Sympathy for the devil? Voter support for illiberal politicians

Abstract Many democracies are witnessing the rise and continuing success of parties and politicians who oppose fundamental principles of liberal democracy. Recent research finds that voters support illiberal politicians, because they trade off policy congruence against attitudes toward liberal democracy. Other studies, however, suggest that authoritarian and populist voters might actually have a preference to vote for illiberal candidates. We argue that both factors interact: Authoritarian and populist voters are more willing to trade off policy representation against support for liberal democracy. To test this mechanism, we rely on a survey experiment conducted in Germany. The results clearly demonstrate that voters indeed trade off policy congruence against liberal democracy. Moreover, this effect is particularly strong for populist and authoritarian voters. Overall, the results have important implications for understanding when and which voters support or oppose liberal democracy.


Introduction
In recent years, many countries have witnessed the election of illiberal politicians who oppose and threaten fundamental principles of liberal democracy (Levitsky and Ziblatt, 2018;Urbinati, 2019;Hellmeier et al., 2021). Why do voters support such illiberal politicians and sometimes even reelect instead of punishing them at the ballot box for their behavior in office (Frederiksen, 2022)? Apparently, adhering to principles of liberal democracy does not seem to be a precondition for voter support, and opposition against these principles does not necessarily result in the loss of votes. However, the mechanisms that underpin voters' willingness to support illiberalism are not entirely clear (Claassen, 2020).
The potential reasons for voters' support for illiberal politicians are currently subject to scholarly debate. Two explanations stand out. First, a recent strand of research suggests that voters do not necessarily have a clear preference for illiberal actors. Instead, respective studies argue that voters support politicians despite of their illiberal positions (Carey et al., 2022;Svolik, 2020;Touchton et al., 2020), for instance, by choosing partisan loyalty or policy congruence over democratic attitudes (Graham and Svolik, 2020;Fossati et al., 2022). According to this view, voters value liberal democratic principles, but they are willing to 'trade-off' this preference against other desirable characteristics of politicians. Second, the literature on populist and authoritarian attitudes argues that these sentiments are in conflict with liberal democracy (e.g., Foa and Mounk, 2016;Müller, 2016;Pappas, 2019). Therefore, voters holding such attitudes might actively prefer illiberal over liberal politicians.
In this paper, we contribute to this debate by building a bridge between these two explanations and explore the relationship between illiberal attitudes and policy congruence. Hence, instead of only treating these as two distinct mechanisms, we argue that both inform each other and form a particular behavior when it comes to the support for illiberal politicians. While we agree that voters are likely to 'trade-off' support for liberal democracy against other favorable characteristics of a politician, our argument is that populist and authoritarian voterswho both have distinct illiberal attitudesare particularly likely to do so. In other words, we expect that non-populist and non-authoritarian voters are less willing to trade-off their policy preferences against support for liberal democracy.
We test these expectations based on a survey experiment conducted in Germany in which we present the participants with a choice between a politician who supports the institutions of liberal democracy and another politician who wants to abolish them. Thus, respondents essentially make a choice between liberalism and illiberalism. As experimental treatment, we randomly vary the policy positions of the two competitors and measure their policy congruence with the respondents. The results of our study shed light on the puzzle between high levels of support for democracy, as research has shown on several occasions (Welzel, 2014;van Houwelingen and Dekker, 2021), and the ongoing success of illiberal actors.
First, we demonstrate that policy congruence trumps voters' normative support for democracy. Presented with a choice between an illiberal and a liberal candidate, we find that voters are inclined to support the illiberal over the liberal alternative if they agree more with the policy positions of the former. Second, the described relationship is strongly moderated by the level of populist and authoritarian attitudes. While all voters tend to increase their support for the illiberal politician with increasing levels of policy congruence, the increase is much stronger for populist or authoritarian voters. For these voters, illiberalism represents a candidate's feature that matches their own preferred regime principles. In this context, we find that populist and authoritarian attitudes are not correlated in our data, suggesting that both factors have indeed an independent effect, thus pointing at different forms of illiberalism leading to similar results. Finally, we demonstrate that comparable interaction effects cannot be found for left-right ideology, political interest or satisfaction with democracy. These findings indicate that 'thin' ideologies, such as populism and authoritarianism, are the relevant moderators of illiberal attitudes and not necessarily thick ideologies. 1 Overall, our findings contribute to the growing literature on citizens' support for liberal democracy and highlight when and which voters support illiberal candidates.

Theoretical expectations
Despite their ideological variety, the common defining features of illiberal parties is their hostility against core characteristics of liberal democracy, such as the rule of law, the division of powers and minority rights. Once in power, these actors have proven to execute a respective agenda by assaulting liberal institutions such as constitutional courts and the media, as examples in Poland and Hungary demonstrate (e.g., Ágh, 2016;Huber and Schimpf, 2017;Cianetti et al., 2018;Caamaño and Bértoa, 2020;Juon and Bochsler, 2020).
The electoral success of illiberal parties and politicians has been stimulating academic investigations for a while. Countless works attempt to analyze support for different forms of illiberalism, such as anti-establishment parties, populist radical right and left parties, extremist parties, and others (Zulianello, 2018). At the macro level, their rise can be explained by a variety of factors, such as economic decline (Rodrik, 2021) or cultural change in societies which provokes backlash by those who feel culturally 'left behind' (Norris and Inglehart, 2019). At the micro level, 1 Populism and authoritarianism are both thin ideologies in the sense of Freeden (1996) as their core is restricted to a certain domain and both ideologies lack 'the capacity to put forward a wide-ranging and coherent programme for the solution to crucial political questions' (Stanley, 2008, 95) on their own. explanations for voters' decisions to actively support an illiberal political actor over a liberal alternative are the subject of a vivid academic debate.
In a recent study on the USA, Graham and Svolik (2020) find that although citizens might support democracy at the attitude level, they are often not willing to act according to their commitment and 'choose democracy over partisan loyalty'. In fact, they demonstrate 'that only a small fraction of Americans prioritizes democratic principles in their electoral choices when doing so goes against their partisan identification or favorite policies' (Graham and Svolik, 2020: 406). In a similar fashion, Fossati et al., (2022) show that support for democracy depends on partisan cues. In other words, this research suggests that when voters are presented with a trade-off between political preferences (issues, partisanship) and pro-democratic values, they are likely to choose the former. Thus, they might agree with some policy positions of the party for example, on migration or taxeswithout necessarily supporting democratic illiberalism as well. For example, it has frequently been shown how voting for populist radical right parties is driven by strong issue preferences, especially regarding the issue of migration (Rooduijn, 2018;Neuner and Wratil, 2022).
Based on these works, we suppose to consider the decision to choose between two politicians as a trade-off between the conflicting positions of the politicians in two dimensions: their position on liberal democracy and their positions on political issues (Graham and Svolik, 2020;Svolik, 2020). The trade-off takes place when a voter is presented with a choice between politicians of which one is illiberal but displays high congruence on policy issues with the voter and the other is liberal but distant from the voter in terms of issue positions. Graham and Svolik (2020: 395) express this situation based on a formal model which essentially argues that the difference in the policy congruence between the illiberal and liberal candidate with the voter is key (see also Svolik, 2020). In this line of thought, because support for liberal democracy is assumed to be a valence attribute that all voters agree with, the policy congruence between the illiberal candidate and the voter has to be larger than the policy congruence between the liberal candidate and the voter. Put differently, the illiberal candidate needs a 'relative policy congruence'-advantage that compensates for the illiberal stance on democracy. The larger this relative policy congruence is, the more likely it is that it will actually compensate for the illiberal stance on democracy and a voter is willing to support this candidate. We thus formulate our first hypothesis as follows: Hypothesis 1 The higher the relative policy congruence between a voter and an illiberal politician, the more likely a voter is to support the illiberal politician.
Following the model of Graham and Svolik (2020), one would expect that voters are more likely to turn a blind eye on illiberal democratic attitudes when they receive compensation for it in terms of a higher degree of policy congruence. So far, however, the implications of the model have been discussed based on the assumption that all voters have the same attitude toward liberal democracy, that is, each voter 'punishes' illiberal candidates for their illiberal stance equally strongly. This assumption is unrealistic given the variance of attitudes toward democracy in a society. 2 While it seems reasonable to assume that the vast majority of voters actually have a preference for liberal democracy, it is also plausible to suppose that some voters are more easily willing to trade-off liberal democracy against policy congruence while some others are less willing to do so. Some voters might even actively prefer illiberal over liberal positions. This assumption has important consequences for the relationship between the policy congruence and a voter's readiness to support an illiberal candidate. Essentially, the weight a voter assigns to liberal democracyhence, their support of liberal democracyshould be a moderator of the relative policy congruence (as expressed in H1).
Consider again the situation in which one candidate is supportive and the other candidate opposes liberal democracy. Furthermore, assume that the illiberal politician has a slightly higher policy congruence with the voter than the liberal politician, i.e., a positive relative policy congruence for the illiberal politician. Whether this advantage in policy congruence is sufficient for a voter to support the illiberal candidate depends on how strongly the voter values the principles of liberal democracy. If a voter puts no weight to liberal democracy, even a tiny relative policy advantage will be sufficient for the illiberal candidate. But if a voter attaches a strong weight to it, a small advantage will not be sufficient. The implication of the model is thus that the effect of the relative policy congruence (H1) depends on how strongly a voter values liberal democracy. Therefore, the crucial question is which factors influence a voter's support for liberal democracy.
Directly measuring the weight voters attach to liberal democracy is challenging. In fact, the 'weight' that comes into play are normative orientations with regard to liberal democracy as a regime rather than an evaluation of its performance. Therefore, we suggest to discuss this aspect with regard to other attitudes that represent such normative orientations and are assumed to be at odds with liberal democracy. Specifically, we examine authoritarian and populist attitudes as factors which should decrease a voter's support for liberal democracy. Despite the variety of definitions, a core feature of authoritarian attitudes is the 'desire or tendency to impose one's own will on others' (Ray, 1976: 319). As a result, authoritarian attitudes represent a hierarchical understanding of society and hostility against those who are perceived as threats to a quasi-natural social order (Feldman, 2003). In this regard, several scholars have pointed at the importance of authoritarian beliefs for the decline of support for liberal democracy. For instance, Foa and Mounk (2016) have found that an increasing number of citizens are willingly supporting authoritarian alternatives and becoming more and more 'cynical' about liberal democracy.
A second set of attitudes that is likely to contribute to choosing illiberal over liberal candidates is populism. Populism (Mudde, 2004: 543) is most commonly defined as an 'ideology that considers society to be ultimately separated into two homogeneous and antagonistic groups, "the pure people" versus "the corrupt elite," and which argues that politics should be an expression of the volonté générale (general will) of the people.' As a consequence, populism embraces an antipluralist approach to politics (Müller, 2016;Pappas, 2019) and 'fundamentally rejects any type of limitation on the power of the majority' (Mudde, 2021: 581). Hence, citizens who believe that all politics should mirror the homogeneous will of the 'pure people' are more likely to vote for the respective parties (e.g., Akkerman et al., 2014;Van Hauwaert and Van Kessel, 2018).
Once in power, populists implement illiberal policies, based on the re-interpretation of democracy as 'radical majoritarianism' (Urbinati, 2019: 192), therefore claiming to represent the 'true' people and executing their will. This results in levelling institutions that guarantee the division of power, such as constitutional courts or the media. In this regard, populism does resemble authoritarianism, and it is no surprise that some scholars speak of 'populist authoritarianism' (Inglehart and Norris, 2017;Norris and Inglehart, 2019). This might suggest that populism and authoritarianism are essentially the same concept. Whereas we shall not dive into the details of this debate at this point, empirical research shows that at the attitude level, both are indeed different and not necessarily correlated (Schäfer, 2021). Authoritarianism is about social order, mirrored in a strict political hierarchy, but not necessarily based on the claim that the people are politically homogeneous. Populism, in turn, is based on the idea that the people are defined by their common will and does, therefore, prefer 'majoritarian and unmediated forms of political representation' (Zaslove and Meijers, 2021: 16), hence representing an illiberal alternative to liberal democracy (Mudde, 2021) that does not necessarily demand an autocratic political order. However, both are opposed to the principles of liberal democracy, hence different means to similar ends when it comes to choosing illiberal candidates over liberal alternatives. While populism and authoritarianism are different sets of attitudes, we assume that both attitudes lower a voter's support for liberal democracy. Thus, coming back to the model described above, populist and authoritarian voters should be more likely to support an illiberal candidate compared to non-populist or non-authoritarian voters. Crucially, as we argued above, these attitudes moderate the relationship between relative policy congruence and voting for an illiberal politician, that is, we expect that given the same degree of relative policy congruence, populist and authoritarian voters should be more likely to support the illiberal politician compared to non-populist and non-authoritarian voters. These expectations are our second and third hypotheses: Hypothesis 2 The effect of the relative policy congruence on support for illiberal politicians (H1) is stronger the higher the authoritarian attitudes of a voter are.
Hypothesis 3 The effect of the relative policy congruence on support for illiberal politicians (H1) is stronger the higher the populist attitudes of a voter are.

Experimental design
The hypotheses are tested based on a choice experiment conducted in Germany. The experiment largely follows the standard design of a conjoint experiment as suggested by Hainmueller et al. (2014) with some minor (but important) differences which we shall discuss later. We present respondents with the profiles of two politicians who vary with regard to different attributes. Respondents then have to choose which of the two politicians they prefer based on the presented information. In our experiment, each respondent had to make four of such comparisons. Each candidate profile consists of nine attributes (Table 1). The first two attributes are gender and age. They are included in the experiment because such information is usually known about politicians and therefore make the experimental situation more realistic. More importantly, however, we included four policy positions of the competitors on issues that are of high salience in German politics: taxation, climate protection, unemployment benefits and the regulation of migration. On each of these policies, a politician could either support the status quo or take a more left-wing/progressive or more right-wing/conservative position.
Finally, we include three attributes in the experiment that reflect a politician's position on fundamental aspects of liberal democracy. Each of these attributes has only two levels: one that reflects the status quo in Germany and is thus compatible with liberal democracy and one that opposes the idea of liberal democracy. Specifically, the first attribute concerns the selection of judges for the constitutional court. The illiberal position is that the judges should be selected directly by the government, while the status quo (liberal) position is that the judges have to be appointed by both chambers of the German parliament (Bundestag and Bundesrat). The second attribute addresses the influence of the government on public broadcasting. The illiberal position is that the government should have more rights to influence the content of public broadcasting companies while the liberal position is that the government has no such competence. Finally, the third attribute is concerned with the transfer of legislative power to the executive branch. The illiberal position is that the government can pass bills without the support of parliament while the liberal position is to maintain the legislative power of the assembly. The formulation of the attributes reflects three important considerations. First, while the illiberal positions are incompatible with liberal democracy, they do not reflect a too obvious antidemocratic attitude. In the empirical world, liberal democracy is mostly not under threat by politicians who bluntly advocate turning democracies into dictatorships, but rather by a process in which liberal norms and institutions are continuously weakened whereas the electoral regime remains widely intact (Levitsky and Way, 2002;Levitsky and Ziblatt, 2018;Svolik, 2020). Second, the illiberal positions in the experiment are backed up by real world examples, such as democratic backsliding in Poland or Hungary. Third, while the attributes on policies reflect the political discourse in Germany, they are not restricted to this context since climate protection, unemployment benefits and regulation of migration represent relevant issues in almost all European countries.
In contrast to standard conjoint experiments in political science research, we designed the experiment in such a way that one profile always contains all illiberal or liberal views on the three respective attributes. Moreover, in our design, it is not possible for both politicians to take a liberal or illiberal position at the same time. In other words, in each comparison between the two politicians, a respondent must select between a liberal and an illiberal politician. In choosing this design, we account for the question of when voters decide to support an illiberal politician even though a liberal alternative exists. Moreover, by always combining all three liberal or illiberal positions, we reduce the risk that some respondents might evaluate one of these attributes as less relevant than the others. It also makes the experiment more realistic since it is rather unlikely that a politician takes a liberal position on one attribute and an illiberal position on a different one. In most empirical cases, illiberal positions tend to cluster.
This also implies that we do not follow the standard approach in conjoint experiments and do not estimate Average Marginal Component Effects (AMCE; Hainmueller et al., 2014) or Marginal Means (MMs; Leeper et al., 2020). Instead, we estimate logistic regressions in which the dependent variable is 1 if a respondent selected the illiberal politician and 0 otherwise. As respondents conducted four comparisons, we have four observations for each respondent and use clustered standard errors to account for the non-independence of the observations.

Relative policy congruence
The main independent variable of interest is the relative policy congruence between the respondent and the politicians' profiles. We estimate this variable as follows.
In the pretreatment questionnaire, we retrieved a respondent's position on the exact same four policy issues that were also displayed in the candidate choice experiment. We also offered them the three respective levels as the response scale (left-wing position, status quo, or right-wing position). This allows us to estimate the policy distance between a respondent and each of the randomly created profiles of the politicians (compare Graham and Svolik, 2020). We denote this distance between a voter (i) and a candidate profile (p) on a certain issue (k) as d i;p;k . If a politician and a respondent have the same position, we assign a value of 0; if one of them supports the status quo and the other takes a left-or right-wing position, we assign a value of 1. When a respondent and a politician take directly opposed positions on an issue, we assign a value of 2. We then estimate the 'policy advantage' of the illiberal candidate (p = IL) over the liberal candidate (p = L) for each issue as pa i;k d i;IL;k d i;L;k . The value of pa i;k is positive when the policy distance between the liberal candidate and the voter is larger than the policy distance between the illiberal candidate and the voter on a certain issue. Respectively, it takes a negative value when the liberal candidate is closer to the voter on the issue.
Before aggregating the values of pa i;k into a single score, we first consider how significant an issue is to the voter. To do so, we have asked respondents how important each of the four issues is to them on a scale from 0 ('not important at all') to 3 ('very important'). We then create weights for each issue by dividing the respective issue importance by the sum of all issue importance scores. We denote these weights as w i;k . For example, if a voter does not care about taxes and climate but finds migration and unemployment subsidies very important, the first two issues receive a weight of 0 and the other two a value of 0.5.
The final relative weighted policy advantage (rwpa i ) of the illiberal candidate is then estimated by multiplying the issue specific policy advantage of the illiberal candidate pa i;k with the respective weight and then summing all these values up: This variable can take values between −2 and 2. A value of 2 indicates that the illiberal candidate has the highest possible policy advantage over the liberal candidate, while a value of −2 describes the opposite situation. The value of 0 denotes a scenario in which both candidates have the same policy distance to the voter.

Populism, authoritarianism and controls
In line with the hypotheses, we measured the populist and authoritarian attitudes of the respondents. Populist attitudes are retrieved based on the scale developed by Akkerman et al. (2014). Authoritarianism, in contrast, is measured based on three items taken from Nießen et al., (2020), indicating the degree of willingness to submit to strong leadership. Both scales have a range between 1 (i.e., minimum level of populism/authoritarianism) and 5 (maximum level of populism/authoritarianism). In addition, we include gender, age, left-right self-placement, political interest, and party choice in the analysis. Regarding the interpretation of the effects, it is important to note that the relative policy advantage allows for a causal interpretation due to the randomization of the policy positions between the profiles. In contrast, the other independent variables do not necessarily have a causal interpretation. Instead, they capture the probability of selecting an illiberal politician while controlling for other observable characteristics.

Sample
We conducted the experiment in June 2021 with a sample of 999 German respondents. Germany is a particularly interesting context for our study. On the one hand, it is a stable democracy both in terms of its institutions and the widespread presence of pro-democratic attitudes. This might suggest that support for illiberal attitudes should be rather low in Germany. On the other hand, Germany has witnessed the rise of a populist radical right party in the Alternative for Germany (AfD). The AfD's electoral success can be traced back to both policy positions (first and foremost on the immigration issue) as well as populist (Steiner and Landwehr, 2018) and antidemocratic attitudes (Donovan, 2019: 458-459). On the other side of the political spectrum, The Left is an established populist radical left party (Hough and Koß, 2009). 3 Although located at opposite ends of the spectrum, voters of The Left, similar to those of the AfD, are dissatisfied with democracy and have a populist profile (Olsen, 2018). Therefore, the German context provides two populist parties from different ends of the spectrum and an electorate with respective sentiments. At the same time, both the party system and the support for democracy have remained stable over time. Therefore, besides these parties, Germany might be seen as a 'least likely' case for the support of 3 It should be noted that the degree of populism of The Left is substantially lower compared to the AfD. For example, the established POPPA populism scores assign a populism value of 5.55 to The Left, while it is 9.43 for the AfD ). Yet, the value for The Left is still high enough to describe it as a populist party and research on left-wing populism usually treats The Left as a populist party.
illiberal candidates. While we would expect it to be rather common among the supporters of the two populist parties, we assume it is presumably lower among all other voters if our hypotheses are false.
The experiment was embedded in an online survey; the sample was provided by the internationally operating access sample provider respondi. 4 We used sampling quotas for vote choice in order to guarantee that the resulting data included enough respondents from the fringes of the political spectrum. Specifically, the sample consists of 33% respondents who indicate voting for the populist radical right AfD, 33% respondents who would vote for the populist left-wing party The Left (Die Linke) and 33% respondents stating that they would vote for one of the other nonpopulist established parties (CDU/CSU, SPD, Greens, FDP). This sampling strategy is chosen to ensure that we observe a sufficient number of respondents with rather radical attitudes. However, at the same time, the strategy makes our sample less representative. This is not a major drawback for our study as we are primarily interested in the causal effect of policy congruence and the impact of certain attributes. Moreover, we control for vote choice in the regression models so that clear differences between supporters of different parties should become visible. No other quotas were used for sampling. A more detailed description of the sample is given in Table A2 in the online appendix to this paper.

Results
We display the effect of the relative weighted policy advantage of the illiberal candidate over the liberal candidate in Figure 1. The x-axis displays the 'relative weighted policy advantage' of the illiberal politician over the liberal politician. Thus, positive values on the x-axis denote the Figure 1. Effect of relative weighted policy advantage of illiberal candidate on probability to select the illiberal candidate.
Note: Figure displays the predicted probabilities of supporting the illiberal candidate conditional on the relative weighted policy advantage of the illiberal candidate over the liberal candidate (solid purple line). Shaded areas are 95% confidence intervals. Full regression results are depicted in Table A1 in the online appendix. Visualization is based on Model 1. situation in which the illiberal candidate has an advantage over the liberal candidate and negative values indicate the opposite scenario. At 0, both candidates are equally distant to the voter in terms of weighted policy preferences. On the y-axis, the figure displays the predicted probabilities that a voter supports the illiberal candidate. We derived these probabilities (solid purple line) and the respective 95% confidence intervals (shaded purple area) using the procedure suggested by King et al. (2000).
The results clearly demonstrate that policy congruence matters. When both candidates have the same distance to the voter, that is, at 0 on the x-axis, the probability of supporting the illiberal candidate is below 0.25. This indicates that voters indeed punish illiberalism: three out of four respondents prefer the liberal democratic candidate over the illiberal candidate when the respective policy distance is equal. In line with this finding, the probability of selecting the illiberal candidate decreases toward zero the higher the advantage of the liberal candidate is over the illiberal candidate. Again, this implies that a majority of voters prefers to vote for candidates who support liberal democracy. However, we also find a strong increase in the probability to support the illiberal candidate when this candidate has a substantial policy advantage. The predicted probability increases from around 0.2 to almost 0.8 in the right half of the plot. In other words, when policy congruence is perfect with the illiberal candidate and the liberal alternative is fully distant from the voter, then four out of five respondents choose to support the illiberal candidate. Overall, these findings lend strong support for our Hypothesis 1 that voters are willing to trade off support for liberal democracy against policy congruence.
In the next step, we investigate whether populist or authoritarian attitudes interact with the relative weighted policy congruence, as expressed in Hypotheses 2 and 3. Figure 2 displays the effect for low and high levels of populist attitudes, Figure 3 for authoritarian attitudes respectively.  Table A1 in the appendix. Visualization is based on Model 2.
In both cases, we find strong support for an interaction effect. 5 While all respondents seem to agree not to select illiberal politicians when the relative policy congruence is low, clear differences between populist and non-populist, as well as authoritarian and non-authoritarian voters, become visible when relative policy congruence is high. For example, when both candidates have the same level of congruence with the respondent, the predicted probability of non-populist or nonauthoritarian voters to choose the illiberal politician is approx. 0.1. In contrast, the probability for populist or authoritarian respondents to select the illiberal politician is around 0.4 in this situation. When the illiberal politician has a clear advantage over the liberal candidate, the probability to prefer the illiberal politician increases to over 0.85 for populist or authoritarian respondents, but it remains substantially lower for non-populist or non-authoritarian voters. Overall, these findings demonstrate that populist and authoritarian voters are especially likely to select politicians who oppose liberal democracy. However, it should be noticed that even respondents with minimum levels of populism and authoritarianism have a probability of 0.4 to 0.6 of selecting an illiberal politician when policy congruence is high. Thus, even such voters are not immune to supporting illiberal politicians.

Probing deeper: populism and authoritarianism
Because the results of populism and authoritarianism are so similar, one might suspect that they measure the same underlying attitude, even if existing research indicates otherwise (Schäfer, 2021). Thus, we suggest to analyze the relationship between both concepts in more detail.  Table A1 in the appendix. Visualization is based on Model 3.

5
Notice that the interaction term in Table A1 is not significant. However, as described in the literature on interaction effects in more detail, a nonsignificant interaction term does not necessarily indicate the lack of an interaction effect. As, for example, Brambor et al. First, we provide empirical evidence that authoritarianism and populism measure different attitudes in our data, that is, both variables are not strongly correlated. In fact, Pearson's correlation coefficient is 0.024. Second, based on the observation that both sets of attitudes appear to be uncorrelated, one might argue that respondents who are populist and authoritarian are most likely to support illiberal candidates, whereas a non-populist (or non-authoritarian) attitude might decrease the impact of an strong authoritarian (or populist) attitude. In other words, only a fully illiberal respondent might have a preference for illiberal candidates, while a high value on one of both variables cannot compensate for low values in the other variable (comparable to the argument presented in Wuttke et al., 2020). To test this mechanism, we interact the relative weighted policy advantage with populism and authoritarianism. Running a three-way interaction is asking a lot from the data and therefore the analysis should be taken with a grain of salt.
The results are displayed in Figure 4. The left panel displays that when populism is low, authoritarianism does not seem to have a strong effect. While the high degree of uncertaintyexpressed through the large range of simulated predicted valueshas to be acknowledged, there does not appear to be a clear difference between non-authoritarian and authoritarian respondents when populism is low. When populism is high, the results are different. In this case, authoritarianism clearly matters. For example, respondents who are both strongly populist and authoritarian have a probability of supporting the illiberal candidate in 50% of the cases in which both candidates have the same distance to the respondent. For populist but non-authoritarian respondents this value is substantially lower and below 0.25. When the illiberal candidate has a policy advantage over the liberal candidate, the probability of supporting the illiberal candidate converges toward 1 for populist and authoritarian voters. Yet, the probability also increases strongly for the nonauthoritarian populists. Overall, these results indicate that the combination of populist and authoritarian attitudes is indeed a particularly strong predictor for supporting an illiberal candidate.

Effect of other covariates
Moreover, one might wonder whether similar interaction effects can be found between the relative weighted policy congruence and other respondent characteristics. The results of the previous sections might appear less interesting if one finds that other variables show a similar effect, that is, the effect of the policy advantage variable might vary also with regard to the level of other variables. To test for this possibility, we analyze the interaction effect of left-right ideology, vote choice as well as political interest and satisfaction with democracy with the weighted policy advantage variable.

Left-right ideology
Given that particularly radical right parties often threaten liberal democracy (Huber and Schimpf, 2017), one could suspect that left-right attitudes might have an effect similar to authoritarian attitudes. Thus, we run additional interactions in which the relative weighted policy advantage is interacted with the left-right placement of respondents. Specifically, we run two such analyses: One in which the left-right placement is included as a single term and a second one in which left-right placement is additionally interacted with itself to allow for a nonlinear relationship in which respondents at both ends of the political spectrum become more illiberal.
The results of these analyses are displayed in Figures 5 and 6. Figure 5 indicates no interaction between the variables. For left-wing, right-wing, and centrist respondents, the curves look very similar. When using the squared term for the left-right placement, we also find no evidence that the ends of the political spectrum are especially likely to support illiberal candidates. However, under this model specification, a stronger increase for right-wing respondents can be observed, but given the complexity of the specification, this pattern should not be overemphasized. In other words, whether voters choose an illiberal over an liberal politician due to the former's policy advantage does not depend on the voter's ideological position. This means that the readiness to vote for an illiberal politician applies to voters across the spectrum, not only a specific segment.

Vote choice
Similar to the analysis of left-right, one might also suspect that party preferences moderate the observed effect. Supporters of populist parties might be more inclined to support an illiberal  Table A1 in the appendix. Visualization is based on Model 5. candidate compared to voters of non-populist parties. To control for this aspect, we also run analyses in which we interact vote choice with the relative weighted policy advantage. We assign voters to three groups of respondents: (1) voters of the populist radical right AfD, (2) voters of the populist radical left The Left, and (3) voters of non-populist mainstream parties. We do so because we deliberately oversampled voters from radical parties (see above). In the online appendix (see Section C), we also display the results in which we run the interaction using the raw vote choice (i.e., without grouping voters of non-populist parties in one category). The results are displayed in Figure 7. They indicate that there are no major differences between radical party supporters and supporters of mainstream parties. For example, for all groups of voters, the predicted probability of supporting the illiberal candidate is approximately 0.25 when the policy distance is equal for both candidates. However, when the illiberal candidate has a clear policy advantage, then the  Table A1 in the appendix. Visualization is based on Model 6.  Table A1 in the appendix. Visualization is based on Model 9. radical left and radical right voters are slightly more likely to support the illiberal candidate. The observed effects are, however, substantially weaker compared to the differences observed in the interaction with populism or authoritarianism.

Political interest
We further analyze whether political interest is a moderator of the policy distance effect. As understanding the details of politics is a complex issue and many voters have not much interest in politics, it could be expected that voters with lower levels of political interest do not comprehend the negative consequences for democracy when politicians hold illiberal attitudes. If this is the case, voters with low levels of political interest might be more inclined to vote for an illiberal politician. As can be seen in Figure 8, this assumption does not find empirical support. The effect of the policy distance is identical among respondents with low, moderate, and high levels of political interest.

Satisfaction with democracy
Finally, we run interaction effects between the level of satisfaction with democracy in Germany and the relative weighted policy advantage. The rationale for this analysis is that voters who are dissatisfied with democracy might prefer a more illiberal candidate out of protest, because this candidate is likely to express their dissatisfaction with the democratic status quo. Support for democracy is measured on a five-point scale ranging from 1 ('very satisfied') to 5 ('very dissatisfied') and asks respondents specifically about their level of satisfaction with the state of democracy in Germany. Figure 9 presents the results. They do not indicate a significant interaction between the two variables. Respondents who are dissatisfied or satisfied with the democracy in Germany react quite similarly to the policy advantage of the illiberal candidate. If anything, respondents who are satisfied with the level of democracy are more likely to vote for the illiberal candidate, but these effects are small and nonsignificant. In light of our previous findings on populist and authoritarian attitudes, this supports our argument that choosing the illiberal over the liberal candidate is a substantial choice rather than mere protest.  Table A1 in the appendix. Visualization is based on Model 7.

Conclusion
The research question of this article was straightforward: when do voters support illiberal politicians? Following recent debates about the effect of both policy congruence as well as populist and authoritarian attitudes, we have argued that populist and authoritarian voters are particularly likely to support illiberal politicians when policy congruence is sufficiently high. To test these hypotheses, we conducted a survey experiment with German voters in which we confronted the respondents with a choice between liberal and illiberal politicians and systematically varied the degree of policy congruence between the voter and the liberal and illiberal politician.
Based on the findings from our experiment, we can answer our questions as follows: The selection of illiberal politicians is strongly driven by policy congruence. Once policy congruence is high, many voters are likely to support illiberal politicians. This effect, however, depends on the presence of a viable alternative. When the liberal politician has a similar or higher level of policy congruence, voters are rather unlikely to support the illiberal competitor. This finding is important for two reasons. First, it demonstrates that voters are indeed rather making a trade-off between policy positions and attitudes toward liberal democracy instead of having a clear preference for supporting the illiberal politician. Second, the finding is important as it demonstrates that voters are willing to support a liberal politician unless the illiberal alternative clearly outweighs the former with regard to their policy supply. Given that many illiberal politicians often come from radical parties which are located at the fringes of the ideological spectrum (Rooduijn and Akkerman, 2017), a potential dilemma emerges. Because parties at the margins are usually rather isolated with regard to their policy positions, their voters might feel pressured to support an illiberal party or politician to achieve their policy goals despite not being willing to abandon liberal democracy. These findings also suggest that less radical parties could suffer more strongly from taking illiberal positions, because their voters might consider more parties as viable alternatives than voters of radical parties. This applies particularly to multiparty systems.  Table A1 in the appendix. Visualization is based on Model 8. Furthermore, our results demonstrate that authoritarian and populist attitudes contribute to supporting illiberal politicians, indicating at least some demand for illiberal supply among these voters (Foa and Mounk, 2016). As already highlighted in the discussion of the results, this has important implications when contrasted with the finding that party preferences are not a meaningful predictor for the selection of illiberal politicians. Apparently, it is not the voters of particular parties who are more likely to support illiberal politicians, but rather individual attitudes toward democracy are decisive. In this regard, we have shown that illiberal politicians are preferred by voters with populist, and those with authoritarian, attitudes. Considering the similarity of the results for each set of attitudes, we are presented by distinct forms of illiberalism that lead to similar results when it comes to choosing respective politicians. At the same time, we were able to identify somewhat like a 'populist-authoritarian' attitude, which represents the strongest predictor to choose illiberal over liberal politicians when the former has a policy advantage over the latter.
Naturally, our study has several limitations. One is that our experiment does not take partisanship into account as it is the case in some previous studies (e.g., Carey et al., 2022;Graham and Svolik, 2020). Especially in highly polarized systems, such as the USA, partisanship might be at least equally relevant compared to policy congruence. Another natural limitation is that the study focuses on Germany, a case in which illiberal parties are comparatively weak. Results might be different when the same experiment is conducted in a country such as Poland or Hungary where illiberal parties represent the mainstream in the respective system and democratic backsliding has already taken place.
Despite such limitations, our paper provides important insights for our understanding of voters' tendency to select illiberal politicians. Overall, our findings are concerning. They demonstrate that support for fundamental principles of liberal democracy appears to be conditional, perhaps even fragile across the board. Even if policy congruence seems to represent an aisle for liberal actors by closing the policy-related gap and thus pulling voters away from illiberal politicians, this is rather unlikely to happen: While parties might attract some voters by shifting their policy positions, such a maneuver could also backfire through losing others (Chou et al., 2021). There is also no guarantee that such a policy shift toward illiberal parties reduces their electoral success (Krause et al., 2022;Lewandowsky and Wagner, 2022). Therefore, based on our results, it would be premature to conclude that democracy can be strengthened by mainstream parties adopting the position of radical and illiberal competitors .