From Understandings of Democracy to Support for Political Candidates

Our main argument posits that political culture – specifically, citizens’ heterogeneous understandings of democracy – shapes political behaviour and, in particular, vote choice in contexts of democratic backsliding in ways that enable authoritarian-leaning elites to access and retain power despite engaging in violations of liberal democratic standards. At the macro level, this implies that democratic backsliding is likely to result where divergent understandings of democracy among the citizenry meet political elites willing to exploit such heterogeneous attitudes to retain their grip on power. Our study theorizes and empirically explores the micro-level foundations of this overarching argument.

Earlier explorations of heterogeneous understandings of democracy among citizens have often focused on a wide range of divergent conceptualizations. In their pioneering study in this field, Schedler and Sarsfield (Reference Schedler and Sarsfield2007) explore ‘democrats with adjectives’ as a mirror image of earlier debates around ‘democracies with adjectives’ (Collier and Levitsky Reference Collier and Levitsky1997). Focusing on Mexico, they distinguish six groups of respondents – liberal democrats, intolerant democrats, paternalistic democrats, homophobic democrats, exclusionary democrats, and ambivalent non-democrats – with only the first group fully in line with liberal democracy. In a similar effort, a comparative study among twelve Latin American countries identifies groups of citizens based on their support for polyarchy, pitting those who support all five dimensions of polyarchy (the ‘polyarchs’) against four mixed profiles of power constrainers, power checkers, power delegators, and power restrainers (Carlin and Singer Reference Carlin and Singer2011). Most recently, Davis, Gaddie and Goidel (Reference Davis, Gaddie and Goidel2022) used a combination of open- and closed-ended survey questions to detect latent classes of indifferent respondents, proceduralists, moderates, and social democrats in the USA.

While these studies have the merit of highlighting the presence of distinct understandings of democracy in a given population, their main limitation consists in their case-specific nature, which makes it hard to draw insights beyond the specific context in which the typologies were developed. Adopting a more deductive approach, Ferrín (Reference Ferrín, Ferrín and Kriesi2016) conducted a comparative study of Europeans’ attitudes towards democracy that probes citizens’ support for three alternative models of democracy: liberal democracy, focused on the electoral process and the rule of law; social democracy, articulated around distributive justice and social and economic rights; and direct democracy, emphasizing direct legislation by citizens. Drawing on a comprehensive item battery they fielded within the European Social Survey (ESS), Kriesi and Morlino (Reference Kriesi, Morlino, Ferrín and Kriesi2016, 308) conclude that despite differences in the relative emphasis upon the three main models of democracy tested empirically, ‘the basic principles of liberal democracy are universally endorsed across Europe’. A forthcoming update of the analysis based on a more recent wave of ESS data confirms the continued centrality of free and fair elections and the rule of law for Europeans’ understandings of democracy but notes a partial erosion in support for other elements such as the protection of minority rights (Hernandez, Reference Hernandez, Kriesi and Ferrınforthcoming).

We adopt a similar approach by zooming in on three predefined categories of democratic attitudes. Whereas Kriesi and Morlino (Reference Kriesi, Morlino, Ferrín and Kriesi2016) explore citizens’ support for distinct conceptions of democracy that all qualify as equally democratic, we are interested in differentiating between conceptions that may grant more or less leeway to elected elites seeking to expand their executive powers by dismantling democratic checks and balances. We distinguish these three conceptions primarily based on the supposed source of democratic legitimacy.

Our baseline conception of democracy is a liberal understanding that goes beyond a general regime preference for democracy to embrace pluralism, executive constraints, as well as equal rights and civil liberties for all citizens (O’Donnell Reference O’Donnell1998). In contexts of democratic backsliding, it is typically the liberal aspects of democracy, most notably minority rights protection and various forms of constraints on the executive, that come under pressure. The main conflict line thus runs between liberal and non-liberal forms of democracy rather than between electoral vs. non-electoral regime types.

The second conception of democracy we include in our analysis picks up on the ‘majoritarian threat to liberal democracy’ identified by Grossman et al. (Reference Grossman, Kronick, Levendusky and Meredith2022). Reflecting a populist emphasis on power lying with ‘the people’, voters holding majoritarian views consider decisions supported by the political majority as democratic per se, including when they go against central precepts of liberal democratic conceptions such as pluralism and minority protection (Grigoriadis Reference Grigoriadis2018; Urbinati Reference Urbinati2017). As a result, they grant the elected government considerable leeway to limit executive constraints or pursue critical media in an effort to implement its political programme, making them potentially more open to tolerating political leaders who undermine traditional checks and balances.

Finally, we add a conception that has been qualified as an authoritarian view of democracy, according to which the legitimacy of a political system derives primarily from its ability to maintain social order and prevent chaos. Earlier studies qualify such ‘authoritarian notions of democracy’ as ‘democracy misunderstood’ (Kirsch and Welzel Reference Kirsch and Welzel2019) or ‘democracy confused’ (Kruse, Ravlik and Welzel Reference Kruse, Ravlik and Welzel2019) to signal their incompatibility with liberal democratic orientations. While an authoritarian understanding of democracy may, therefore, appear as a conceptual oxymoron, authoritarian attitudes have been shown to empirically exist among citizens in many democracies (Singh and Dunn Reference Singh and Dunn2013), including in Europe (Vasilopoulos and Lachat Reference Vasilopoulos and Lachat2018), Asia (Dore Reference Dore2014), and Latin America (Cohen and Smith Reference Cohen and Smith2016).

By studying the effects of diverse understandings of democracy on citizens’ political behaviour, we respond to calls to investigate how the strength of democratic beliefs (Carlin Reference Carlin2018, 419) and the liberal-democratic quality of citizens’ regime preferences (Wuttke, Gavras and Schoen Reference Wuttke, Gavras and Schoen2022) relate to vote choice and eventual democratic outcomes. We ask: how do divergent understandings of democracy shape vote choice in contexts of democratic backsliding? We posit two distinct mechanisms that may account for the supposed linkage between understandings of democracy and political behaviour at the ballot box.

On the one hand, the relationship may be uniquely driven by citizens who display high levels of support for liberal democratic conceptions. Such individuals can be expected to be particularly adamant about seeing the liberal dimension of democracy protected by political candidates and, thus, are more prone than others to punish candidates for holding alternative views. By contrast, respondents with majoritarian or authoritarian attitudes can be thought to prioritize alternative features in a candidate’s profile. We thus posit liberal democratic commitment as the first mechanism linking understandings of democracy to vote choice:

On the other hand, the relationship between democratic attitudes and political behaviour may be based on a more general congruence between voters’ understandings of democracy and the democratic positions expressed by political candidates. Such value congruence has been amply studied to explain the linkage between citizens’ democratic values and regime type (Almond and Verba Reference Almond and Verba1963; Welzel Reference Welzel2007, Reference Welzel2021) at the macro level, with citizen demand for democracy and civil liberties thought to create pressures to adjust the supply of such freedoms by the political system and elites (Welzel and Klingemann, Reference Welzel and Klingemann2007, Reference Welzel and Klingemann2008). Zooming in on the micro-level relationship between voters and candidates’ democratic views, we expect an overlap between the two to drive vote choice in this case, with respondents preferring candidates whose positions mirror their own understanding of democracy, irrespective of whether this is liberal, majoritarian, or authoritarian. We therefore hypothesize:

In sum, we expect divergent understandings of democracy among citizens to affect their positioning in electoral contests when competing candidates express a range of democratic positions, some of which openly conflict with liberal democratic norms. We posit two mechanisms that may explain these linkages, namely citizens’ specific commitment to liberal democratic norms or the more general congruence between citizens’ understandings of democracy and the democratic positions expressed by political candidates. To account for partisan-related dynamics, we also assess the interplay between understandings of democracy and partisan preferences and the relevance of partisan voting in explaining respondents’ assessments of competing candidates and, notably, their rejection of non-liberal candidates. We expect such dynamics, where present, to act in parallel and thus in a complementary fashion to our emphasis upon democratic attitudes.

Study Design

Our analytical approach leverages a candidate choice conjoint experiment. This design allows us to integrate alternative elements alongside the democratic positions in candidates’ profiles to capture the potential trade-offs in which voters engage (Schedler Reference Schedler2019; Svolik Reference Svolik2020). The resulting multidimensional set-up allows us to assess the weight of democratic positions when it comes to respondents’ evaluations of competing candidates. We are thus able to probe the linkages between divergent understandings of democracy and citizens’ responses to concrete manifestations of democratic backsliding while controlling for partisan-related factors.

In our study, we placed respondents into a hypothetical election situation and asked them to choose between two competing profiles of candidates running for seats in the national lower house (Sejm).Footnote ² We use the conjoint setting to effectively manipulate elite behaviour – the supply side of our argument – by varying the positions in our candidate profiles regarding the nomination of judges and the role of public media. We strive to capture divergent views of democracy on the elite side by formulating the levels for the two democratic attributes in line with the liberal, majoritarian, and authoritarian understandings we developed for the citizens’ perspective. Table 1 displays our democratic attributes along with the levels reflecting distinct understandings of democracy.

Table 1.

Democratic attributes and levels

Our selected attributes concern two elements – judicial independence and media freedom – that are crucial to liberal democracy but also offer a range of options as to how they may be implemented in a democratic system. Using two distinct democratic attributes allows us to conduct two separate tests of our argument regarding the linkages between understandings of democracy and voter responses to democratic transgressions. Our transgressions capture violations of liberal democratic norms rather than outright violations of the law (Ahmed Reference Ahmed2023), thus enabling us to probe the overall salience of candidates’ democratic views as well as the relative impact of different variations to strong liberal views upon respondents’ candidate choice. For judicial appointments, we complement the liberal position that judges should be based on cross-party consensus with alternatives that foresee a selection by the government (majoritarian) or by the leader of the ruling party (authoritarian). Regarding the role of public media, we let candidates express the liberal view that their role consists of reporting independently on political developments, suggest that they should justify government policy towards the wider public (majoritarian), or claim their role is to defend government policy against criticism (authoritarian).

We deliberately choose more subtle deviations from liberal democracy to model the gradual nature of democratic backsliding, which consists precisely of rather discrete ways of chipping away at checks and balances that only jointly amount to dismantling democratic standards (Scheppele Reference Scheppele2013). Moreover, we decided to refrain from including positions so extreme that they would draw near universal condemnation, making it difficult to discriminate whether such condemnation is driven by an actual commitment to liberal democratic norms or due to considerations of social desirability. As discussed below, despite their subtlety, respondents are able to discriminate among the three distinct levels of democratic positions we introduce for our two democratic attributes (see also Appendix Figure B.3).

We partnered with the Warsaw-based market research company Inquiry – YouGov’s representative for Central and Eastern Europe – to recruit a representative sample of Polish respondents based on age, gender, geographic origin, and vote choice at the last national election for our online survey, into which we embedded our conjoint experiment. The survey was conducted between 12 July and 12 August 2021 ( $N = 2,910$ ). As specified in our preregistration, we removed speeders and those respondents who failed attention checks from our sample (Berinsky, Margolis and Sances Reference Berinsky, Margolis and Sances2014), bringing the final sample we used for our analysis to 1,979 respondents. We report measures of sample representativeness for the final sample and full results for alternative sample specifications in the Online Appendix (Tables A.2 and A.3 and Section B.5).

We asked respondents to complete twelve discrete choice tasks, each time choosing between two candidates (forced-choice) and rating each candidate on a 7-point Likert scale ranging from ‘strongly disapprove’ (1) to ‘strongly approve’ (7). Each candidate profile was identified with a neutral label (‘Candidate A’ vs. ‘Candidate B’) and displayed randomized information on seven attributes, with the order of attributes fully randomized for each choice task. Alongside candidates’ respective democratic views,Footnote ³ our competing profiles contained information on their gender, age, policy positions, and partisanship. For partisanship, we presented respondents with a mix of candidates from all parties or party coalitions that scored above 5 per cent of vote share according to polls in June 2021 when the survey design was finalized. Choice situations also included runoffs between candidates of the same party background. We include the full attribute table in Table A.2 in the Appendix.

Average Marginal Component Effects (AMCEs) and their Limits

Our design seeks to estimate how much importance respondents assign to multidimensional candidate characteristics. First, we replicate the traditional approach by computing AMCEs for our candidate choice experiment (see Appendix Figure B.3). AMCEs allow researchers to estimate the effect of an individual treatment component over the joint distribution of the remaining attributes (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014, 10). Focusing on the attributes of the judiciary and media, the AMCEs suggest that, on average, respondents approve less of candidates who make majoritarian or authoritarian statements compared to liberal ones.Footnote ⁴

However, AMCEs do not allow us to draw inferences about the individual level. Specifically, the observed pattern may reflect a shared adherence to reasonably liberal democratic attitudes across our sample. However, AMCEs may just as well mask considerable divergence of choice behaviour within our sample, with some respondents punishing democratic transgressions very harshly, while others are indifferent or even approve of candidates expressing non-liberal democratic views. Yet such divergent behaviours at the individual level underpin the causal mechanism we seek to probe in our study, leading us to adopt a recently proposed alternative approach to conjoint analysis via individual marginal component effects.

Individual Marginal Component Effects (IMCEs) as a Measure of Individual-level Candidate Preferences

Individual marginal component effects (IMCEs) overcome some of the limitations of analyses focused on AMCEs (Zhirkov Reference Zhirkov2022). In this approach, each respondent rates a relatively high number of profiles (in our case twenty-four candidate profiles in twelve election runoffs) on a rating scale,Footnote ⁵ allowing us to estimate the effects of each candidate attribute level on the respondent’s rating for candidates. For instance, if a respondent repeatedly rates liberal candidate profiles more highly, their IMCE on the corresponding liberal attribute will be higher. Substantially, higher values indicate a stronger preference for liberal democratic candidates. We detect this preference by regressing a respondent’s ratings of the twenty-four candidates on each of the candidate’s attributes separately:

(1) $${{\bf{y}}_i} = {\alpha _{il}} + {\pi _{il}}{\bf{X}}{{\rm{'}}_{il}} + {\varepsilon _{il}},$$

where ${{\bf{y}}_i}$ is a vector of ratings for each candidate profile made by respondent $i$ , ${\bf{X}}{{\rm{'}}_{il}}$ a vector of values of attribute $l$ shown to respondent $i$ , and ${\varepsilon _{il}}$ a vector of respondent-specific errors.Footnote ⁶ We define ${\hat \pi _{il}}$ as the IMCE for attribute $l$ . In our study, we focus on individuals’ IMCEs for the two attributes relating to candidates’ statements towards democracy (that is, judicial appointments and the role of public media).Footnote ⁷ To assess the empirical relevance of congruence between respondents’ and candidates’ democratic views for respondents’ candidate evaluations, we regress the IMCEs of the two democratic attributes on the three understandings of democracy. This allows us to examine the relative importance of congruence separately for all three understandings, enabling us to evaluate whether only respondents with strong liberal democratic attitudes lend greater weight to candidates’ democratic views or whether there is a generalized preference for candidates expressing democratic views that are congruent with those of the respondent. As mentioned above, we observe the ratings separately for each of the two democratic attributes in order to assess how the area in which a democratic transgression occurs may concern respondents to a different extent.

Previous research has proposed to divide a population into subgroups of interest (for example, based on gender or partisanship) and study average conjoint behaviour separately for these groups (Leeper, Hobolt and Tilley Reference Leeper, Hobolt and Tilley2020). However, defining such subgroups for our attitudinal concept of understandings of democracy would require imposing arbitrary thresholds to distinguish different subgroups from one another. IMCEs, by contrast, enable us to assess different understandings of democracy on a continuous scale and examine their relationship with political choice. Besides, aggregate analyses usually allow for examining only one covariate of interest at a time. However, we often expect the covariate of interest to vary with other covariates, raising concerns about omitted variable bias.Footnote ⁸ By contrast, determining to what extent individual respondents care about candidates’ stances toward democracy allows us to consider a range of explanatory variables jointly in a regression framework.Footnote ⁹

IMCEs rely on the same set of assumptions as AMCEs. That is, only when the assumptions of (1) stability and no carryover effects, (2) no profile-order effects, and (3) completely independent randomization of the profiles in a conjoint experiment hold can IMCEs be estimated independently for each respondent (Hainmueller, Hopkins and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014; Zhirkov Reference Zhirkov2022). For our candidate choice experiment, we verified assumption 1 (see Online Appendix B.2.2), and assumptions 2 and 3 are true by design, allowing us to proceed with estimating IMCEs. In Online Appendix D.2, we also implement an alternative machine-learning approach to estimating IMCEs (Robinson and Duch Reference Robinson and Duch2024) and find similar results.

Independent Variable: Understandings of Democracy

Empirical studies often equate democratic commitment with citizens’ support for the generic concept of democracy (Wuttke, Gavras and Schoen Reference Wuttke, Gavras and Schoen2022; Foa and Mounk Reference Foa and Mounk2016, Reference Foa and Mounk2017). This narrow understanding – and corresponding measurement – of democratic commitment is increasingly recognized as a key limitation in accurately assessing citizens’ democratic beliefs (Inglehart Reference Inglehart2003; Ananda and Bol Reference Ananda and Bol2021; Alonso Reference Alonso, Ferrín and Kriesi2016; König, Siewert, and Ackermann Reference König, Siewert and Ackermann2022). To explain heterogeneity in individuals’ evaluations of candidates expressing different views of democracy, we instead implement a more fine-grained measurement model to gauge individuals’ respective scores for distinct understandings of democracy. We introduce the resulting individual factor scores as independent variables into a regression model, controlling for party preference and socioeconomic variables. In essence, our research design thus assesses to what extent respondents’ understandings of democracy in the abstract translate into a willingness to punish democratic transgressions in a concrete candidate choice situation.

To measure respondents’ understandings of democracy, we revise and expand an item battery from the World Value Survey (WVS) (Haerpfer et al. Reference Haerpfer, Inglehart, Moreno, Welzel, Kizilova, Diez-Medrano, Lagos, Norris, Ponarin and Puranen2020) and implement the measurement model outlined in our preregistration, asking respondents to rate how essential they find each item to be for democracy on a scale from 1 to 7. We pretested the majoritarian and authoritarian items for internal validity and statistical benchmarks for confirmatory factor analysis in a dedicated pretest among a smaller sample of Polish respondents. Table 2 reports the retained items as included in our preregistration, while Table A.4 in the Appendix compares our items to those included in the WVS.

Table 2.

Item battery of understandings of democracy

Based on these observed items, we implement an ordered confirmatory factor analysis with three separate latent variables corresponding to a liberal, majoritarian, and authoritarian understanding of democracy.Footnote ¹⁰ We compute individual factor scores for each latent variable based on the model. Since the different understandings may be correlated with one another,Footnote ¹¹ we allow covariance between the three latent variables and choose to assess the relative strength of each respondent’s support for the three distinct understandings of democracy separately. The model indicates a good model fit ( ${\chi ^2}$ = 175.91, CFI = 0.99, RMSEA = 0.06), suggesting that our measures are internally valid (see details of measurement model in Online Appendix B.1). In Online Appendix B.3, we implement an exploratory factor analysis and find strong support for the internal validity of our measures.

To further probe the robustness of our measurement, we assess to what extent respondents’ understandings of democracy correspond to their evaluation of the democratic attributes we included in the conjoint. We asked respondents outside the actual candidate choice experiment to rate how democratic they thought each of the statements towards judges and media was (see Online Appendix B.7). The observed patterns indicate that our independent measure of respondents’ understandings of democracy maps onto their evaluation of the items we chose to include in the conjoint, making these a salient measure of the congruence between respondent-level and candidate-level democratic views.

In another test of our measurement, we examine the extent to which divergent understandings of democracy correspond to party preferences. Earlier experimental findings indicate that citizens may rationalize democratic violations when they are carried out by an actor whose policy preferences align with their own (Krishnarajan Reference Krishnarajan2023). Such behaviour is particularly prevalent in contexts of democratic backsliding, allowing citizens to alter their very perception of what is democratic or undemocratic to convince themselves that they are getting both their preferred policy and democracy. A previous study in Poland has suggested that backsliding leaders are able to maintain themselves in power precisely because their leadership style aligns with a distinct understanding of democracy among their electorate (Reykowski Reference Reykowski2020). In this scenario, partisanship would largely overlap with distinct democratic attitudes, potentially undermining our argument that these attitudes represent a distinct dimension influencing citizens’ voting choices. Besides, a recent study examining the role of incumbency shows that citizens can adjust their understanding of democracy to their partisan interest, with incumbent supporters typically more majoritarian in their orientation than opposition supporters (Bryan Reference Bryan2023). We therefore examine to what extent partisan preferences are associated with respondents’ understandings of democracy in ways that might affect their responses to democratic violations.

To do so, we investigate what percentage of variance in understandings of democracy can be explained by party preferences. If an overwhelming share of the variance in understandings of democracy were to be explained by partisan affiliation, this would provide strong support for the claim that party supporters in backsliding countries substantially differ in the notions of democracy to which they subscribe. Table 3 breaks down the variance in understandings of democracy explained by partisan affiliation. Specifically, 3.51, 2.75, and 1.48 per cent of the variance in liberal, majoritarian, and authoritarian understanding scores, respectively, can be attributed to differences in partisanship. This suggests that, although some differences in understandings of democracy are linked to party preference, a significant portion of these differences transcends partisan groups, indicating that citizens within the same party have varied understandings of democracy.

Table 3.

Variance explained by party preferences in ANOVA models for a liberal, majoritarian, and authoritarian understanding of democracy

Lastly, we consider the possibility that the Polish government party PiS disseminated a more majoritarian and authoritarian narrative of what democracy constitutes and thereby increased its supporters’ endorsement of such an understanding of democracy. To examine this possibility, we draw on a survey panel collected before and after the Polish PiS government came to power in 2015 (see Appendix C). These data, in turn, allow us to compare levels of support for different notions of democracy over time and whether PiS supporters or even the entire Polish electorate became more supportive of non-liberal understandings of democracy once this party entered power. Our results do not provide evidence that government supporters or voters in general became more supportive of majoritarian or authoritarian notions of democracy before and after PiS assumed power in 2015, suggesting that the government did not significantly shift Polish citizens’ understandings of what democracy constitutes. Although PiS may still have affected its voters’ views of democracy while in government, our analysis shows that political culture is not solely determined by the discourse and actions of political parties and can, therefore, be treated as an independent factor shaping vote choice.

The various robustness checks indicate the validity of our measurement of citizens’ understandings of democracy. On this basis, we proceed to examine to what extent these distinct democratic attitudes are related to support for non-liberal candidates.

Empirical Strategy

To assess how divergent understandings of democracy affect vote choice, we implement OLS models regressing individuals’ IMCEs for democratic attributes ( ${\hat \pi _{il}}$ ) on a vector of their understanding of democracy ( ${\bf{X}}{{\rm{'}}_i}$ ), controlling for a vector of partisanship and sociodemographic variables ( ${\bf{Z}}{{\rm{'}}_i}$ ):

(2) $${\hat \pi _{il}} = {\alpha _i} + {\bf{X}}{{\rm{'}}_i}{\beta _1} + {\bf{Z}}{{\rm{'}}_i}{\beta _2} + {\varepsilon _i}.$$

This approach allows us to evaluate the relevance of divergent democratic attitudes while controlling for party preference and socio-demographic variables. Controlling for respondents’ preferred party allows us to rule out that different partisan attachments confound the relationship between understandings of democracy and revealed democratic attitudes. Similarly, adding socio-demographic variables (age, gender, education, income, perceived economic status) helps mitigate concerns over omitted variables bias, as socioeconomic status could also feed into respondents’ level of democratic commitment as measured in the candidate experiment.

The Aggregate Relationship between Understandings of Democracy and Candidate Preferences

Our main argument holds that divergent understandings of democracy feed into citizens’ political choices in contexts of democratic backsliding. To provide an aggregate overview of the relationship between divergent understandings of democracy and choices between candidates who advance different democratic views for the entire survey electorate, we examine a subset of vote choices that pit a candidate whose democratic positions are consistent across both democratic attributes against a candidate expressing mixed views.

We first focus on choices in which respondents are confronted with one candidate with consistent liberal attributes and another candidate holding either consistent majoritarian or authoritarian views. Figure 1(a) shows the fraction of choices made for the purely liberal candidate and plots the share along with respondents’ extent of liberal understanding of democracy. The more respondents’ liberal orientation increases, the more they prefer the liberal candidate to their non-liberal contender. Substantially, from the least liberal to the most liberal respondents in the sample, we find an average increase of about 25 per cent in electoral support for liberal candidates over non-liberal candidates.

Figure 1.

The fraction of vote choices for a consistent over a non-consistent candidate at varying levels of respondents’ understandings of democracy. Generalized additive model (GAM) slopes are shown. Ribbon represents a 95 per cent confidence interval.

We find a similar but weaker pattern for choice situations where a fully authoritarian candidate runs against either a fully liberal or majoritarian candidate (Figure 1(c)): the more respondents subscribe to an authoritarian understanding, the higher the vote share for consistent authoritarian candidates, with an overall increase in electoral support of 10 per cent. However, note that the overall vote share for purely authoritarian candidates does not exceed 40 per cent, even among strongly authoritarian respondents, indicating an overall rejection of candidates expressing authoritarian positions by respondents.

A substantially weaker pattern emerges for a majoritarian understanding of democracy (Figure 1(b)). Only marginally growing with respondents’ attitude toward a majoritarian understanding of democracy, the average vote share increases from just below to just above 50 per cent, suggesting that a majoritarian understanding is substantially more weakly associated with preferences for fully majoritarian candidates than liberal and authoritarian understandings with support for congruent political candidates.

Individual-level Understandings of Democracy and Candidate Preferences

The descriptive overview provides insight into the aggregate relationship between divergent understandings and support for political candidates. To investigate the individual-level association between divergent understandings of democracy and the evaluation of candidates expressing distinct positions on liberal democratic safeguards, we turn to the analysis of individual-level candidate evaluations by regressing respondents’ IMCEs on their liberal, majoritarian, and authoritarian understandings scores. Table 4 reports our main findings.

Table 4.

OLS regression of candidate attribute preferences (IMCEs) on liberal, majoritarian, and authoritarian understandings of democracy, controlling for party preference and socioeconomic controls. Robust standard errors are reported. The full regression table can be found in Table B.5

${{\rm{\;}}^{{\rm{***}}}}p \lt 0.001$ ; ${{\rm{\;}}^{{\rm{**}}}}p \lt 0.01$ ; ${{\rm{\;}}^{\rm{*}}}p \lt 0.05$ .

The results indicate that the more respondents subscribe to a liberal understanding of democracy, the less supportive they are of candidates who make (1) majoritarian and (2) authoritarian claims about the judiciary and (3) endorse authoritarian-leaning views of government media. In turn, a higher majoritarian understanding is associated with stronger support for candidates delivering majoritarian or authoritarian statements about the appointment of judges. By contrast, a majoritarian understanding does not predict the approval of candidates proposing majoritarian or authoritarian views on the role of public media. Similarly, a higher authoritarian understanding is not positively related to respondents’ support for majoritarian or authoritarian candidates. We add party controls to our models to show that the effect of distinct understandings of democracy holds even when we account for citizens’ partisan affiliation and other party-related dynamics that we discuss further below.

Overall, our findings allow us to conclude that divergent democratic attitudes play a discrete and non-negligible role in shaping citizens’ vote choice in contexts of democratic backsliding. As shown in Figure 1, there is a consistently positive relationship between a liberal understanding of democracy and voting for liberal candidates. In other words, even a small increase in a liberal understanding of democracy at any range of the scale is associated with substantive vote share increases for liberal candidates. Divergent understandings of democracy within the citizenry appear to enable democratic backsliding primarily due to distinct levels of liberal democratic commitment among citizens, thus providing support for Hypothesis 1a.

By contrast, we find no support for our Hypothesis 1b on the overall congruence between respondents’ and candidates’ democratic views as a predictor of candidate preference. If this were the case, we should find not only that respondents expressing strongly liberal attitudes are most likely to support candidates with liberal views. In addition, respondents holding more majoritarian or authoritarian views of democracy should similarly endorse candidates expressing corresponding democratic positions, possibly even to the point of rating lower those candidates who espouse liberal democratic views. In other words, this would imply that certain voters support specific candidates, not despite the undemocratic practices they sponsor but precisely because these candidates profess views that align with their own views. However, our analysis suggests instead that respondents who endorse non-liberal understandings of democracy appear to lend less weight to candidates’ democratic views rather than actively supporting candidates who propose democratic transgressions that correspond to their understanding of democracy.

Partisan Voting and Rejecting Non-liberal Candidates

Our findings so far provide evidence that citizens’ level of commitment to liberal democracy shapes their willingness to support candidates making majoritarian and authoritarian claims. A large body of literature suggests that partisanship is a main driver of citizens’ tolerance towards violations of democratic principles by co-partisan politicians (Ahlquist et al. Reference Ahlquist, Ichino, Wittenberg and Ziblatt2018; Carey et al. Reference Carey, Clayton, Helmke, Nyhan, Sanders and Stokes2020; Graham and Svolik Reference Graham and Svolik2020). The key mechanism underpinning this argument is that voters are unwilling to switch to an out-party candidate if their own co-partisan behaves undemocratically. To examine to what extent partisans’ willingness to shift to an out-party candidate if their co-partisan candidate adopts non-liberal positions is associated with divergent understandings of democracy, we focus on a subset of choices, namely those between a co-partisan who adopts at least one non-liberal (that is, majoritarian or authoritarian) position, and who runs against a purely liberal out-party candidate. We implement the following linear probability regression:

(3) $${Y_{ij}} = {\alpha _i} + {\bf{X}}{{\rm{'}}_i}{\beta _1} + {\bf{Z}}{{\rm{'}}_i}{\beta _2} + {\varepsilon _i},$$

where $Y$ is respondent $i$ ’s preference for the non-liberal co-partisan in choice $j$ , ${\bf{X}}{{\rm{'}}_i}$ is the vector for the respondent’s understanding of democracy scores, and ${\bf{Z}}{{\rm{'}}_i}$ is a vector of socioeconomic controls as defined in Equation 2.

This analysis allows us to test whether partisans’ willingness to shift from a non-liberal co-partisan candidate to a liberal out-party candidate is associated with divergent understandings of democracy. Table 5 displays the results. Controlling for partisanship and socioeconomic variables, the more respondents subscribe to a liberal understanding of democracy, the less likely they are to vote for the non-liberal co-partisan candidate over a liberal out-party candidate. We find the reverse effect for an authoritarian understanding of democracy, but the association diminishes when jointly regressing vote preference on all understandings of democracy. A majoritarian understanding of democracy is unrelated to voting for a non-liberal co-partisan over a liberal out-party candidate.

Table 5.

Linear probability model (OLS) of voting a non-liberal co-partisan candidate over a liberal out-party candidate on liberal, majoritarian, and authoritarian understandings of democracy, controlling for party preference and socioeconomic controls. Robust standard errors clustered at the respondent level are reported

${{\rm{\;}}^{{\rm{***}}}}p \lt 0.001$ ; ${{\rm{\;}}^{{\rm{**}}}}p \lt 0.01$ ; ${{\rm{\;}}^{\rm{*}}}p \lt 0.05$

In sum, the results are consistent with our finding that divergent degrees to which respondents embrace a liberal understanding of democracy are associated with the extent to which voters reject non-liberal candidates: when respondents’ co-partisan candidate adopts a non-liberal position and runs against a liberal out-party candidate, the extent to which partisan voters abandon the non-liberal co-partisan varies with respondents’ liberal understanding of democracy.Footnote ¹²