Are radical right and radical left voters direct democrats? Explaining differences in referendum support between radical and moderate voters in Europe Open Access

Referendums are increasingly popular in established democracies (Qvortrup 2017). According to participatory democrats, direct involvement in political decisions leads to more responsive governments, better informed citizens, and vigorous public debate (Matsusaka 2004, Pateman 1970, Roberts 2004). Modernization theories claim that due to advances in education and communication technology, citizens in established democracies are better-equipped to participate in politics, and therefore, want to decide on specific issues, beyond merely electing representatives. Other theories argue that some citizens, particularly those from disadvantaged backgrounds, increasingly feel their interests are not represented by political elites, and therefore, also want to decide on specific issues (Dalton et al. 2001).

However, referendums recently held in Switzerland, Hungary, the Netherlands, and the United Kingdom, have raised concerns about the damaging effects of direct democracy. Newspaper headlines include ‘referendums break democracies so best to avoid them’ (Harford 2018) and ‘referendums are problematic yet more popular than ever’ (Henley 2016). One concern is that these instruments provide a medium for semi-authoritarian, populist forces seeking to bypass legislatures (Qvortrup 2017). For example, Mudde notes that many Radical Right (RR) parties in Europe, such as the Dutch Party for Freedom, the French National Front, and Alternative for Germany, have called for the introduction or increased use of referendums (2007: 152), but is this desire for direct democracy also expressed by their voters?

Studies investigating a link between RR voting and referendum support have yielded mixed results (Bowler et al. 2017, Pauwels 2014, Rooduijn 2017) and few studies have investigated whether these instruments are also appealing to Radical Left (RL) voters, despite certain similarities between these voter groups. Using data from the 2012 European Social Survey and the 2017 Polpart survey, we investigate whether RR and RL voters are more favorable towards the use of referendums than moderate voters. Such a finding has strong implications for democracy: it indicates that direct citizen participation is most appealing to voters often expressing dogmatic and exclusionary preferences and suggests potential for radical populist leaders to continue mobilizing support for referendums targeting immigrants or financial and political elites.

Building on this research question, we first develop a cross-national hypothesis testing whether the association between voting for a RR or RL party and referendum support depends on the electoral success of these parties. Referendums provide political outsiders with a means of bypassing legislatures (Bowler et al. 2002), suggesting these instruments are more appealing to parties occupying a marginal role in politics. However, in some countries radical parties have also become mainstream parties. Second, we try to explain at the individual-level why RR and RL voters are more favorable towards referendums than moderate voters. The most obvious explanation is that these voters are dissatisfied with representative institutions and political elites. However, referendums also enable them to influence mainstream politics with regards to their respective political agendas. Whereas RR voters may seek more control over political decisions to promote xenophobic goals, RL voters may seek more control to push for redistributive policies.

Rydgren (2018) defines the RR as parties which share ethnonationalist xenophobia, anti-establishment populism, or sociocultural authoritarianism. However, these parties are increasing being rebranded as populist RR, owing to their condemnation of political elites and their promise to return power to the people (Akkerman et al. 2016, Mudde 2007, Pauwels 2014, Pirro 2015, Rooduijn 2017). In the literature, there is a growing consensus around Mudde’s definition of populism as:

Populist parties defend the general will as the most important aspect of democracy, superseding human rights or constitutional guarantees (Mudde 2007:23). Corrupted elites are perceived as having strayed from the general will by implementing policies that are not in accordance with what most citizens want. Referendums provide a means of undermining elite decision-making and realigning politics with the interests of the majority. The emphasis on popular sovereignty, inherent in the populism of RR parties, is what links these parties and their voters to a preference for direct democracy. Given that almost all RR parties are referred to as populist (Mudde 2007), we hypothesize that RR voters express greater support for referendums than moderate voters (H1a). By moderate voters we refer to all voters not supporting a RR or RL party.

Based on survey data from Canada, Australia, and New Zealand, Bowler et al. (2017) conclude that right-wing populist voters are not direct democrats. By contrast, Pauwels (2014) demonstrates that in Germany, Belgium, and the Netherlands, dissatisfaction with politics, RR voting, and referendum support go hand-in-hand, suggesting that this association is specific to Europe. Rooduijn (2017) finds that the voter bases of 15 European populist parties, whether left or right, do not always consist of individuals with a preference for direct democracy.

The RL party family consists of conservative and reform communists as well as populist socialists. These parties are united by their struggle against economic inequality, through the promotion of collective ownership, economic planning, or redistribution and their desire for an alternative economic and political system (Fagerholm 2016). Despite their continued presence in European democracies, RL parties have not received as much scholarly attention as their right-wing counterparts (Ramiro 2016, Visser et al. 2014). Scholars have recently started taking an interest in left-wing variants of populism in Europe, exemplified by parties such as Syriza (Greece), Podemos (Spain), Die Linke (Germany) and the Dutch Socialist Party (Kioupkiolis and Katsambekis 2018, March 2007, 2009, Rooduijn 2017). In the past, RL parties were characterized by their emphasis on the role of the revolutionary vanguard in securing change for the proletariat, prescribed in the ideology of Marxism-Leninism (March 2007: 74). The collapse of communism and the rightwards movement of social liberal parties, particularly in the aftermath of the financial crisis, has prompted a phase of reconstruction among RL forces. Some of these parties have started abandoning their democratic centralism in favor of a more bottom-up approach to political organization (Rooduijn and Akkerman 2017: 196). According to March, the contemporary RL combines its critique of capitalism with aspirations for direct democracy and/or local participatory democracy (2009: 126). Although the extent to which contemporary RL forces can be considered populist remains subject to investigation, some studies demonstrate that RL parties (Rooduijn and Akkerman 2017) and their voters (Akkerman et al.2017) are just as populist as their right-wing counterparts. Based on the ideological transformation these parties are undergoing, we hypothesize that RL voters are more favorable towards the use of referendums than moderate voters (H1b).

Research on direct democracy and its consequences for political parties maintains that referendums and initiatives are particularly conducive towards the emergence of smaller opposition parties. Direct democracy enables parties with less chance of winning seats in Parliament to remain influential, either by threatening governing parties with a referendum, or by using referendums to keep their issues on the political agenda (Ladner and Brändle 1999). In most countries except Switzerland, Greece, Hungary, and Poland, RR and RL challengers are peripheral forces critical of the monopoly held by mainstream parties, which is perceived as contributing to their exclusion from the party system (Taggart 2000). However, once these parties gain a bigger following, and their chances for representation in legislative institutions are improved, the drive to undermine existing political configurations may subside. In a study conducted in Canada, New Zealand, and the United States, Bower et al. (2002) demonstrate that legislators with more access to power are less sympathetic to direct democracy than those with limited access. Accordingly, voters whose interests are well represented in parliament may see little need for instruments that circumvent parliamentary procedures. Therefore, we hypothesize that referendum support among RR and RL voters is weaker in countries where these parties are more successful in elections (H2).

The following section argues that referendum support among RR and RL voters is an expression of distrust in representative institutions and anti-elitism, which are treated as separate motivations. However, these voters may also pursue direct democracy as a strategy for realigning policymaking with their respective policy preferences. Therefore, it is also argued that referendum support among RR voters is partly explained by anti-immigrant attitudes, whereas referendum support among RL voters is partly explained by a preference for income redistribution.

Wide-sweeping societal developments such as globalization, regional integration, and mass immigration have carved a new political cleavage between the ‘winners’ and the ‘losers’ of economic and cultural transformations (Kriesi et al. 2008). For example, the enlargement of the European Union has opened the borders to competition from economic migrants and foreign companies. RR and RL parties claim that representative institutions have failed to defend the interests of ordinary citizens who have been adversely affected by these developments.

One of the key factors contributing to the rise of radical forces is party convergence, which means that political parties no longer offer clear alternatives to one another (Kitschelt and McGann 1997). Whereas the RR denounces conservatives for having gone soft on immigration the RL accuses social democrats of fraternizing with neoliberalism (March 2007). Another factor contributing to the rise of these parties is cartelization. Through inter-party cooperation and accommodation, established parties consolidate their hold on government while excluding newcomers to the party system (Katz and Mair 1995). By providing a direct link between citizens and policy-making, referendums enable radical parties and their voters to break up existing majorities (Bowler et al. 2002, Ladner and Brändle 1999). Therefore, we argue that referendum support among both RR and RL voters is partly explained by distrust in representative institutions (H3). We emphasize that this relationship is only partly explained because we expect that other explanatory factors also play a role.

According to populist theories, one of the defining characteristics of radical parties is their anti-elitism (Mudde 2007, Taggart 2000). Populism’s central message is that citizens have been excluded from power by politicians that are selfish, unresponsive, and corrupt (Canovan 2002). Politicians are blamed for the growing divide between policymaking and the peoples’ interests (Rooduijn and Akkerman 2017). Referendums keep politicians in check by ensuring that they seek approval from the general public before taking decisions. Therefore, we expect that referendum support among RR and RL voters is partly explained by anti-elitist sentiments (H4).

Whereas some scholars argue that populist forces are diametrically opposed to the principle of representation (Rosanvallon 2008, Taggart 2000), others argue that populists are not against representation per se, but against representation by the wrong people, that is: the corrupt elite, and want to see their own politicians in power instead (Mudde 2007, Rooduijn and Akkerman 2017). According to Pirro, RR parties cannot be regarded as anti-system parties because they do not seek to overturn the democratic system, but rather to remold certain aspects of it (2015: 3). This debate suggests that distrust in representative institutions and anti-elitism should be considered separately as explanations for referendum support among RR and RL voters.

Several studies demonstrate that opposition to immigration is the most important attitude explaining RR party support (Ivarsflaten 2008, Van Der Brug et al. 2000, Zhirkov 2014). Mudde refers to these parties as ‘nativist’, arguing that they are not just opposed to foreign persons, but to all foreign elements including ideas and influences (2007: 22). Linked to this is the claim that populists perceive society as a homogenous entity, and therefore, reject the existence of diverse groups and interests (Taggart 2000). The RL might also be opposed to immigration if, for example, immigrants were willing to work for lower wages resulting in greater inequality. However, previous research finds that anti-immigrant attitudes are not associated with a RL ideology (Visser et al. 2014) and left-wing populism is generally more inclusionary than right-wing populism (Mudde and Kaltwasser 2013).

Referendums are predicated on the strength of numbers, which means they can provide a tool for majority action against unpopular minorities, as demonstrated by the Swiss and American experience with direct democracy (Gamble 1997, Haider-Markel, Querze and Lindaman 2007). RR parties in Europe have used direct democracy to directly or indirectly target immigrants. For example, the Swiss People’s Party put forward an initiative to deport foreigners accused of minor felonies and Hungary’s ruling party Fidesz initiated a referendum against refugee reception quotas imposed by the European Union. Therefore, we argue that referendum support among RR voters is partly explained by anti-immigrant attitudes (H5). This explanation is less likely to hold for RL voters who are not as strongly characterized by such attitudes.

The RL is characterized primarily by its rejection of the capitalist economic system (Ramiro 2016). These parties criticize market-liberal policies as the cause of existing economic and political inequalities and demand that governments take measures to reduce income differences (March 2009). Previous research demonstrates that support for income redistribution is one of the main determinants of support for a RL ideology (Akkerman et al. 2017, Visser et al. 2014). Most accounts of the RR tend to argue that these parties vacillate between market-liberal and state-oriented policies depending on the context, however, a recent cross-national study by Zhirkov (2014) finds those who oppose economic redistribution are more likely to vote for the RR.

Classical theories of democracy suggest that public involvement in policymaking stimulates a transfer of wealth from the rich to the poor (Wagschal 1997). Because referendums benefit the majority of citizens, they are expected to prevent the concentration of wealth in the hands of a select few. Income distribution is generally skewed to the right which means that the median voter’s income is below average, therefore, popular votes in which the majority decide will always favor redistribution (Freitag and Vatter 2006). In Switzerland, direct democracy has been used to address economic injustices, for example in 2016 citizens voted on establishing a universal basic income (although this proposal was rejected). Therefore, we argue that referendum support among RL voters is partly explained by a preference for income redistribution (H6). This explanation is unlikely to hold for RR voters who are not as favorable towards redistribution as RL voters.

The data were obtained from two sources: the sixth wave (2012) of the European Social Survey (ESS) and the 2017 Polpart survey conducted in seven European countries (Netherlands, United Kingdom, Germany, Switzerland, Hungary, Romania, and Greece) between June and August 2017. The Polpart survey was administered by global research company Kantar-TNS and respondents were recruited from nationwide online panels based on quotas for age, sex, education, and employment status that were applied similarly in each country.¹ Merging these datasets provides more cases for testing our expectation that the association between radical voting and referendum support is weaker in countries where these parties secured a greater percentage of votes (H2). The Polpart survey includes three countries that are not in our ESS sample (RO, GR, UK) and it adds a second observation to the four countries that are in both surveys (due to elections in the five-year time-lag between the surveys, the RR and RL vote-shares have changed).

In the pooled dataset, there are 48,249 respondents from 26 European countries and 30 country-years, four countries having participated in both surveys. Non-European countries and countries without radical voters in the sample were excluded. Some countries do not have both RR and RL voters (see Appendix 1) but excluding these countries leads to very similar results (it makes the differences in referendum support between voter groups slightly stronger), therefore, we decided to keep them in the sample. After dropping respondents who were ineligible to vote (7.39%) and those with missing values on referendum support (3.33% of eligible voters) our ESS analytical sample includes 40,091 respondents from 23 countries. The Polpart survey did not ask respondents about voter eligibility and there were no missing values on referendum support as this was a forced-response question. After dropping non-citizens who would be ineligible to vote in elections our Polpart analytical sample includes 8158 respondents from 7 countries.

Our individual-level hypotheses about the attitudes explaining the association between radical voting and referendum support (H3–H6) were tested on a smaller sample. Firstly, we only used the 2017 Polpart survey as the 2012 ESS is less current, having been conducted several years before Brexit and other controversial anti-immigrant and populist referendums held in Europe (more recent ESS waves do not include an item on referendum support). Furthermore, the ESS lacks items for measuring attitudes towards political elites which according to the populism literature are very closely related to ideas about popular sovereignty. However, because the ESS includes a greater number of countries and more representative samples than the Polpart survey, we also investigated whether our individual-level explanations for the link between radical voting and referendum support also hold using data from the 2012 ESS. Although most of the key results are similar, we noted some differences between the surveys which are discussed in Appendix 2.

Second, we tested our individual-level explanations for the association between radical voting and referendum support in Western European countries only. As demonstrated by the grand means in Table 1, in Central and Eastern Europe RL voters scored the same on referendum support as moderate voters and the difference between RR and moderate voters was very small. In Western Europe, the differences in referendum support between voter groups are greater. Furthermore, previous research has argued that radical parties in Central and Eastern Europe are different to their Western European counterparts: RR parties are more openly anti-democratic and hostile to popular sovereignty (Allen 2017, Bustikova 2018) and RL parties have preserved the party elitism of Marxist-Leninism (Musto 2017). Therefore, the analytical sample for testing our individual-level hypotheses includes 5959 respondents from the 5 Western European countries in the Polpart survey, after excluding respondents with missing values on the additional individual-level predictors included in these analyses (0.2%).

Support for referendums. In both surveys, respondents were asked ‘how important is it for democracy that citizens have the final say on political issues by voting in referendums?' (0 = not at all important/10 = very important).

Voting Behavior. In the ESS respondents indicated which party they voted for in the last national elections and in the Polpart survey respondents indicated which party they would vote for if national elections were held tomorrow. Dummies were created for (1) RR voters; (2) RL voters; (3) moderate voters; and (4) respondents who didn’t vote (or intended not to vote in Polpart), voted blank, didn’t know, or refused to say. Moderate voters were used as the reference category. The proportions of respondents per voter group are provided in Table 2 as means. In the Polpart survey, the percentage of respondents who said they didn’t know which party they would vote for was very high, especially in Romania (22.64), Hungary (21.53), and Greece (26.35). This could be attributed to widespread disillusionment with politics and the volatility of party systems in these countries.

To identify RR and RL parties we referred to the works of several scholars (Akkerman et al. 2016, Kioupkiolis and Katsambekis 2018, March 2009, Mudde 2007, Pirro 2015, Ramiro 2016, Rooduijn 2017, Rydgren 2018) as well as the descriptions of the parties provided by the ESS (ESS 2012). Previous research by Mudde (2007) and Rooduijn (2017) suggests that RR and RL parties in general contain elements of populism, therefore, we adopted a broader classification of these parties instead of focusing only on those commonly referred to as populist. Furthermore, as we argued in the theory section, there may be other attitudes besides populism linking RR (anti-immigrant attitudes) and RL (income redistribution) voters to referendum support. Despite being ruling parties, Fidesz (Hungary), the Swiss People’s Party, Law and Justice (Poland), and Syriza (Greece) are often identified as radical parties. However, because Fidesz ran for office together with the Christian Democratic Party (KDNP) in 2010 and 2014, the 2012 ESS does not distinguish between Fidesz and KDNP voters. Therefore, we only include Fidesz voters among the RR in the Polpart sample. For the classification of parties refer to Appendix 1.

RR and RL Vote-Shares. Two variables were created representing the percentage of votes secured by the RR and RL in the last parliamentary elections. The data were obtained from the European Elections Database (2012), compiled by the Norwegian Centre for Research Data. When two or more RR or RL parties obtained votes in the same election the percentages were added. If elections were held during the fieldwork period the percentage for the current rather than the preceding elections was used. We assume that when taking the survey, respondents are more likely to be influenced by current estimates of their party’s popularity than by election results from several years ago. The vote-share percentages for each country-year are presented in Table 1. The vote-share variables were centered on the mean of the country-year percentages in order to facilitate the interpretation of interaction models. Using the percentage of seats won by radical parties in legislative bodies would yield similar results as this measure is highly correlated with the percentage of votes (r = 0.97 for RL and r = 0.98 for RR).

Controls. A dummy variable representing Western (0) versus Central and Eastern (1) European countries was included as a control in the cross-national analyses. All countries classified as CEE were state socialist countries.

Distrust in representative institutions (also referred to as Political Distrust). Respondents indicated how much trust they have in Parliament, Political Parties, and Politicians on a scale ranging from 0 (‘complete distrust’) to 10 (‘complete trust’). The responses were reverse coded so that a higher score means less trust. The mean scores on these items were averaged, creating one scale. Factor analyses showed that the items loaded onto one scale in each country, with the alphas ranging from 0.82 to 0.92.

Anti-elitism. The mean scores on the following four items were averaged: ‘elected officials talk too much and take too little action’; ‘most politicians make a lot of promises but do not actually do anything’; ‘I don’t think politicians care much what people like me think’; and ‘most politicians are in politics only for what they can get out of it personally’ (0 = strongly disagree/4 = strongly agree). These items loaded onto the same factor in all countries, with alphas ranging from 0.82 in Greece to 0.92 in the Netherlands. A factor analysis demonstrated that the variables on distrust in representative institutions and anti-elitism represented two different dimensions.

Anti-immigrant attitudes. The mean scores on the following three items were averaged: ‘immigration is good or bad for the [country]’s economy?’ (0 = good/10 = bad), ‘[country]’s cultural life is undermined or enriched by immigrants?’ (0 = enriched/10 = undermined), and ‘immigrants make [country] a worse or better place to live?’ (0 = better/10 = worse). Factor analyses showed that the items loaded onto one scale in each country, with the alphas ranging from 0.83 to 0.93.

Support for income redistribution. Respondents were asked whether ‘The Government should redistribute income from those who are better off to those who are worse off’ (0 = strongly disagree/4 = strongly agree).

Controls. Age, sex (1 = female; 0 = male); education (7-point scale); and a subjective measure of income were included as controls. The measurement of education is derived from the International Standard Classification of Education and ranges from less than lower secondary education to higher tertiary education. As a subjective measure of income, respondents indicated which of the following descriptions comes closest to their feelings about their household income nowadays: ‘very difficult on present income’ (0); ‘difficult on present income’ (1); ‘coping on present income’ (2); and ‘living comfortably on present income’ (3).

Descriptives for the individual-level variables are provided in Table 2.

First we examined the differences in referendum support between voter groups based on descriptives per country-year (Table 1). After this, we tested whether the differences between voter groups were significant in the pooled dataset combining the ESS and Polpart samples (N = 48,249). Using the same sample, we tested whether the differences in referendum support between radical and moderate voters were smaller in countries were radical parties obtained a larger percentage of votes (H2). This was done by including cross-level interactions between the voter group dummy variables and the vote-share variables. In order to account for the nested structure of the data, we analyzed the data with multilevel regressions consisting of three levels (respondents + country-years + countries). Random slopes for the voter group dummy variables were included at the survey-country level because the vote-share variables are measured at this level.

Our individual-level hypotheses were tested on the Polpart sample of Western European countries (5959) using OLS regressions with fixed effects for countries and robust standard errors (as we are now using fewer countries and no longer testing country-level effects). We included the individual attitudes in separate models and observed whether there was a reduction in the b-coefficients of the RR and RL dummies. Based on this we could assess to what extent our individual attitudes ‘explained away’ the association between voting for a RR or RL party and referendum support, however, we could not actually test whether the reductions in these associations were statistically significant. Hence, the results do not provide a formal test of our hypotheses on the individual attitudes linking RR and RL voters to referendum support. However, we do test whether the associations between these voters and referendum support are still significant when all confounders are included. All analyses were done in STATA 14.

Starting with the descriptives presented in Table 1, the grand means demonstrate that in Western Europe RR and RL voters both scored higher on referendum support than moderate voters with RR voters scoring the highest. In Central and Eastern Europe, RL and moderate scored the same and the difference between RR and moderate voters was much smaller than in Western Europe. Looking at the means by country-year (i.e. country-survey combination), RR voters scored higher than moderate voters in 24 out of 26 country-years, whereas RL voters scored higher than moderate voters in 15 out of 22 country-years. Interestingly, UKIP voters scored almost the same as moderate voters in the Polpart survey, which was conducted one year after the Brexit Referendum. The United Kingdom was not included in our ESS sample because UKIP voters could not be identified in the data (they were included in the ‘other party’ category, most likely because UKIP secured only 3.1% of votes in the 2010 elections). For the countries participating in both surveys (NL, DE, HU, CH), the average scores on referendum support for all groups were much lower in the 2017 Polpart survey than in the 2012 ESS.

In Table 3, Model 1 we test the differences in referendum support between the voter groups on the pooled dataset combining the ESS and Polpart samples. RR and RL voters are significantly more favorable towards the use of referendums than moderate voters, confirming H1a and H1b. The difference in support between RR and moderate voters is almost twice as large as the difference between RL and moderate voters. However, an additional analysis in which we ran the same model but changed the reference category demonstrated that the difference between RR and RL voters is not significant.

The random effects of the constant in Table 3 demonstrate that there is variation in referendum support at both the country-year and country levels but the variation is higher at the country level. The random slopes of the dummy variables for RR and RL voters indicate substantial variation in the effects of RR and RL voting on referendum support across country-years, therefore, multilevel analysis with cross-level interactions is considered appropriate. The cross-level interactions in Models 2 and 3 test whether support for direct democracy among RR and RL voters is weaker in countries where these parties secure a greater percentage of votes (H2). Note that countries without RR voters were excluded from Model 2 and countries without RL voters were excluded from Model 3. There are no main effects of the RR and RL vote-shares on referendum support. The main effect of RR voting in Model 2 indicates that in a country-year with the mean RR vote-share (10.82%), these voters are predicted to score 0.61 points higher on referendum support (which is measured on an 11-point scale) than moderate voters. The significant interaction coefficient in Model 2 (b = −0.016) demonstrates that referendum support among RR voters decreases as the percentage of votes secured by RR parties in the last parliamentary elections increases. This means, for example, that an increase in the RR vote-share of 10 percentage points is associated with a decrease in the effect of RR voting on referendum support by 0.16 points. Hence, in a country where the RR secured 20.82% of votes (10 percentage points above average) RR voters are predicted to score 0.77 points higher than moderate voters, instead of 0.61 in the ‘average country’. The non-significant interaction coefficient in Model 3 (b = −0.008) demonstrates that referendum support among RL voters does not depend on the electoral success of RL parties. Therefore, H2 is confirmed for the RR but rejected for the RL.

The individual-level explanations for the differences in referendum support between radical and moderate voters are tested in Table 4 on the Polpart sample of Western European countries. Model 1, which only includes the voter group dummies and the control variables, demonstrates that RR and RL voters are more favorable towards the use of referendums than moderate voters even when controlling for age, sex, education, and income. The difference between RR and moderate voters is almost two times bigger than the difference between RL and moderate voters. An additional model in which we changed the reference category (not shown) revealed that the difference between RR and RL voters is significant, indicating that RR voters are the most favorable towards direct democracy of the three voter groups. Turning to the controls, referendum support is stronger among older people, women, lower educated people, and those who have more difficulties making ends meet.

In the subsequent models, we included the individual attitudes separately and observed whether there was a reduction in the association between radical voting and referendum support. Model 2 demonstrates that distrust in representative institutions (parliament, parties, and politicians) has a positive effect on referendum support. The b-coefficient of the RR in the model with political distrust (Model 2) is 23% smaller than the b-coefficient of the RR in the model without political distrust (Model 1), suggesting that this attitude partly explains why RR voters score higher on referendum support than moderate voters. By contrast the b-coefficient of the RL in Model 2 is the same as the b-coefficient of the RL in Model 1. Therefore, our claim that distrust in representative institutions partly explains referendum among radical voters (H3) only holds for RR voters.

Anti-elitism is positively related to referendum support (Model 3) and including this attitude in the model explains away 48% of the difference in referendum support between RR and moderate voters, but only 6% of the difference between RL and moderate voters. Therefore, our claim that anti-elitism partly explain referendum support among radical voters holds for RR voters but hardly for RL voters (H4).

Model 4 demonstrates that anti-immigrant attitudes have a small positive effect on referendum support. The coefficient for RR voters in the model with anti-immigrant attitudes is 17% smaller than the coefficient for RR voters in Model 1. By contrast the coefficient for RL voters in Model 4 is larger than the coefficient for RL voters in Model 1. As expected, Model 4 suggests that referendum support among RR voters is partly explained by anti-immigrant attitudes (H5).

Finally, support for income redistribution is positively related to referendum support (Model 5) and including this attitude in the model explains away 41% of the difference in referendum support between RL and moderate voters but hardly any of the difference between RR and moderate voters. As expected, Model 5 suggests that referendum support among RL voters is partly explained by a preference for income redistribution (H6).

In the final model, all controls and attitudes were included at the same time to see whether taken together these variables entirely explain the association between radical voting and referendum support. The results show that a significant difference in referendum support between RR and moderate voters as well as between RL and moderate voters remains, even when accounting for all hypothesized explanations. Model 6 also demonstrates that the effect of political distrust turns negative when controlling for anti-elitism, suggesting that anti-elitism predicts referendum support better than distrust in representative institutions. In order to compare the strengths of the effects, standardized coefficients (Betas) are also included in Model 6, indicating that anti-elitism is by far the strongest predictor of referendum support. As explained in the Data and Method section, we also tested our individual-level explanations using the 2012 ESS, the results for which are available in Appendix 2.

Our key finding is that in Western European countries, RR and RL voters are more favorable towards the use of referendums than moderate voters. This finding confirms previous research on the link between RR voting and referendum support conducted in Germany, Belgium, and the Netherlands (Pauwels 2014), but contrasts with similar research conducted in Canada, Australia, and New Zealand, suggesting that this link is more characteristic of European democracies (Bowler et al. 2017). Unlike Rooduijn (2017), who only compares German and Dutch RL voters to other voter groups and also controls for left-right self-placement, we demonstrate that RL voters in Western Europe also share a preference for referendums, suggesting that these parties are shedding their predecessors’ emphasis on the party elite (Rooduijn and Akkerman 2017). While the differences in referendum support between voter groups are much smaller in Central and Eastern Europe, overall support for the use of these instruments is much stronger than in Western Europe. Higher levels of corruption and political instability in Central and Eastern Europe might explain why the desire for alternative ways of doing politics is more widespread in this region (Kriesi 2014, Stanley 2017).

The association between radical voting and referendum support may be cause for concern, particularly among participatory democrats who claim that participation in political decision-making leads to a more inclusive base of support for politics as well as a greater sense of community and stronger commitment to democratic values among participants (Matsusaka 2004, Pateman 1970, Roberts 2004). RR and RL parties are often described as authoritarian, exclusionary, or fundamentally opposed to certain key principles of liberal democracy (Mudde 2007), making it hard to imagine that their voters would demand referendums for the same reasons as participatory democrats. Indeed, our results suggest that xenophobia is one of the motivations for referendum support among RR voters. Critics of direct democracy have warned that these instruments may cause harm to minorities, as illustrated by the referendums directed against immigrants in Switzerland (2014) and Hungary (2016) and against the LGBT community in Croatia (2013), Slovakia (2015), and Romania (2018). While our results suggest potential for RR leaders to mobilize voters around calls for such referendums, measures could be taken to protect minorities. For example, by introducing constitutional restraints on the subjects available for popular vote or opting for advisory over binding outcomes (Rojon et al. 2019).

Furthermore, if the demand for direct citizen participation in political decisions is not evenly spread throughout the electorate, then referendums may become stigmatized as catering to the views of RR and RL voters. Completely denying citizens the right to decide on important matters affecting their lives might worsen trust in government, therefore, efforts must be made so that direct democracy represents a means for change among all segments of the political spectrum and not just RR and RL voters. For example, other political parties might increase support for direct democracy among their electorates by starting initiatives for referendums on issues concerning their voters.

Mudde has argued that for populist RR parties the introduction of direct democracy only represents a temporary measure to circumvent an out-of-touch elite. Once an elite that is representative of their views has been established, direct democracy is no longer necessary (Henley 2016). Although we did not conduct longitudinal analyses to test Mudde’s claim, we demonstrated that the difference in referendum support between RR and moderate voters is smaller in countries where the percentage of votes obtained by RR parties is higher. However, the means in Table 1 indicate that even in countries where RR parties have been most successful, RR voters score higher on referendum support than moderate voters (Hungary being the exception). By contrast, the link between RL voting and referendum support does not depend on the electoral success of these parties suggesting that for RL voters direct democracy does not represent a temporary measure to replace an unresponsive elite.

Given the long-term decline in political support across established democracies, participatory democrats have argued that more inclusive decision-making processes might stimulate renewed trust in politics, particularly among the politically disaffected and socio-economically disadvantaged (Smith 2009, Qvortrup 2017). Indeed, our results demonstrate that referendum support is stronger among politically distrusting, anti-elitist, lower educated, and lower income individuals, and that referendum support among RR voters is partially explained by dissatisfaction with politics. Therefore, greater use of referendums and other direct democratic initiatives might convince these individuals that their opinions are being counted. These findings challenge previous research from the United States claiming that despite increasingly low levels of trust in government, most citizens do not want to be involved in political decision-making (Hibbing and Theiss-Morse 2002)

Furthermore, our results demonstrate that anti-elitism is not only the main explanation for referendum support among RR voters, but also the strongest predictor of referendum support in general. Anti-elitism overrules distrust in representative institutions as a predictor of referendum support suggesting that the demand for more control over political decisions is not driven by opposition to representative politics, but by distaste for political elites. This finding may be cause for concern if citizens are using referendums to vote against unpopular elites instead of providing input on specific policy questions. Such concerns might be addressed by making greater use of electoral recall, an instrument designed to remove unresponsive or misbehaving representatives from office by popular vote, thereby separating issue-specific concerns from anti-elitist motivations.

Whereas anti-elitism seems to be a key explanation for referendum support among RR voters, neither political distrust nor anti-elitism explain much of the difference in referendum support between RL and moderate voters, undermining our claim that these voters seek greater involvement in political decisions out of frustration with politics. Instead, our results suggest that RL voters perceive direct democracy as a vehicle for policies addressing economic inequality. In general, we find that individuals who think ‘the government should redistribute income from those who are better off to those who are worse off’ express greater support for referendums. This would support our claim that referendums are appealing to individuals in favor of redressing economic inequalities as these instruments, by nature of their alignment with the median voter’s preferences, may stimulate redistribution of wealth.

Because referendums benefit the majority of citizens, they are expected to prevent the concentration of wealth in the hands of a select few. Income distribution is generally skewed to the right which means that the median voter’s income is below average, therefore, popular votes in which the majority decide will always favor redistribution

In conclusion, the evidence suggests that RR voters seek greater control over political decisions out of frustration with an out-of-touch elite and immigration, while RL voters are motivated by their desire to reduce differences in income levels. However, it is important to note that even after controlling for a range of explanations for the association between radical voting and referendum support, RR and RL voters still score higher on referendum support than moderate voters. Either there is another explanation we have not accounted for or there is a direct causal relationship between radical voting and referendum support. Our correlational design does not allow for causal inferences, therefore, future research might investigate whether radical voters have a deep-seated preference for direct democracy, which influences their voting behavior, or whether populist leaders incite preferences for direct democracy among these voters by calling for referendums against corrupted elites, immigration, or capitalist injustice (or both). To better determine whether RR and RL are truly direct democrats, future research might also investigate whether these voters participate more in referendums than moderate voters and whether they would support these instruments even when their political preferences do not receive majority support.

No potential conflict of interest was reported by the author(s).

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Sebastien Rojon is a post-doctoral researcher working on the ERC-funded project “Non-elected politics: Cure or Curse for the Crisis of Representative Democracy” at the Université libre de Bruxelles. During his PhD at the Vrije Universiteit Amsterdam, he investigated public support for citizen participation in political decision-making through referendums and deliberative meetings. More specifically, he looked at which groups of citizens are most favourable to direct decision-making and how this differs across countries.

Arieke J. Rijken is a researcher at the Netherlands Interdisciplinary Demographic Institute. This article was written while she was affiliated with the Vrije Universiteit Amsterdam, Department of Sociology. She received her PhD from Utrecht University, the Netherlands, in 2009 and has published in the fields of family sociology, demography, criminology, and political science.

Although the 2012 ESS is less recent and does not include measures for anti-elitism, it has more countries (13 Western European countries) and more representative samples than the 2017 Polpart survey. Therefore, we investigated whether our individual-level explanations for the association between radical voting and referendum support also hold using data from the ESS. The variables were measured identically in both surveys with the exception that political distrust was measured on an 11-point scale in the ESS instead of a 5-point scale, hence this variable was standardized to make the effects presented in Table A2 comparable. Factor analyses showed that the multi-item scales created for political distrust and anti-immigrant attitudes are also reliable in the ESS, with the alphas ranging from 0.73 to 0.90 in each country for political distrust and from 0.78 to 0.90 in each country for anti-immigrant attitudes. See Table A1 for descriptive statistics.

Comparing Models 1 and 2 in Table A2 demonstrates that in both surveys, RR voters are more favorable towards the use of referendums than moderate voters, even when controlling for all confounders, and referendum support is generally stronger among individuals who are dissatisfied with representative institutions and believe that the government should take measures to reduce differences in income levels.

Turning to the differences between the key findings: first, whereas the association between voting for a RL party and referendum support is explained away in the 2012 ESS, this association remains significant in the 2017 Polpart survey. A potential explanation for this is that over time RL voters increasingly perceive referendums as an important feature of democracy, rather than a channel for expressing dissatisfaction with politics or redressing economic inequalities. Second, whereas in the ESS anti-immigrant attitudes have a small negative effect on referendum support, in the Polpart survey these attitudes have a small positive effect. Given the xenophobic undertones of recent referendum campaigns, namely, the Swiss Mass Immigration Initiative (2014), Brexit (2016), the Hungarian Migrant Quota Referendum (2016), and the Dutch initiative against an EU-Ukraine rapprochement, these instruments may have become more relevant to anti-immigrant forces. Looking more closely at the results by country revealed, for example, that in the Netherlands anti-immigrant attitudes are negatively related with referendum support in 2012 but positively related with referendum support in 2017 (results upon request).