Party-System Polarization and Individual Perceptions of Party Differences: Two Divergent Effects on Turnout

Abstract

In this article, we argue that party-system polarization and subjective perceptions of ideological party differences are conceptually and empirically distinct phenomena that affect electoral participation differently. Looking at 84 elections worldwide, we show that party-system polarization, and the sharp conflicts associated with it, depresses turnout because many citizens are put off by extreme party positions and unrewarding polemics. By contrast, the individual perception of differences between parties increases turnout because more citizens can find a party that is close to their own position and identify others as being further away. These opposite effects are possible because party-system polarization leads only some individuals to perceive differences between parties but leads others to avoid the emotionalized political arena. Moreover, individuals' ability to recognize differences between parties is not necessarily a consequence of party-system polarization. The contradictory findings in previous research are due to a conceptual and empirical blurring of these two essentially different aspects.

Populist and extreme (right-wing and left-wing) parties are challenging democracies around the world and mobilizing substantial parts of their countries' electorates. Often, these are new parties that take extreme ideological positions and mobilize against the so-called ‘mainstream party cartel’, which they accuse of ignoring the real interest of the people. By forcing the traditional parties to react to their challenge, the extreme parties contribute to an increasing polarization in party systems. Moreover, almost all representative democracies have experienced a substantial decline in turnout during the last several decades – the most basic and most essential form of participation in representative democracy (Franklin 2004; Lijphart 1997). In this article, we concentrate on turnout and examine the role of polarization in explaining individual electoral participation.

To date, evidence on the impact of polarization on turnout is ‘deeply ambiguous’ (Rogowski 2014: 479, 481; see also Leininger and Meijers 2021; Moral 2017: 936). As we argue and show, this ambiguity is caused by both a conceptual and an empirical blurring of two distinct mechanisms: (1) party-system polarization depresses turnout among moderate citizens and non-partisans, and (2) individual perceptions of party differences increase electoral participation among all citizens. In the literature on turnout, party polarization aids those voters who see no differences between the parties; otherwise, they would not know which party to vote for – and thus abstain (Downs 1957a; Franklin 2004). The higher the level of polarization, the easier it is for even those not so politically sophisticated citizens to recognize parties' diverse ideological standpoints and issue agendas (Ellger 2023; Lachat 2008; Levendusky 2010; Lupu 2015). As a result, they find some parties to be close to their views and others to be far away. Polarization increases choice and the individual value of the vote and is thus essentially ‘good’ for participation.

We argue that, although both narratives talk about polarization, they refer to two essentially different phenomena: party polarization as a system trait on the one hand, and polarization as individual perceptions of differences between parties on the other. We further argue that these two notions of polarization call for different modes of operationalization. At present, large parts of this rich body of literature fail to differentiate between the macro-level effects of party-system polarization and the micro-level effects of perceived party differences (for rare exceptions, see e.g. Enders and Armaly 2019; Lupu 2015; Moral 2017). As we point out, party-system polarization is only one of several, partly idiosyncratic, sources of individual perceptions of party differences. System polarization and individual perceptions of difference correlate only weakly (see the empirical findings below). This is the case because there are citizens in heavily polarized party systems who do not discern any differences between parties in their country's party system and, vice versa, citizens in centrist party systems who perceive huge differences between parties. Under the condition of system polarization, we expect low turnout because it disengages moderate citizens and non-partisans, which – as we show – represent substantial parts of the electorate in most countries. By contrast, an individual's perception of party differences will always increase their likelihood of participating in elections. Our results confirm the expected differential effects of party-system polarization and perceived party differences. System-level polarization decreases turnout, especially among moderate citizens and non-partisans, whereas perceiving differences in parties' ideological positions mobilizes citizens, irrespective of their ideological position or party identification.

The next section introduces these two stories of polarization in detail. We present our data and operationalization in the subsequent section. Thereafter we discuss the empirical findings and robustness checks and conclude with an evaluation of our results in light of the theoretical discussion.

Political polarization and its effects on turnout

Story 1: The harmful consequences of party-system polarization

This first story refers to system effects. Polarization is a system constellation with two parties or party blocs that are similarly strong in terms of vote share and thus block one another. This results in a dysfunctional government or government gridlock (Abramowitz 2013; McCarty 2015). This system tears the society apart and endangers social cohesion because the two segments are unable to communicate with each other (Hetherington 2009). In a polarized society, ‘consent over basic values is rarely achieved’ (Rapp 2016: 35). Both sides believe that they are right and that the other side is evil, corrupt or simply stupid (Abramowitz 2013; Hetherington 2009; Mason 2015). Thus, the society exists in a permanent state of political warfare, where everything is focused on winning the next election and there is no focus on governing between elections (Mason 2015; Theriault 2015). In this situation, no one can remain neutral, the centre begins to ‘disappear’ (Abramowitz 2010) and everyone is eventually partisan (Mason 2015; Stonecash 2015). The bipolarity no longer concerns single issues but rather entire issue packages that are increasingly intertwined with individuals' basic identities and value orientations (Carsey and Layman 2015; Jacoby 2014; Layman and Carsey 2002; McCarty 2015). Instead of cleavages based on economic status or class, society is torn apart by a ‘culture war’ (Abramowitz 2013, 2015; see already Hunter 1991). In this situation, politics focuses less on finding solutions for political problems, but rather on invoking social and political identities. It is dominated by two partisan tribes that aim to defeat one another. Note that bipolarity is not at all restricted to two-party systems. In the contemporary multiparty landscape of Europe, for example, polarization is driven by the advent and lasting electoral success of right-wing radical or populist parties (e.g. Bischof and Wagner 2019; Ellger 2023; Leininger and Meijers 2021; Tavits 2006).

Voter turnout will be initially high and party competition fierce in such a Clausewitzian ‘friend’ or ‘foe’ constellation because it is impossible not to take sides. But will the political community survive such a permanent conflict in the long run? For Joan-María Esteban and Debraj Ray, ‘polarization is closely linked to the generation of tensions, to the possibilities of articulated rebellion and revolt, and to the existence of social unrest in general’ (1994: 820). However, many non-partisans and ideologically moderate voters are not at all polarized.

Hence, ordinary citizens are often moderate, they cluster around the ideological centre and are turned off by polarized party positions, permanent partisan hatred and ongoing ‘negative campaigning’ (Ansolabehere and Iyengar 1997; Ansolabehere et al. 1994, 1999). Those involved in the polarized camps engage in extreme mobilizing efforts (Kleiner 2018; Rogowski 2014), which negatively affects those in between. When politics does not revolve around policies but anger and hatred between two groups, moderates and non-partisan citizens perceive political polarization not as policy or ideological polarization but instead as a noisy battle arena which is not of concern to them since they do not belong to any battle camp. Those in between no longer perceive the policy differences behind the noise because they turn their backs on politics, repelled by the constant hatred and attack campaigning. They do not feel represented and thus abstain from voting (Dassonneville and Çakır 2021; Fiorina et al. 2005; Kleiner 2018; Levendusky and Malhotra 2016; Moral 2017).

Story 2: The beneficial consequences of party polarization

This second story relates to an individual-level process. Although most authors who emphasize that polarization increases turnout use the term polarization, they do not mean polarization in terms of a bi-modal or ‘two-spike’ (Esteban and Ray 1994: 825) distribution of party systems, but rather in terms of party distribution along the ideological continuum (Sani and Sartori 1983: 329). Hence, this literature is inspired by the spatial model as put forth by Anthony Downs (1957a). Downs assumed that parties and voters can be aligned along a left–right continuum that serves as a kind of ‘super-issue’ – that is, an ideological space that encompasses all issues and provides a shortcut for parties and voters (see also Dalton 2008: 904; Fuchs and Klingemann 1990; Laver and Hunt 1992; Lipset et al. 1954; Lupu 2015: 338; Sani and Sartori 1983: 310; Van der Meer et al. 2009; Weßels and Schmitt 2008). Put simply, the spatial model predicts that if the nearest party is far away from a citizen's viewpoint, that citizen will abstain from voting. This is exactly the case for moderate citizens in a heavily polarized party system or for extreme citizens in centrist party systems (Ford 2015: 128). However, abstention and indifference are also very likely if parties are equidistant to the voter's position (Downs 1957a), or if a citizen's expected party differential is close to zero (Downs 1957b: 139) because all parties look alike.

Polarization, understood as the distinctiveness of parties concerning their ideological position, increases voter turnout (Dassonneville and Çakır 2021; Ellger 2023; Hobolt and Hoerner 2020; Key 1966; Leighley and Nagler 2014; Lupu 2015; Moral 2017; Schäfer and Debus 2018). Accordingly, heavily polarized party systems and heavily centrist party systems may similarly depress voter turnout because both ideal types ignore the ideological positions of portions of the electorate (Downs 1957a: 117ff.): ‘Polarization leaves moderate voters without a voice in politics; centrism leaves voters at the extremes without a voice, and similarly unhappy about it’ (Ford 2015: 128). In other words, electoral choice options can be both too distinct and not distinct enough, and ‘the quality of electoral democracy suffers at both ends of the scale’ (Schmitt and Freire 2012: 66). The mechanism is that citizens are able to perceive differences between parties and thus differentiate between a party close to their own position and others that are further away (Blais et al. 2014; Hobolt and Hoerner 2020; Leighley and Nagler 2014: 8). Polarization increases ‘the clarity of the cues elites send to voters’ (Levendusky 2010: 114) and thus allows voters to make more sophisticated decisions at a ‘lower cognitive cost’ (Lachat 2008: 688). Elections are ultimately about choices between candidates or parties – and choice is a perception of differences and not necessarily a perception of party-system polarization.

In this line of reasoning some scholars ascribe a corrective function to populism, since it increases ideological representation and fosters electoral participation among those population segments that are alienated from the party system (Huber and Ruth 2017; Mudde and Rovira Kaltwasser 2012; see, however, Anduiza et al. 2019; Immerzeel and Pickup 2015; Schwander et al. 2020).

Assuming that citizens' preferences are spread along the left–right continuum, an ideal matching party system would thus not be polarized but encompass parties with very diverse ideological positions, both moderate and extreme. In such a situation, turnout is high because all voters, regardless of their own ideological position, find a party close to their own standpoint and see others as being distant (Moral 2017: 941).

Is there a third story?

How can one make sense of these two stories that lead to divergent expectations regarding turnout? The remainder of this article shows that both types of polarization are conceptually and empirically distinct. Party-system polarization dampens turnout in one segment of the electorate and increases electoral participation in another. Moderates and non-partisans tend to abstain because they only find parties that are far away from their own standpoint (Dassonneville and Çakır 2021). By contrast, extreme and partisan citizens are incentivized to cast their vote since they find parties both close to their own views and further away from other parties. This is the case because partisans and extreme voters behave as engaged players in their respective party teams (e.g. Mason 2015). Those who are not part of the competing teams are taken aback by negative campaigning, noise and hatred that characterize party-system polarization. This constant noise obscures the parties' problem-solving strategies for those citizens who are already not strongly involved; those citizens resemble neutral observers who came to enjoy the match and are shocked by the hate choirs and pyrotechnics in the fan blocks. They resolve not to return to the stadium.

Why do we believe that Story 2 is also correct? This is the case because the level of party-system polarization is only one of many different factors that contribute to an individual's perception of differences between parties. However, these perceptions influence behaviour regardless of whether they are rooted in party-system polarization or another source (Adams et al. 2011; Dalton 2008: 909; Enders and Armaly 2019: 816; Kenyon 2014). We do not even know whether citizens register party positions (Lupu 2015: 334). Individual perceptions of party differences might as well originate from perceptions about candidates' images, their impressions of the campaigns, media consumption, or the amount of attention they pay to politics. They might be a result of discussions among the family, peers, co-workers or members of organizations with which a person is associated. They might also be filtered by partisan identity, projections of disgust or preference for specific parties, conflicts about single issues that are important to a voter, or based on simple misperceptions. They may even be the result of more or less polarized party constellations during a person's formative years, or simply correlate with a person's age, educational background, level of political sophistication or socioeconomic status. What matters is whether or not individuals subjectively perceive differences between the choice set. Citizens who perceive clear differences should participate in elections – regardless of their ideological position – because one party will be closer to their views than other parties.

Thus, we agree with Mert Moral (2017: 942) that individual perceptions of party differences will increase turnout, but we disagree with his expectation about the positive impact of party-system polarization. Based on Story 1, we expect that the noise, negativism and the spirit of non-compromise related to party-system polarization leads instead to lower turnout rates – particularly if an electorate has many non-partisans and moderate citizens. In short, while we argue that party-system polarization depresses turnout among non-partisans and moderates, many citizens may be unaware of party-system polarization. Or, they overwrite the ‘noise’ of hostile party competition with their own idiosyncratic perceptions. In any given electorate both processes can be at work at the same time. Thus, we contend that many citizens are not attentive to party competition but base their perception of party differences on the manifold options spelled out above. As Moral says, ‘citizens’ perceptions of and responses to party polarization are hardly uniform’ (2017: 936).

Thus, we conclude that both stories about the impact of polarization are correct. Too often, however, they are conceptually blurred, which leads to the ambiguity that characterizes the present state of ‘polarization’ research (Moral 2017; Rogowski 2014). It is no coincidence, then, that analyses basing the polarization measurement on subjective perceptions show a mobilizing effect, while analyses based on more ‘objective’ measures conclude that polarization depresses turnout (Moral 2017: 938). Strictly speaking, only system polarization – that is, a bipolar and uncompromising party-system constellation – is a clear indication of polarization. On the individual level, by contrast, polarization refers to citizens' ability to perceive differences between the competing parties. Therefore, we contend that the term polarization is misleading for describing this micro-level process. Naturally, party-system polarization might be one arrangement that enables individuals to sort their parties along the ideological spectrum. However, as we argued above, it is only one of many possible origins for individuals' perceptions of party differences. Consequently, party-system polarization and individual perceptions of party differences call for different modes of operationalization. The next section addresses this point in greater detail.

From this discussion on party-system polarization and individually perceived party differences, we derive the following hypotheses. The first two concern the divergent main effects:

Hypothesis 1: As party-system polarization increases, the level of individual turnout decreases.

Hypothesis 2: As individuals perceive more and greater differences between parties, individual turnout increases.

These divergent effects are caused by two independent mechanisms, as spelled out in the following hypotheses:

Hypothesis 3: Individual perceptions of party differences increase turnout among all voters, which includes moderate and more extreme citizens, partisans and non-partisans alike.

Hypothesis 4: Party-system polarization decreases turnout among moderate citizens and non-partisans and increases turnout of more extreme voters and partisans.

Data and operationalization

We use data from the Comparative Study of Electoral Systems (CSES), which is a collaborative research programme of election study teams from around the world. Participating countries include a common module of survey questions in their post-election studies (see http://www.cses.org/). Specifically, this article uses data from waves 3 and 4 – that is, data collected between 2006 and 2011 (Module 3), and 2011 and 2016 (Module 4).¹ Single case studies have dominated past research, most of which focus on the USA (e.g. Abramowitz 2010; Abramowitz and Stone 2006; Enders and Armaly 2019; Fiorina et al. 2005; Hetherington 2008, 2009; Levendusky and Malhotra 2016; Rogowski 2014; Westfall et al. 2015; for other single case studies, see, e.g. Ford 2015; Munzert and Bauer 2013; Wolf and Roßteutscher 2013). Some studies use aggregate data (Dalton 2008; Siaroff and Merer 2002; Steiner and Martin 2012) and the few comparative studies based on individual data are limited in scope, as they focus exclusively on (Western) European countries (Kleiner 2016, 2018; Lachat 2008; Moral 2017). Hence, this is one of the first truly comparative accounts of the impact of polarization on voter turnout worldwide (most recently Dassonneville and Çakır 2021; see also Lupu 2015 or Ezrow et al. 2014, but both with a focus on partisanship). The original data set contains information on 144,381 respondents from 95 election studies in 47 countries; 11 additional studies were excluded as their questionnaires lacked decisive questions.² Belarus (2008) was also dropped because most candidates were not affiliated with any party. On the individual level, we excluded respondents with missing values on the dependent or core predictor variables. As a result, the analyses are based on 88,397 individuals from 84 election studies in 43 countries (see Table A1 in the Supplementary Material for final case numbers per country and election year included in this study).³

The dependent variable is a dummy that differentiates between respondents who self-reported either participating in (1) or abstaining from (0) the election. Our main predictor variables are the two measures of polarization: party-system polarization and individual perceptions of party differences. There are several options for measuring polarization.⁴ We decided in favour of the formula developed by Russell Dalton (2008; similar Ezrow 2008; Lupu 2015). First, it integrates parties' positions on the left–right continuum and accounts for party size (see also Esteban and Ray 1994; Van der Meer et al. 2009). We believe that it makes a difference in terms of the degree of polarization whether parties at the extreme poles of the ideological spectrum are tiny or large (Alvarez and Nagler 2004; Dassonneville and Çakır 2021; Ezrow and Xezonakis 2011). Vice versa, the size of centre parties is also crucial. If centre parties encompass a majority of the popular vote, party systems are less polarized even if there are niche parties at the extremes (see also Lupu 2015; Moral 2017). By contrast, if extreme parties are large and centre parties are negligible in terms of voter support, a party system is heavily polarized. Second, unlike measures based on the mean absolute differences between parties, this model does not suffer from biases originating from differences in the absolute number of parties (Schmitt and Franzmann 2017). Finally, it accounts for parties in the ideological centre of a party system, in contrast to measures based on the ideological range of a party system or the mere number of – subjectively defined – extreme parties (e.g. Abedi 2002; Crepaz 1990).⁵ Among the marginally different calculation formulas available that acknowledge dispersion and size, all produce highly similar results (see Moral 2017 for a comparison). The formula thus is as follows:⁶

$$\eqalign{{\rm Party\ System\ Polarization} = & {\rm \;SQRT} \bigg \{{\sum ( {{\rm party\ vote\ shar}{\rm e}_i} ) } \cr & \quad\times {( [ {{\rm party}\;L/R\;{\rm scor}{\rm e}_i} } \cr & \quad{{ { {- {\rm \;weighted\ party\ system\ average}\;L/R\;{\rm score}} ] /5} ) }^2} \bigg \} , \; } $$

where i represents individual parties.

The measure of individual perceptions of party differentiation builds on individual perceptions of party positions. All respondents were asked to position their country's parties on the left–right continuum. We thus calculate an individual score of perceived party differences for each respondent. Although there are more dimensions that organize party competition,⁷ the left–right measure is still the most general measure of party conflict that encompasses, at least partially, other potentially important divides (see, e.g. Freire 2008; Fuchs and Klingemann 1990; Huber 1989; Knutsen 1995; Roßteutscher and Scherer 2013). Moreover, it is the only dimension for which expert and individual party ratings are available. Respondents could not always position every party, and there were particularly high rates of missing values for the smaller parties. Hence, we consider only parties with a vote share of 4% or more.⁸ Note that we decided against imputing mean values because we assume that parties that individuals cannot rate are also not part of their choice set.⁹ All respondents with only one or no single party estimate were dropped from the analyses. Since there is no measure of perceived party size available, we used the relative vote shares of the positioned parties. As CSES surveys are conducted directly after an election, we assume that respondents possess a fairly correct idea about the relative strength of their national parties. That said, using an index formula with unweighted party positions yielded identical results (see section on robustness below).¹⁰ Hence our findings are independent of whether this assumption holds or not.

As we argue above, party-system polarization is essentially different from individual perceptions of party differences. Due to a lack of independent sources, scholars often use the aggregated individual perceptions as a measure of system polarization.¹¹ We believe this is a poor idea because individuals draw on different, hardly comparable, sources to estimate their parties (Lupu 2015: 338; Rogowski 2014: 481). Country means only reflect a valid measure of system polarization if most individuals share a minimal understanding of party positions. If, however, some believe that party differences are large, while others think there are no differences, the mean value hardly reflects reality. Such differences emerge naturally if individuals' perceptions of party differences are based on the very diverse and idiosyncratic sources spelled out above. Moreover, if the perception of party differences is systematically related to other theoretically relevant variables, aggregation may simply produce a distorted measure. This is exactly the case here. Individual perceptions of party differences correlate positively with extreme positions on the ideological scale (Pearson's r = 0.27, p < 0.001) and partisanship (strong partisanship r = 0.15, p < 0.001, weak partisanship r = 0.06, p < 0.001; no partisanship r = −0.16, p < 0.001).¹² In short, country means are highly dependent on the distribution of these characteristics in the respective sample.¹³

Therefore, to arrive at a measure of party-system polarization, we refer to the expert ratings obtained by CSES (see also Lupu 2015). The experts include the members of the consortium or the academic project team responsible for data collection in a given country, who, like respondents, were asked to position their country's political parties on the left–right continuum. These experts, all of them scholars of political science and electoral research, are politically highly sophisticated and, we assume, closely familiar with the state of their respective party system. Hence, the expert ratings are used as an exogenous measure of party-system polarization – that is, all respondents in a given country/election year are coded with an identical party-system polarization score based on the country's expert ratings.¹⁴

Applying the Dalton formula to both, we now have an expert measure for party-system polarization and an individual measure for the subjective perception of party differences (for distributions see Figures A3 and A4 in the Supplementary Material). Hypothetically, the two measures stretch from 0 (all parties on the exact same spot of the left–right continuum) to 10 (two identically large parties positioned at the extremes).

With regard to our core independent variables on the individual level, we distinguish moderate voters (self-assessed 4, 5, 6 on the 0 to 10 left–right scale) from extreme voters (0–3 and 7–10).¹⁵ We further differentiate between non-partisans, those who weakly identify with a party (‘leaners’), and those who strongly identify with a party, by constructing three corresponding dummy variables. Based on the rich literature on turnout (e.g. Abendschön and Roßteutscher 2015; Armingeon and Schädel 2015; Franklin 2004; Gallego 2015; Leighley and Nagler 2014; Lijphart 1997; Schäfer et al. 2020), we include the following variables as controls because we know that they also explain electoral participation: we use educational levels as a proxy for political sophistication, since no better suited variables are available.¹⁶ We distinguish those who have completed higher education (upper secondary or higher), coded as 1, from those with less than upper secondary education (0). We differentiate first-time voters (depending on voting age regulations, i.e. roughly 21 or younger), from more experienced voters (coded 0). Further, we include age in years, gender (1 for male, 0 for women), current employment status (1 for unemployed, 0 for not unemployed), satisfaction with democracy measured on a scale ranging from 1 ‘not at all satisfied’ to 4 ‘very satisfied’ and individual self-placement on the left–right scale (0 to 10). All continuous variables are z-transformed.

On the country level, we control for the share of moderate and non-partisan citizens – that is, the relative share of the electorate in a given country that self-assesses as a 4, 5, or 6 on the left–right scale and the relative share that claims to be unaffiliated to a party, respectively (for distributions see Figures A8 and A9 in the Supplementary Material). As further control variables, we include election year and, from the World Bank, the general unemployment rate and the GDP per capita one year before the election. Democratic regime age is extracted from the Polity IV project to account for the possibility that long-standing democracies have higher (or lower) turnout. To control for party-system fragmentation, we include the Corrected Effective Number of Electoral Parties (Taagepera 1997). As voting is compulsory in some of the countries, we include a binary variable for enforced compulsory voting schemes. Finally, to account for the fact that survey data are heavily biased with regard to turnout because of over-reporting and selection bias in particular (Sciarini and Goldberg 2016), we include a macro-level variable that represents the difference between aggregated turnout based on survey responses and the officially registered turnout.¹⁷ All continuous variables were once again z-transformed. To account for the multilevel data structure and the nature of the dichotomous dependent variable, we subsequently conduct multilevel logistic regressions with two levels – the individual and the election level. Although our main interest is in the hypothetically divergent impact of party-system polarization and the subjectively perceived party differences (H1 and H2), our further hypotheses call for interaction terms: between individually perceived party differences and moderate/non-partisan positions (H3) and party-system polarization and moderate/non-partisan positions (H4). Table A2 in the Supplementary Material describes all of the variables involved.

Empirical findings

We claim that party-system polarization and individual perceptions of party differences are conceptually and empirically distinct. The correlation between political polarization based on expert ratings and individual (not aggregated) perceptions of party differences is very low (Pearson's r = 0.17, p < 0.001, see also Figures A3–A5 in the Supplementary Material for the correlation and univariate distributions).¹⁸ Hence, operationalizing polarization by aggregating country means reveals very distinct results compared to expert judgements. Thus, aggregating individual perceptions is a suboptimal measure of system polarization. We subsequently use the expert ratings as an exogenous measure of party-system polarization.

Now we test our hypotheses with multivariate analyses. Table A7 (in the Supplementary Material) presents the full model estimates. First, we present a main effects model and then test our hypotheses based on interaction terms. As for the main effects hypotheses (H1 and H2), both are statistically and substantially significant. Party-system polarization depresses turnout (odds ratio of 0.78**), but an individual's perception of party differences increases the likelihood of voting (odds ratio of 1.21***). Figure 1 presents the probabilities of individual turnout (a) under the condition of party-system polarization and (b) under the condition of individually perceived party differences.

Figure 1. Direct Turnout Effect of Party System Polarization (Left) and Individuals' Perceptions of Party Differences (Right) with 95% Confidence Intervals

Note: x-axis scale units denote changes in standard deviation from mean.

In both instances, the effect size is not trivial and, as assumed, it runs in different directions. In the case of low party-system polarization (left panel), turnout rates are more than 90%, compared to a maximally polarized system where turnout is roughly 12 percentage points lower.¹⁹ In terms of individual perceptions of party differences, the probability of turnout changes by about 10%, depending on an individual's ability to perceive clear party differences. Hence, system polarization relates to lower turnout and individual difference perceptions correlate with higher participation.²⁰ Other variables in the model (see M1 in Table A7) reveal that moderate citizens have a lower turnout rate than extreme citizens (see also Dassonneville and Çakır 2021) – and there is no difference between a left-leaning and a right-leaning respondent (for similar findings, see Van der Meer et al. 2009: 1439). We also find that partisans vote at much higher rates than non-partisans. Moreover, as an electorate's share of moderate citizens increases, so too does abstention; as the share of non-partisans increases, so too does the turnout.²¹ As for our controls, evidence confirms the results of turnout research. Education increases turnout and newly enfranchised cohorts tend to abstain. Older citizens vote at higher rates, while unemployment decreases participation, and there is no gender gap. However, satisfaction with democracy exerts clear positive effects. Looking at macro-level controls, neither regime age, GDP per capita, unemployment rate, turnout over-reporting nor survey year impact electoral participation. Compulsory voting exerts a strong positive effect and the corrected effective number of electoral parties a negative effect.

In H3, we expected that difference perceptions would increase turnout among all voters – moderate and extreme voters, partisans and non-partisans alike. Figure 2 displays the interaction between difference perceptions moderated by individual ideological position and partisanship (see Table A7, Models M2 and M3). Looking at individuals' ideological positions (left panel in Figure 2), extreme citizens always show a slightly higher turnout rate than moderates. Note, however, that confidence intervals overlap. Hence, this difference is not statistically significant. However, in both groups, turnout increases continuously with higher difference perceptions. The change in turnout probability amounts to roughly 10 percentage points, comparing citizens with the lowest and the highest difference perceptions. Turning to partisanship (right panel in Figure 2), non-partisans benefit more from increasing difference perceptions (an increase in turnout of about 11 percentage points) than partisans. Note, however, that partisans, particularly those who identify strongly, turn out at very high rates either way. Even those who perceive no differences whatsoever have a turnout rate close to 90%.

Figure 2. The Effect of Perceived Party Differences by Moderate/Extreme Position (Left) and Different Levels of Partisanship (Right) with 95% Confidence Intervals

Note: x-axis scale units denote changes in standard deviation from mean.

Identifying with a party is a sufficient motivation for voting (for similar findings, see Lachat 2008: 694), and because of ceiling effects, difference perceptions can only moderately increase partisans' turnout rates. In summary, increases in perceptions of differences between parties increase turnout among all citizens alike, as predicted in H3.

Subsequently, we test our hypotheses concerning the moderating effects of party-system polarization. H4 assumes that system polarization decreases turnout among moderates and non-partisans (compared to extreme citizens and partisans). Figure 3 visualizes the interaction effects from the multilevel logistic regression (see models M4 and M5 in Table A7). Beginning with ideological positions (left panel), both moderates and extremists have the highest predicted turnout (above 90%) when system polarization is low. Turnout decreases steadily with increasing levels of polarization. Despite being true for both groups, the turnout decline is stronger among moderates. Turnout declines among moderates by 15 percentage points when comparing a party system with the lowest polarization level to one with the highest polarization level. Among extreme citizens, the decrease amounts to roughly 10 percentage points. The difference between both groups is statistically significant, yet confidence intervals overlap (see model M4 in Table A7). Party-system polarization suppresses turnout in both groups, but to a much lower degree among extremists than moderates.

Figure 3. The Effect of Party System Polarization by Moderate/Extreme Position (Left) and Levels of Partisanship (Right) with 95% Confidence Intervals

Note: x-axis scale units denote changes in standard deviation from mean.

A different picture emerges for partisanship (right panel). Strong partisans remain unaffected by system polarization. In every constellation, their turnout is clearly above 90%. By contrast, persons with weak party identification react to increasing system polarization with declining turnout. However, the decline is modest (clearly above 90% to slightly below 90%) compared to non-partisans. When party-system polarization is low, non-partisans vote at a rate of 88% – that is, with only slightly lower levels than both partisan groups. From low to high levels of system polarization, their turnout rate decreases by about ten percentage points. Hence, hostile party environments lead non-partisans to disengage, while partisans are unaffected by centrist party systems. Although Figure 3 tells a rather clear story, this effect is statistically insignificant (model M5 in Table A7), hence we cannot be certain that our hypothesis is confirmed.

Robustness

To verify whether our results are robust, we tested alternatives concerning case selection, variable construction and model estimation. To address case selection, we reran the models with five specifications: (1) excluding systems with compulsory voting schemes²² (for the results see Table A10 in the Supplementary Material); (2) excluding non-OECD countries (see Table A11); (3) excluding two-party systems (see Table A12); (4) excluding party-system polarization outliers – that is, with index values lower than 2 and higher than 6²³ (see Table A13) – and (5) excluding countries that Freedom House rates either non-free or partially free (i.e. with a score above 2.5)²⁴ (see Table A14). On the variable level, we model seven variants: (1) we include parties with a vote share of above 2% (instead of 4%) (see Table A15); (2) we impute means for missing individuals' party ratings (compared to treating them as missing values) (see Table A16); (3) we operationalize ‘moderateness’ on a continuum (instead of the dummy variable) (see Table A17); (4) we use polarization indices unweighted by party size (see Table A18); (5) we use the ideological distance between the most extreme parties as polarization indices (see Table A19); (6) we use aggregated perceived party differences as a party-system polarization measure (see Table A20); and (7) we use our party-system polarization measure but control for aggregated perceived party differences (see Table A21). Regarding model estimation, we rerun the analyses using election-fixed effects (see Table A22) and country-fixed effects (see Table A23). Table 1 summarizes the results.

Table 1. Robustness Checks

Note: x indicates the confirmation of the hypothesis. ⁺Indicates p < 0.1 significance level.

Regarding the main effects of system polarization and individual difference perceptions (H1 and H2), which are this article's focus, every possible variant results in identical findings: strongly depressing effects of system polarization and strong positive effects of perceived party differences. Hence, our core hypotheses concerning the divergent impact of those two conceptually distinct aspects are highly robust. The same applies to our assumption regarding the participatory impact of perceived party differences among all groups of an electorate (H3), although the effect is significantly smaller for strong partisans than for non-partisans in some instances. Regarding the cross-level interaction between system polarization and moderate and non-partisan citizens (H4), robustness checks support our results. The significant effect of moderateness holds in nine out of fourteen model specifications; the non-significant effect of partisanship repeats in all but one of the models. Thus, H4 is confirmed for moderateness but rejected for partisanship.²⁵

Conclusion

This article aims to show that party-system polarization, on the one hand, and individuals' perceptions of party differences, on the other, are two conceptually and empirically distinct phenomena that impact individual turnout differently. Most citizens, particularly if they are ideologically moderate and non-partisan, are disengaged by hostile party positions, negative campaigning and party elites who prohibit compromise. Hence, under party-system polarization, turnout is relatively low. By contrast, if individuals perceive differences between parties, the likelihood is great that, among these parties, one is closer to their views than the others. In this constellation, electoral outcomes are relevant to an individual and, hence, turnout is relatively high. Perceptions of party differences may be caused by party-system polarization. However, many other reasons exist as to why some individuals perceive differences and others do not – ranging from socialization heritage, candidates' images, specific issues and private networks to simple misconceptions. As we demonstrated, the perception of differences between parties increases turnout among all citizens, including ideological moderates and extremists, as well as partisans and non-partisans. However, party-system polarization exerts depressing effects on electoral participation, particularly among moderates and non-partisans. By contrast, more extreme citizens and especially partisans are unaffected by system polarization; they participate in elections regardless – even in centrist party systems (for similar findings, see, e.g. Putnam 2000: 342; Van der Meer et al. 2009: 1430). Extreme positions and, in particular, strong partisanship are sufficient motives for participation (Lachat 2008).

That said, our findings are based on cross-sectional data and, hence, we cannot capture dynamic aspects of polarization. For instance, the results do not reveal whether partisans have a temporal limit of tolerance. How long do they endure a party system that neglects their extreme and partisan attitudes? We also do not know whether moderates in heavily polarized systems tend to develop more extreme positions and, in the longer run, join the competing camps, as suggested by Alan Abramowitz (2010), Jeff Stonecash (2015) or Lilliana Mason (2015). This limitation also implies important methodological issues as there is a possible non-equivalence between cross-sectional and longitudinal relationships and concerns about causality and omitted variable bias (see, for example, Fairbrother 2014; Schmidt-Catran and Fairbrother 2016). Lacking comparative panel data, however, we conclude, for the time being, that unpolarized party systems provide the best conditions for high turnout. Such systems coincide with the demand of moderates and non-partisans (who represent large fractions of the electorate) and deter neither extremists nor partisans from voting. Future research should address whether centrist party systems with high shares of extreme and partisan citizens lead to changing party positions or the emergence of new parties at the ideological fringes in the longer run.

To end on a more methodological note, constructing a metric for party-system polarization from the aggregated means of individually perceived ideological differences between the parties in a country is inferior to exogenous measures and may produce misleading results that contribute to the unsatisfactory state of current polarization research. This is the case because the perceptions do not mirror party-system polarization, but rather individuals' impressions of whether there are parties close to their views and others further away. As we argue above, such individual perceptions may be due to several factors, of which party-system polarization is only one. Hence, aggregated means represent the average perceptions of differences (whatever the source) – and not of system polarization. Although it was beyond the scope of this article, future studies will have to examine the sources of individual perceptions of party differences.