to partisan­
mainly by increasing negative feelings towards competitors. Positive
ship. Indeed, Fig. 2 shows that the mean-distance measure is somewhat
partisanship may lead citizens to perceive greater differences to other
more closely related to partisanship than the spread-of-scores measure,
parties (Medeiros and Noël, 2014; Greene, 1999; Kelly, 1989), which
though this difference is not very large.
will be seen mainly as competitors and rivals (Weisberg, 1980). The
Interestingly, the spread-of-scores measure is somewhat more closely
biased information processing and directional motivated reasoning
related to negative partisanship than the mean-distance measure. This
associated with partisanship may exacerbate such tendencies (Taber and
indicates that the spread-of-scores measure is better able to capture out-
Lodge, 2006; Greene, 1999; Campbell et al., 1960). Medeiros and Noël
group dislike as a key component of affective polarization. Again,
(2014) argue that positive partisanship is likely to come before negative
however, the differences between the measures are visible but not very
feelings about other parties, so it is plausible that dislike of opponents is
large.
driven by positive affect for one party rather than vice versa.
However, while partisanship and affective polarization are intrinsi­
cally linked concepts, they are not coterminous. For one, partisans vary 6.2. Affective polarization and left-right polarization and ideology
in the extent they dislike outgroups, as has been shown for the US
(Iyengar et al., 2012). As argued above, affective polarization can also be Affective polarization is associated with, yet distinct from, ideolog­
measured for those without a positive party identification. Indeed, in ical extremism and polarization (Rogowski and Sutherland, 2016;
many European party systems, dislike of an out-party on the radical left Lelkes, 2019; Reiljan 2019; Ward and Tavits, 2019). Here, I will examine
or right might be stronger than in-group affect with a favoured party. two phenomena: individual-level ideologial extremism and
This implies that affective polarization captures more than the individual-level perceptions of party system polarization. Again, I will
established concept of partisanship. To examine the extent to which consider the overall association with affective polarization as well as
affective polarization and partisanship capture similar phenomena, we differences between measures.
can consider differences visible in the CSES data. Partisanship is First, individual extremism summarizes people’s political beliefs,
measured using the standard question included in the CSES: ‘Do you which can range from centrist to extreme on a simplified left-right
usually think of yourself as close to any particular political party?‘.6 It spectrum. Such extremism should be positively associated with affec­
would of course have been useful to be able to use more detailed mea­ tive polarization: the more extreme a citizen is, the more he or she
sures such as that proposed by Bankert et al. (2017), but these are rarely should develop positive and negative feelings towards parties. West­
available in cross-national surveys. wood et al. (2018) find that partisans distrust those further away ideo­
To assess the association between affective polarization and parti­ logically from them the most (see also Lelkes, 2019). This association is
sanship, I run a series of models with party identification as the outcome also indicated by research on negative partisanship, which finds that
variable and each of the measures of affective polarization as the pre­ ideology is a key predictor of negative feelings towards parties
dictor. For these analyses, the measures of affective polarization were (Medeiros and Noël, 2014).
standardized to have a mean of 0 and a standard deviation of 1. These The perceived level of ideological polarization between parties
models include fixed effects for each survey (i.e. country-election in­ should also be associated with affective polarization (Ward and Tavits,
dicators) to account for time- and country-specific differences in levels 2019). Some voters think that many parties stand for more or less the
of party identification, also relating to how the question is asked in each same policies, while others detect large differences in party pro­
survey. These models aim to assess associations, not causal grammes. The more voters think parties are polarized in left-right
relationships. ideological terms, the more they will see parties as distinct choices
Fig. 2 shows that affective polarization has a clear positive effect on and the more they will perceive there to be high stakes associated with
positive partisanship. A standard-deviation increase in affective polari­ electoral outcomes (Rogowski and Sutherland, 2016).8
zation is associated with about a 15 percent increase in the probability of I measure (weighted) ideological polarization using a 0–10 left-right
having a positive party identification. However, this association, while scale and the following formula:
strong, leaves a lot of space for other factors; affective polarization and √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ P
partisanship are not tautologous concepts. √∑ ( )2
Ideological polarizationi = √ vp positionip − positioni
A similar pattern is present if we take negative partisanship, which is p=1
measured exclusively in Round 3 of the CSES. Respondents were asked

6 7
Note that (1) this does not code people who only respond ‘yes’ to the follow- The precise question is ‘And were there any [party or parties/candidate or
up question (‘Do you feel yourself a little closer to one of the political parties candidates] that you would never vote for?’
8
than the others?‘) as partisans and (2) there is considerable heterogeneity across Of course, causality may also run the other way, as affective polarization
surveys within the CSES project in terms of the precise questions measuring may affect information seeking and information processing, leading to greater
partisanship. levels of perceived ideological polarization. More simply, affect can lead to
assimilation and contrast effects in how survey respondents place political
parties (Merrill et al., 2001; Drummond, 2011).

7
M. Wagner Electoral Studies xxx (xxxx) xxx

Fig. 2. Probability of partisan ID: marginal effect of affective polarization. Note: Figure presents the average marginal effect of affective polarization. For calculation
of affective polarization measures, refer to text. Country dummies included in model. Left panel: Partisanship measured as a binary variable using the main
partisanship question in the CSES, i.e. without the follow-up ‘squeeze’ question. Analysis uses CSES Modules 1 to 5. Right panel: Negative partisanship measured as ‘Is
there a party you would never vote for’ in CSES Module 3. Full results in Appendix A.2 and A.3.

where position is the left-right position of the party. The mean left- 7. Association with democratic values and participation
right position should itself be weighted by party size and is calculated as:
The aim of this final section is also twofold. First, it assesses whether
∑
P
( )
positioni = vp * positionip affective polarization holds any explanatory power for key democratic
p=1 values and political participation that goes beyond the traditional con­
cepts of partisanship and ideological polarization. Second, this section
Fig. 3 presents the results of models where the measures of affective
assesses whether the measures differ in their explanatory power for
polarization are the predictor and ideological polarization the outcome
these outcome variables.
variable; again, survey dummies are included. The Figure shows that
As shown by Ward and Tavits (2019), affective polarization affects
there is a clear positive relationship between ideological and affective
perceptions of political choice as well as turnout and participation. This
polarization. Hence, there is unsurprisingly a strong and clear associa­
is because strong positive and negative affect towards the parties
tion between the two concepts measured by survey responses. However,
competing will increase the perception that there is a lot at stake in each
the association is not perfect. The effect of a one standard-deviation
election (Ward and Tavits, 2019). Moreover, the in-group and outgroup
chagne in each measure in Fig. 3 is less than half a standard deviation
identity-based aspects of affective polarization are likely to increase
for ideological polarization. Moreover, the correlation coefficient is just
attention to and engagement with politics (Huddy et al., 2015). Simi­
0.44 for the weighted spread-of-scores measures and 0.41 for the
larly, Abramowitz and Stone (2006) show that the polarization of
weighted mean-distance measure. So, even though these concept are
evaluations of George W. Bush was an important predictor of turnout
measured in the same survey in similar ways, they are not that strongly
and activism as it energizes voters.
correlated.
This section also assesses whether affective polarization holds
The right panel of Fig. 3 shows the association between affective
explanatory power that goes beyond two key existing variables estab­
polarization and ideological extremism, measured on a 0–5 scale as the
lished in prior research, partisanship and ideological polarization.
distance from the midpoint on the standard 11-point left-right scale. A
Existing research on partisanship indicates that it leads to greater
standard-deviation increase in affective polarization is associated with a
engagement in politics (Campbell et al., 1960), higher levels of turnout
half-unit increase in ideological extremism.
(Smets and Van Ham, 2013) and greater participation in general (Verba
Fig. 3 also indicates how the measures differ in their association with
et al., 1995; Rosenstone and Hansen, 1993). The relationship between
ideological polarization and extremism. We should expect the link be­
partisanship and satisfaction with democracy is more complex, as
tween perceived party left-right polarization and the spread-of-scores to
‘winners’ have higher levels and ‘losers’ lower levels of satisfaction
be higher, as this measure captures overall divisions within the party
(Anderson and Guillory, 1997; Anderson and Tverdova, 2001; Nadeau
system. This is indeed what the left panel of the Figure shows. The right
and Blais, 1993; Craig et al., 2006). Existing research on ideological
panel shows that all measures are similarly linked to ideological
polarization indicates that it leads to higher levels of turnout (Dalton,
extremism, with the exception of the unweighted spread-of-scores
2008; Crepaz, 1990) and greater satisfaction with democracy (Ezrow
measure.
and Xezonakis, 2011), though Rogowski (2014) finds that increased

8
M. Wagner Electoral Studies xxx (xxxx) xxx

Fig. 3. Affective polarization and Left-Right polarization. Note: Figure presents the effect of affective polarization. For calculation of affective and ideological
polarization measures, refer to text. Country dummies included in model. Left panel: Effect of standardized affective polarization measures of standardized ideo­
logical polarization measure. Right panel: Effect of standardized affective polarization measures of Left-Right extremism (range: 0–5). Analyses use CSES Modules 1
to 5. Full results in Appendix A.5 and A.4.

electoral choice leads to lower turnout within the US. Given that af­ no partisanship as the reference category. I also control for a variable
fective polarization is related to both partisanship and ideological po­ capturing each respondent’s the highest like-dislike score, so that af­
larization, we will examine the effects of all three factors below. fective polarization does not act as a proxy for simply liking a party; this
The first set of outcome variables are three measures of democratic variable can also be seen as an alternative measure of partisanship.
values. First, I examine the standard question on satisfaction with de­ Furthermore, I control for standard socio-demographic variables (age,
mocracy, which has of course been subject to significant criticism (Linde gender, education, income, religiosity and rural-urban residence). Apart
and Ekman, 2003; Canache et al., 2001). This question is measured on a from gender, all sociodemographic variables are implemented as
five-point scale. Second, I use a question asking respondents whether continuous variables.10 A more substantive control is the left-right
they believe that who they vote for makes a difference. Third, I use a extremism of a voter, measured as above on a 0–5 scale.
question asking respondents whether they believe that it makes a dif­ The models, presented in Fig. 4 and in full in Appendix A.6, show
ference who is in power. These two questions have been used in, for consistent effects for all measures of affective polarization. Each row in
instance, Vowles (2010, 2008) and Dalton and Anderson (2011).9 The the top row of the Figure shows the effect of affective polarization using
two questions are similar to questions on efficacy as they relate to the each of the four measures. Concerning satisfaction with democracy (left
belief that citizens can effect political change at the ballot box (Niemi panel), affective polarization leads to a decrease in satisfaction: the more
et al., 1991; Campbell et al., 1954). polarized people’s interparty hostility, the less they are satisfied with
I run linear regression models predicting each outcome variable. I democracy. This is likely to be related to the fact that losing is worse if
include fixed effects for each survey. Again, I run models for all measures the opponent is viewed more negatively. However, at the same time
of affective polarization. I will compare the effects for affective polari­ affective polarization is associated with believing that it makes a dif­
zation to two other key variables: ideological polarization, measured as ference who people vote for (middle panel) and who is in power (right
described above; and partisanship, included as a dummy variable, with panel).
The two bottom rows show the effects of perceived left-right polar­
ization and party identification in each of the four models, i.e. when
9
The precise questions are: ‘Some people say that it doesn’t make any dif­
ference who is in power. Others say that it makes a big difference who is in
power. Using the scale on this card, (where ONE means that it doesn’t make any
10
difference who is in power and FIVE means that it makes a big difference who is Age is coded in years; education is coded as none (1), incomplete primary
in power), where would you place yourself?’ and ‘Some people say that no (2), primary completed (3), incomplete primary (4), secondary completed (5),
matter who people vote for, it won’t make any difference to what happens. post-secondary completed (6), university undergraduate degree incomplete (7),
Others say that who people vote for can make a big difference to what happens. and university undergraduate degree completed (8); income is coded as the
Using the scale on this card, (where ONE means that voting won’t make any household income in quintiles; religiosiy is coded as: have no religious beliefs
difference to what happens and FIVE means that voting can make a big dif­ (1), not very religious (2), somewhat religious (3), very religious (4); Urban-
ference), where would you place yourself?‘. Note that the exact phrasing of the rural residence is coded as: rural area or village (1), small or middle-sized
question has varied slightly across modules. town (2), suburbs of large town or city (3), large town or city (4).

9
M. Wagner Electoral Studies xxx (xxxx) xxx

Fig. 4. Affective polarization and democratic values. Note: Figure presents the coefficients for affective polarization, ideological polarization and partisanship for
four different models, each using a different measure of affective polarization. Each row of coefficients in each panel is a different model. The dependent variables are
given at the top of the Figure. For calculation of affective and ideological polarization measures, refer to text. Full models shown in Appendix A.6. Analyses uses CSES
Modules 1 to 5.

using different measures of affective polarization. The effects of affective controls are the same as in the previous models, though I also control for
polarization are larger than those for perceived left-right polarization. In efficacy (i.e., who is in power and who people vote for makes a differ­
contrast, the effect of party identification is larger than that of affective ence), as this may impact decisions to participate politically. In the
polarization.11 models predicting participation, I also control for whether the respon­
The effects also depend somewhat on which measure is used. The dent states that he or she voted or not.
spread-of-scores measures has a lower effect on explaining satisfaction The models, presented in Fig. 5 and in full in Appendix A.7, show
with democracy. This is probably because the mean-distance measure that affective polarization is consistently associated with a higher like­
captures better whether one single party is liked by respondents. Such lihood of turning out to vote and with greater participation in politics.
respondents may more often be dissatisfied with democratic outcomes, Again, the magnitude of these effects is clearer and stronger than for left-
especially if they are on the losing side. In contrast, the spread-of-scores right polarization and smaller than for party identification.12Finally, the
measures has a clearly stronger effect on perceptions of who people vote models show that the measure of affective polarization that we choose
for and who is in power makes a difference. does not affect findings for these two outcome variables.
Next, we consider two measures of participation: self-reported
turnout and a broader measure of participation based on a battery of 8. Conclusion
questions from Round 2 of the CSES project. Here, five activities were
asked about: trying to persuade others to vote for a particular party or Applying the concept of affective polarization to multiparty systems
candidate; showing support for a particular party or candidate by, for requires researchers to adapt approaches to measurement. Moving to
example, attending a meeting, putting up a poster, or in some other way; multiparty settings means that researchers have to take into account
contacting a politician or official; taking part in a protest, march or affect across the party system. This creates measurement challenges as
demonstration; or working together with people who shared the same researchers need to decide (a) whether to use a single or multiple in-
concern. To simplify the analysis, I create an index, ranging from 0 to 5, group conceptualization of partisanship and (b) whether the size of
of the number of activities respondents said they had engaged in the past parties should matter for affective polarization.
five years or so. Both questions are based on self-reported behaviour, In this article I have argued that a weighted measure allowing for
which limits their overall validity. However, no alternatives to such multiple in-groups is best suited to understanding affective polarization
questions are available for a broad range of countries. in multiparty systems. This article has also presented measures that
Again, I run regression models with fixed effects for each survey. For allow researchers to implement their own preferred version of affective
voting, I run logistic models; for participation, I run linear models. The polarization using like-dislike scores. Ultimately, the choice of measure
depends on the theoretical standpoint of the researcher.
The second contribution of this article has been to apply these
11 measurement approaches to cross-national data from the CSES. The
Appendix A.8 presents results for partisans only. The results hold even
among partisans: affective polarization has the same type of effects for partisans empirical results show that the choice of measure is less consequential
and non-partisans. That said, the effects of affective polarization are unsur­
prisingly weaker for partisans: these voters all have a strong identity driving
responses to these questions. An interesting question for further study is
12
whether the effect of affective polarization on these attitudes depends on Appendix A.8 again shows that the effects for affective polarization also
winner-loser status in the political system (Blais and Gélineau, 2007). hold for partisans only.

10
M. Wagner Electoral Studies xxx (xxxx) xxx

Fig. 5. Affective polarization and participation Figure presents the coefficients for affective polarization, ideological polarization and partisanship for four different
models, each using a different measure of affective polarization. The dependent variables are given at the top of the Figure. For calculation of affective and ideo­
logical polarization measures, refer to text. Full models shown in Appendix A.7. Analyses uses CSES Modules 1 to 5 (left panel) and CSES Module 2 (right panel). Full
results in Appendix A.7.

than could be expected. Patterns of association do not depend a lot on useful starting points (e.g., Curini and Hino, 2012), but this research also
which measure is used. Empirically, the different measures are highly needs to take into the unique aspects of affective polarization.
correlated, mainly because voters do not have one strong in-party. A second set of questions relates to short-term changes in affective
The analysis of CSES data produces several additional important polarization. Why does affective polarization increase or decrease from
results. First, the United States, hitherto the focus of most research is not one election to another? How does affective polarization rise and fall
a country with particularly large affective polarization among all citi­ depending on context conditions? How does the emergence of radical
zens (see also Westwood et al., 2018; Gidron et al., 2019). Second, af­ competitors on the left and right influence affective polarization? Initial
fective polarization is linked to positive and negative partisanship as findings indicate that affective polarization declines outside of election
well as ideological polarization, but the findings indicate that these periods (Hernandez et al., 2019). Future analyses could provide fresh
concepts do not fully cover each other. Finally, affective polarization has insights into how to decrease affective polarization (Levendusky, 2018;
a stronger association with key measures of citizen engagement with Iyengar et al., 2019; Gidron et al., 2019).
democracy than perceptions of party system polarization. Thus, higher Finally, a cross-national study based on post-election surveys can do
levels of affective polarization are related to higher levels of turnout, little to establish causality. Researchers should examine how affective
participation and perceptions of electoral choice, but lower levels of polarization causally affects perceived levels of ideological polarization
satisfaction of democracy. Here, affective polarization is associated with and connections to the democratic system and vice versa. This research
these outcome variables in ways that add to existing explanations based should use panel data and experimental designs to test for causal di­
on ideological polarization and (positive) partisanship. rections in how these attitudes and behaviours are related.
Analyzing affective polarization cross-nationally opens up a broad
range of potential research questions Gidron et al. (2019). Future work Acknowledgements
should consider explanations for differences in affective polarization
over time and across countries. One set of questions concerns the Previous versions of this article were presented at the sixth annual
long-term origins of affective polarization. Why do some countries have European Political Science Association conference 2016, GESIS in
higher levels of affective polarization than others? Thus, we should Mannheim, and the University of Gothenburg. I received helpful com­
investigate how affective polarization is related to, for instance, existing ments and support from James Adams, Nicolai Berk, Eelco Harteveld,
social cleavages (Robison and Moskowitz, 2019; Westwood et al., 2018) Philipp Mendoza, Marian Quednau and Andres Reiljan. I would also like
or economic inequality (Gidron et al., 2018). Here, the literature on to thank the anonymous reviewers for their useful comments.
partisanship and on ideological polarization may provide additional

11
M. Wagner Electoral Studies xxx (xxxx) xxx

Appendix A. Supplementary data Huddy, Leonie, 2001. From social to political identity: a critical examination of social
identity theory. Polit. Psychol. 22 (1), 127–156.
Huddy, Leonie, Bankert, Alexa, Davies, Caitlin, 2018. Expressive versus instrumental
Supplementary data to this article can be found online at https://doi. partisanship in multiparty european systems. Polit. Psychol. 39, 173–199.
org/10.1016/j.electstud.2020.102199. Huddy, Leonie, Mason, Lilliana, Aarøe, Lene, 2015. Expressive partisanship: campaign
involvement, political emotion, and partisan identity. Am. Polit. Sci. Rev. 109 (1),
1–17.
References Iyengar, Shanto, Sood, Gaurav, Lelkes, Yphtach, 2012. Affect, not ideology: a
socialidentity perspective on polarization. Publ. Opin. Q. 76 (3), 405–431.
Abramowitz, Alan I., Stone, Walter J., 2006. The Bush effect: polarization, turnout, and Iyengar, Shanto, Westwood, Sean J., 2015. Fear and loathing across party lines: new
activism in the 2004 presidential election. Pres. Stud. Q. 36 (2), 141–154. evidence on group polarization. Am. J. Polit. Sci. 59 (3), 690–707.
Abramowitz, Alan I., Webster, Steven, 2016. The rise of negative partisanship and the Iyengar, Shanto, Lelkes, Yphtach, Levendusky, Matthew, Malhotra, Neil, Westwood, Sean
nationalization of US elections in the 21st century. Elect. Stud. 41, 12–22. J., 2019. The origins and consequences of affective polarization in the United States.
Alvarez, R Michael, Nagler, Jonathan, 2004. Party system compactness: measurement Annu. Rev. Polit. Sci. 22, 129–146.
and consequences. Polit. Anal. 12 (1), 46–62. Johnston, Richard, 2006. Party identification: unmoved mover or sum of preferences?
Anderson, Christopher J., Guillory, Christine A., 1997. Political institutions and Annu. Rev. Polit. Sci. 9, 329–351.
satisfaction with democracy: a cross-national analysis of consensus and majoritarian Kelly, Caroline, 1989. Political identity and perceived intragroup homogeneity. Br. J.
systems. Am. Polit. Sci. Rev. 91 (1), 66–81. Soc. Psychol. 28 (3), 239–250.
Anderson, Christopher J., Tverdova, Yuliya V., 2001. Winners, losers, and attitudes about Klar, Samara, Krupnikov, Yanna, Ryan, John Barry, 2018. Affective polarization or
government in contemporary democracies. Int. Polit. Sci. Rev. 22 (4), 321–338. partisan disdain? Untangling a dislike for the opposing party from a dislike of
Bankert, Alexa, Huddy, Leonie, Rosema, Martin, 2017. Measuring partisanship as a social partisanship. Publ. Opin. Q. 82 (2), 379–390.
identity in multi-party systems. Polit. Behav. 39 (1), 103–132. Lelkes, Yphtach, 2016. Mass polarization: manifestations and measurements. Publ. Opin.
Blais, André, Gélineau, François, 2007. Winning, losing and satisfaction with democracy. Q. 80 (S1), 392–410.
Polit. Stud. 55 (2), 425–441. Lelkes, Yphtach, 2019. Policy over party: comparing the effects of candidate ideology
Boxell, Levi, Gentzkow, Matthew, Jesse, M Shapiro, 2020. Cross-country trends in and party on affective polarization. Political Science Research and Methods 1–8.
affective polarization. In: Working Paper 26669 National Bureau of Economic Lelkes, Yphtach, Sood, Gaurav, Iyengar, Shanto, 2017. The hostile audience: the effect of
Research. http://www.nber.org/papers/w26669. access to broadband internet on partisan affect. Am. J. Polit. Sci. 61 (1), 5–20.
Brewer, Marilynn B., Brown, Rupert J., 1998. Intergroup Relations. McGraw-Hill. Lelkes, Yphtach, Westwood, Sean J., 2017. The limits of partisan prejudice. J. Polit. 79
Campbell, Angus, Gurin, Gerald, Miller, Warren Edward, 1954. The Voter Decides. Row, (2), 485–501.
Peterson, Evanston, Illinois. Levendusky, Matthew S., 2018. Americans, not partisans: can priming American national
Campbell, Angus, Converse, Philip, Miller, Warren, Stokes, Donald E., 1960. The identity reduce affective polarization? J. Polit. 80 (1), 59–70.
American Voter. University of Michigan Press, Ann Arbor, MI. Linde, Jonas, Ekman, Joakim, 2003. Satisfaction with democracy: a note on a frequently
Canache, Damarys, Mondak, Jeffery J., Seligson, Mitchell A., 2001. Meaning and used indicator in comparative politics. Eur. J. Polit. Res. 42 (3), 391–408.
measurement in cross-national research on satisfaction with democracy. Publ. Opin. Maggiotto, Michael A., Piereson, James E., 1977. Partisan identification and electoral
Q. 65 (4), 506–528. choice: the hostility hypothesis. Am. J. Polit. Sci. 745–767.
Carsey, Thomas M., Layman, Geoffrey C., 2006. Changing sides or changing minds?Party Marcus, George E., 1988. The structure of emotional response: 1984 presidential
identification and policy preferences in the American electorate. Am. J. Polit. Sci. 50 candidates. Am. Polit. Sci. Rev. 82 (3), 737–761.
(2), 464–477. Marcus, George E., Neuman, W Russell, MacKuen, Michael, 2000. Affective Intelligence
Craig, Stephen C., Martinez, Michael D., Gainous, Jason, Kane, James G., 2006. Winners, and Political Judgment. University of Chicago Press.
losers, and election context: voter responses to the 2000 presidential election. Polit. Mayer, Sabrina Jasmin, 2017. How negative partisanship affects voting behavior in
Res. Q. 59 (4), 579–592. Europe: evidence from an analysis of 17 European multi-party systems with
Crepaz, Markus M.L., 1990. The impact of party polarization and postmaterialism on proportional voting. Research & Politics 4 (1).
voter turnout. Eur. J. Polit. Res. 18 (2), 183–205. Medeiros, Mike, Noël, Alain, 2014. The forgotten side of partisanship: negative party
Curini, Luigi, Hino, Airo, 2012. Missing links in party-system polarization: how identification in four Anglo-American democracies. Comp. Polit. Stud. 47 (7),
institutions and voters matter. J. Polit. 74 (2), 460–473. 1022–1046.
Dalton, Russell J., 2002. The decline of party identifications. In: Dalton, Russel J., Merrill, Samuel, Grofman, Bernard, Adams, James, 2001. Assimilation and contrast
Wattenberg, Martin P. (Eds.), Parties without Partisans. Oxford University Press, effects in voter projections of party locations: evidence from Norway, France, and
pp. 19–36. the USA. Eur. J. Polit. Res. 40 (2), 199–221.
Dalton, Russell J., 2008. The quantity and the quality of party systems: party system Nadeau, Richard, Blais, André, 1993. “Accepting the election outcome: the effect of
polarization, its measurement, and its consequences. Comp. Polit. Stud. 41 (7), participation on losers’ consent. Br. J. Polit. Sci. 23 (4), 553–563.
899–920. Niemi, Richard G., Craig, Stephen C., Mattei, Franco, 1991. Measuring internal political
Dalton, Russell J., Anderson, Christopher J., 2011. Citizens, Context, and Choice: How efficacy in the 1988 national election study. Am. Polit. Sci. Rev. 85 (4), 1407–1413.
Context Shapes Citizens’ Electoral Choices. Oxford University Press. Niemi, Richard G., Jennings, M. Kent, 1991. Issues and inheritance in the formation of
Dinas, Elias, 2014a. Does choice bring loyalty? Electoral participation and the party identification. Am. J. Polit. Sci. 970–988.
development of party identification. Am. J. Polit. Sci. 58 (2), 449–465. Reiljan, Andres, 2019. Fear and loathing across party lines (also) in Europe: affective
Dinas, Elias, 2014b. Why does the apple fall far from the tree? How early political polarisation in European party systems. Eur. J. Polit. Res. 59 (2), 376–396.
socialization prompts parent-child dissimilarity. Br. J. Polit. Sci. 44 (4), 827–852. Richardson, Bradley M., 1991. European party loyalties revisited. Am. Polit. Sci. Rev. 85
Druckman, James N., Levendusky, Matthew S., 2019. What do we measure when we (3), 751–775.
measure affective polarization? Publ. Opin. Q. 83 (1), 114–122. Robison, Joshua, Moskowitz, Rachel, 2019. The group basis of partisan affective
Drummond, Andrew J., 2011. Assimilation, contrast and voter projections of parties in polarization. J. Polit. 81 (3), 1075–1079.
left-right space: does the electoral system matter? Party Polit. 17 (6), 711–743. Rogowski, Jon C., 2014. Electoral choice, ideological conflict, and political participation.
Ezrow, Lawrence, 2007. The variance matters: how party systems represent the Am. J. Polit. Sci. 58 (2), 479–494.
preferences of voters. J. Polit. 69 (1), 182–192. Rogowski, Jon C., Sutherland, Joseph L., 2016. How ideology fuels affective polarization.
Ezrow, Lawrence, Xezonakis, Georgios, 2011. “Citizen satisfaction with democracy and Polit. Behav. 38 (2), 485–508.
parties’ policy offerings. Comp. Polit. Stud. 44 (9), 1152–1178. Rose, Richard, Mishler, William, 1998. Negative and positive party identification in post-
Fiorina, Morris P., 1981. Retrospective Voting in American National Elections. Yale communist countries. Elect. Stud. 17 (2), 217–234.
University Press. Rosenstone, Steven J., Hansen, John Mark, 1993. Mobilization, Participation, and
Garry, John, 2007. “Making ‘party identification’ more versatile: operationalising the Democracy in America. Macmillan Publishing Company.
concept for the multiparty setting. Elect. Stud. 26 (2), 346–358. Shively, W Phillips, 1979. The development of party identification among adults:
Gidron, Noam, Adams, James, Horne, Will, 2018. How ideology, economics and exploration of a functional model. Am. Polit. Sci. Rev. 73 (4), 1039–1054.
institutions shape affective polarization in democratic polities. In: Annual Smets, Kaat, Van Ham, Carolien, 2013. The embarrassment of riches? A meta- analysis of
Conference of the American Political Science Association. individual-level research on voter turnout. Elect. Stud. 32 (2), 344–359.
Gidron, Noam, Adams, James, Horne, Will, 2019. Toward a comparative research agenda Taber, Charles S., Lodge, Milton, 2006. Motivated skepticism in the evaluation of
on affective polarization in mass publics. APSA Comparative Politics Newsletter 29, political beliefs. Am. J. Polit. Sci. 50 (3), 755–769.
30–36. Verba, Sidney, Lehman Schlozman, Kay, Brady, Henry E., 1995. Voice and Equality: Civic
Greene, Steven, 1999. Understanding party identification: a social identity approach. Voluntarism in American Politics. Harvard University Press.
Polit. Psychol. 20 (2), 393–403. Vowles, Jack, 2008. “Does globalization affect public perceptions of ‘Who in power can
Greene, Steven, 2004. Social identity theory and party identification. Soc. Sci. Q. 85 (1), make a difference’? Evidence from 40 countries, 1996–2006. Elect. Stud. 27 (1),
136–153. 63–76.
Helbling, Marc, Jungkunz, Sebastian, 2019. Social Divides in the Age of Globalization. Vowles, Jack, 2010. Making a difference? Public perceptions of coalition, single-party,
West European Politics, pp. 1–24. and minority governments. Elect. Stud. 29 (3), 370–380.
Hernandez, Enrique, Anduiza, Eva, Guillem Rico, 2019. Affective polarization and the
salience of elections. In: Annual Conference of the American Political Science
Association.
Hetherington, Marc J., Rudolph, Thomas J., 2015. Why Washington Won’t Work: Po-
Larization, Political Trust, and the Governing Crisis. University of Chicago Press.

12
M. Wagner Electoral Studies xxx (xxxx) xxx

Ward, Dalston G., Tavits, Margit, 2019. “How partisan affect shapes citizens’ perception Westwood, Sean J., Iyengar, Shanto, Walgrave, Stefaan, Leonisio, Rafael, Miller, Luis,
of the political world. Elect. Stud. 60, 102045. Oliver, Strijbis, 2018. The tie that divides: cross-national evidence of the primacy of
Weisberg, Herbert F., 1980. A multidimensional conceptualization of party partyism. Eur. J. Polit. Res. 57 (2), 333–354.
identification. Polit. Behav. 2 (1), 33–60.

13