Introduction

Despite the obvious negative consequences of political scandals on citizens’ trust in politics and the quality of democracy in general, existing research demonstrates that misbehavior of politicians has often only small or even no effects on voters’ electoral decisions and candidate evaluation (De Vries and Solaz 2017). While some studies find that politicians involved in a scandal are punished by voters (e.g., Funk 1996; Tumber and Waisbord 2004; Maier 2011), others show that corrupt politicians are frequently re-elected (e.g., Chang et al. 2010; Bågenholm 2013; Fernández-Vázquez et al. 2016). The potential reasons for the (re-)election of scandalous politicians are numerous and can range from lacking information on the conducted misbehavior (Chang et al. 2010) to an otherwise good performance of the politician which might compensate for the wrongdoings (Muñoz et al. 2016).

In this study, we aim to shed light on a less extensively studied aspect of electoral behavior after a political scandal, that is, the impact of viable alternatives on voter behavior. As De Vries and Solaz (2017, p. 400) argue “[t]he viability of alternatives is a core factor that conditions voters’ ability to switch party choice” when the otherwise preferred politician is involved in a corruption scandal. While the answer to the question of what constitutes a viable alternative might be highly subjective, there is general agreement that it primarily refers to political alternatives that are evaluated positively by a voter (e.g., Agerberg 2020). At the heart of the ‘viable alternative’-hypothesis lies the assumption that voters are willing to turn a blind eye on corruption if the alternative is to vote for a politician strongly disliked by the voter. In a nutshell, factors such as partisanship and ideological proximity can outweigh a voter’s distaste for corruption (Anduiza et al. 2013; Charron and Bågenholm 2016; Solaz et al. 2019).

The contribution of this paper is threefold. First, we put the ‘viable alternative hypothesis’ to an empirical test based on a survey experiment conducted in Germany. We thus contribute to the emerging literature on this topic (Agerberg 2020). Based on the argument developed in Charron and Bågenholm (2016) as well as Agerberg (2020), we argue that the ‘viable alternative’-argument implies that politicians of radical parties are more likely to get away with misbehavior than politicians of mainstream parties. In other words, we assume that supporters of radical parties are more likely than voters of mainstream parties to support corrupt candidates (Charron and Bågenholm 2016). The main reason for this argument is that voters of radical parties are more probable to hold negative views on most parties (and the ‘political class’ in general) and thus to have fewer ‘viable alternatives’ compared to supporters of mainstream parties. This argument is closely related to spatial models of party competition. For mainstream parties, the distance to other mainstream parties is usually much smaller compared to the distance of radical parties to mainstream parties. It is therefore less challenging for voters of mainstream parties to find an alternative (mainstream) party which also largely reflects a voter’s policy preferences. Based on these considerations, we expect that supporters of radical parties more frequently lack viable alternatives because not supporting the radical party means to support a party that is ideologically quite distant from the voter. Due to this isolation in the policy space, we expect that radical voters are less likely to punish corruption as it would imply that they have to support a candidate from a party that they probably dislike.

The second contribution of our paper is to show that voters of radical parties are not inherently more prone to turn a blind eye on corruption. A problem of the viable alternative hypothesis is that it is challenging to distinguish if radical voters are more likely to support corrupt candidates because radical voters care less about corruption or because they find no competing candidate acceptable. Our argument is that supporters of radical parties are just as likely as other voters to shift away from corrupt politicians if they are presented with an alternative which they find acceptable. Likewise, we also expect mainstream voters to continue to support corrupt candidates when they lack a viable alternative, e.g., when a corrupt mainstream candidate competes against a clean radical candidate.

Finally, we also analyze whether the strength of voters’ partisan attachment affects their willingness to switch their vote away from a corrupt politician. Following Eggers (2014), we argue that strong partisans have a higher likelihood to support their preferred party’s candidate even if she/he has conducted misbehavior. Weak partisans, in contrast, might be willing to cast their vote for another party’s candidate.

We test these expectations based on a vignette experiment conducted in Germany with 1000 voters. As the aim of this paper is to shed light on the candidate choice of radical voters compared to those of mainstream party voters, we make use of a sample in which radical left and radical right voters are deliberately oversampled. While one-third of the 1000 respondents are voters of mainstream parties, the other two-thirds are voters of the radical left or radical right. As radical parties are often less successful than mainstream parties, their voters are often underrepresented in general population surveys. Based on our sampling strategy, however, the preferences of radical party voters can be analyzed in more detail. This oversampling strategy also differentiates our study from the existing literature. Previous studies used representative samples (Agerberg 2020; Charron and Bågenholm 2016) and were hence potentially restricted in predicting radical voters’ behavior in the context of scandals and viable alternatives.

Our findings provide support for the viable alternative hypothesis. In general, respondents are more likely to support a corrupt politician when the alternative candidate comes from a party which they strongly oppose. This pattern is, as expected, particularly pronounced for supporters of radical right parties. However, once we take the different evaluation of alternative parties by radical and non-radical party supporters into account, these differences vanish. In other words, if supporters of radical parties evaluate a different party favorably—which happens less frequently than for supporters of non-radical parties—they are just as likely as non-radical voters to switch their vote away from a corrupt politician. Quite similar, we find that voters of mainstream parties are willing to support corrupt candidates when the alternative is to vote for a non-corrupt but radical party. Additionally, we also find evidence for our assumption that strong partisans have a higher likelihood of supporting a corrupt politician. Voters with weaker partisan attachments are more willing to withdraw the vote from their preferred candidate in case of corrupt behavior. Overall, these findings enhance our understanding of the ‘viable alternative’-hypothesis in corruption and electoral research and also how radical and mainstream voters do (not) differ in the evaluation of corruption scandals.

Corruption scandals and voting behavior

It is often expected that the career of a politician is negatively affected when being involved in a corruption scandal. Anecdotal evidence seems to support this expectation. For instance, in 2002, the Senator of Finance of the German city state of Berlin, Gregor Gysi, resigned after him taking advantage of bonus miles became public (“Flugmeilenaffäre”). Another example is linked to the former German Minister of Defense Karl-Theodor zu Guttenberg, who was forced to step down in 2011 due to massive plagiarism in his doctoral thesis. Furthermore, one of the most prominent examples of a corruption scandal in the last years in Europe is the ‘Ibiza Scandal.’ A secretly recorded video shows the former vice chancellor of Austria and then-leader of the radical right Freedom Party (FPÖ), Heinz-Christian Strache, speaking about his intentions to circumvent party funding laws. The publication of the video led to the breakdown of the ÖVP/FPÖ governmental coalition in 2019.

However, while some politicians resign from office after corruption allegations, many others remain in their position and even re-run for office in the next election. In such cases, voters are able to punish the corrupt behavior of politicians by not casting their ballot for the corrupt candidate. Particularly in times in which voters have become more critical toward the political elite (Maier 2011) and their trust in political institutions decreases (Bowler and Karp 2004), the electoral consequences for politicians that are involved in scandals could be severe. However, existing research only finds mixed empirical evidence for this expectation and demonstrates that corrupt politicians are often even re-elected (e.g., Chang et al. 2010; Bågenholm 2013; Praino et al. 2013; Fernández-Vázquez et al. 2016; De Vries and Solaz 2017). Although free and fair elections provide voters with the power to hold corrupt politicians accountable for their misbehavior, voters not always seem to do so.

According to De Vries and Solaz (2017), three conditions have to be met for a corruption scandal to have an effect on electoral behavior. First, voters need a certain degree of information on corrupt behavior of politicians. In case of misinformation or lack of information, they are not able to punish politicians electorally. Second, voters have to acknowledge politicians’ behavior as misbehavior and attribute blame correctly (De Vries and Solaz 2017). Third, even when voters know about corrupt behavior and even when they correctly attribute the blame for it, voters must be willing to switch their vote away from the corrupt politician, i.e., voters need to show a behavioral response. It is precisely this last step of the causal model developed by De Vries and Solaz (2017) that we are interested in: Why do voters support a corrupt politician even when they know about the corruption scandal and correctly attribute the responsibility to that politician?

Party cues and viable alternatives

Ongoing electoral support for corrupt politicians is often explained by a voter’s degree of partisanship. Party cues do not only matter for voters’ issue positions and opinions on certain policies but also for their candidate choice (Campbell et al. 1960; Zaller 1992; Bartels 2000; Brader and Tucker 2012). Voters thus are most likely to cast their ballot for their preferred party, as they know this party best represents their interests and views on certain issues. Moreover, voters also rely on and trust information on issues given by their preferred party. With regard to corruption, this also implies that voters are often blind toward misbehavior vis-à-vis their preferred party’s politicians. For example, Anduiza et al. (2013) show that voters judge misconduct by politicians of their in-group (i.e., their preferred party) as less severe than misbehavior of candidates from other parties. Further, Ecker et al. (2016) demonstrate that partisans are less likely to react to short-term events, such as corruption allegations, when evaluating politicians.

In many cases, however, it is well known to voters that the politician of the most preferred party is involved in a scandal. Even under such conditions, not all voters are likely to switch their vote away from this candidate. The ‘viable alternative hypothesis’ suggests that in such situations the voters’ willingness to switch votes depends on whom the corrupt politician competes against (Charron and Bågenholm 2016). The argument rests on the assumption that voters dislike corrupt behavior and are thus generally willing not to vote for the corrupt politician (Agerberg 2020). However, voters do not only care about the integrity of a politician, they are also concerned about other characteristics among which the partisanship of a politician is one of the most important factors. Thus, voters probably consider information about corruption in their decision-making process, but it is unlikely to be the only relevant information that matters for them (Incerti 2020). Instead, voters face a trade-off in which they are forced to either vote for a corrupt politician who possesses other characteristics which they prefer, such as ideological proximity or partisanship, or they vote for a politician who is not involved in a corruption scandal but might not represent their preferred party. What follows from this consideration is that the probability to turn a blind eye on corruption should be a function of how close a voter feels to the other candidate (Agerberg 2020).

The impact of radicalness

While the impact of the ideological proximity of a voter and candidate on the probability to select an alternative in elections has been comprehensively demonstrated by Agerberg (2020), we are particularly interested in explaining the behavior of voters from the fringes of the political spectrum. Since we know that the ideological distance of a voter to the clean candidate is important, it is also crucial to acknowledge that not all voters feel equally close to other parties. Thus, how ‘viable’ an alternative is depends on the ideological position of a voter (Charron and Bågenholm 2016). It is more likely that voters with more mainstream ideological positions, and thus preferences for parties located in the center of the political spectrum, possess more viable alternatives compared to voters with more radical ideological positions. Voters of such parties are often ideologically less distant from other parties on the political spectrum and thus have more alternatives that fit their preferences.

In contrast, we assume that voters of parties located at the left and right end of the political spectrum are less likely to abandon their party and to put their policy preferences over corrupt behavior of candidates. Voters on the fringes of the political spectrum are less likely to find a viable alternative to their preferred party choice, as there are less parties located on the right and left fringes (Charron and Bågenholm 2016; Agerberg 2020). We therefore expect that voters of radical right and radical left parties have a lower probability to vote for a clean alternative candidate than voters of more mainstream parties. Our first hypothesis is as follows:

Hypothesis 1

Voters of radical right or radical left parties are more likely to support the corrupt candidate from their preferred party than voters of mainstream parties.

Hypothesis 1 rests on the assumption that voters of radical parties are more likely to support corrupt politicians due to a lack of viable alternatives. However, an alternative explanation that could lead to observationally equivalent results would be that voters of radical parties are inherently more willing to support corrupt politicians. To distinguish between these two mechanisms, we suggest to take a voter’s evaluation of potential alternatives into account. Our argument is that voters of radical parties are not necessarily more likely to vote for corrupt candidates because they find corruption more acceptable. In contrast, we assume that all voters equally oppose corruption among politicians. We thus expect mainstream party and radical party voters to have the same probability of voting for the clean alternative if both voters have the same evaluation of the alternative candidate. Crucially, the same logic applies to voters from mainstream parties. We also expect mainstream voters to avoid voting for a clean alternative when this alternative candidate comes from a party that a voter strongly dislikes. In other words, mainstream voters have a similar negative evaluation of corruption as voters of radical voters. We therefore only expect radical party supporters to be more willing to support a corrupt candidate (see Hypothesis 1) because radical parties have, almost by definition, fewer competing parties that could provide comparable positions. If, however, a radical party supporter finds a competing party acceptable, this voter should be equally likely as a non-radical voter to support this alternative. Our second hypothesis thus reads as follows:

Hypothesis 2

Voters of radical and mainstream parties are equally likely to vote for a corrupt politician when they evaluate the clean alternative candidate similarly.

Strength of partisanship

Finally, whether voters are likely to continue the support for corrupt politicians should be a function of the strength of their party preferences (Eggers 2014). Existing studies on the impact of partisanship in general suggest that voters with strong partisan attachments cast their ballot for their preferred party even though certain policy positions do not perfectly overlap with their preferred party’s position (Ezrow et al. 2014; Zaller 1992). With regard to corruption, Eggers (2014) demonstrates that politicians suffer less from corruption scandals among those voters who have strong partisan attachments. Moreover, only partisans with weaker party preferences are likely to cast their ballot for another than the preferred party. This also means that voters with weak partisan attachments are more open to vote for another candidate that is not corrupt compared to voters with strong partisanship. These voters are less likely to withdraw their vote from their preferred but corrupt politician (Eggers 2014). However, also voters with weak partisan attachments must have a certain preference for another candidate’s party and have to evaluate it at least as somewhat positively. Only if this applies, weak partisans should be willing to cast their vote for any other party. Considering both the viability of an alternative and the strength of partisanship as important predictors for vote choice after corruption occurs, we expect that strong partisans are less likely to switch their vote away to an alternative candidate and that they remain to vote for their preferred, but corrupt politician. Thus, our third hypothesis is as follows:

Hypothesis 3

Strong partisans are less likely to switch away their vote from their preferred party’s candidate compared to weak partisans.

Study design

Sample and case selection

The data used in this study were collected in an online survey fielded in Germany in June 2021. The sample was provided by the survey provider respondi. We use sampling quotas based on the vote intention of the respondents in order to create a sample that includes a sufficient number of respondents with vote intentions for the radical parties on the fringes of the political spectrum, which is necessary and advisable due to the study’s focus on voters from radical parties. Hence, our sample consists of 33% supporters of the radical left-wing party ‘Die Linke,’ 33% of supporters of the radical right-wing party ‘Alternative für Deutschland (AfD),’ and 33% respondents supporting one of the non-populist mainstream parties in Germany, i.e., the conservative CDU/CSU, social democratic SPD, the market-liberal FDP, or the Greens. While this provides us with a sample that is not fully representative of the German electorate, this strategy allows us to avoid under-sampling of radical voters which would be the case with a random sample. This is particularly important for this study, as we are mainly interested in analyzing radical voters’ electoral behavior. Moreover, this sampling strategy allows us to compare the candidate choice of radical voters to those of mainstream voters more explicitly.

We decided to focus on the case of Germany due to several reasons. First, the German party system consists of six relevant parties that are currently represented in the German Bundestag, with two of these parties being radical ones. On the left end of the political spectrum, the Left Party (‘Die Linke’) is located. The Left Party mainly focuses on economic inequalities and has shifted its position toward the utmost left end of the political spectrum during the election in 2021 (Debus 2021). In contrast, the radical right ‘Alternative for Germany (AfD)’ is placed on the right end of the spectrum. The AfD was founded in 2013 as a Eurosceptic party. During its first years, the party fully turned into a populist radical right party. The party’s platform mainly focuses on anti-immigration policies (Berbuir et al. 2015; Jankowski et al. 2017). Both parties therefore represent “polar opposites in terms of ideology and policy prescriptions,” and Left party voters have deflected to the AfD (instead of mainstream parties) in several elections (Olsen 2018, p. 71). Furthermore, both parties are isolated from the political mainstream, although to slightly different extents. The Left embraces a radical, partly extremist profile and has been met with suspicion and reluctance from the other political parties since it came into existence. However, they have never been as stigmatized as any radical right party. While both parties are outsiders in the German party system, we account for different evaluations of the respective parties both in the appendix (see Figure 5) as well as in our analysis. In addition to these two radical contestors, a number of German parties are located more in the center of the political spectrum. These include the two largest parties CDU/CSU (conservative) and the SPD (social-democratic) as well as the Green Party (ecologist) and the FDP (market-liberal).

Second, as previously outlined, in Germany it is rather common that politicians accused of corrupt behavior step down and resign from their office. While Germany in general can be considered as a low corruption countryFootnote 1, from time to time wide-ranging corruption scandals occur. Hence, voters are confronted with corruption scandals of politicians, but also know at the same time that these politicians most likely will resign from office. This implies that especially in the German case, voters should have a strong distaste for corruption and that they are very likely to not support corrupt politicians. This makes it interesting to test how voters react to corrupt behavior of their preferred party’s politician, as corrupt behavior should be viewed as particularly problematic in a country like Germany. Thus, the potentially high probability for corrupt politicians of being punished by the voters and the existence of two radical right parties on each side of the political spectrum makes Germany an ideal case to test our hypotheses.

Experimental design

Our empirical analysis relies on a survey experiment embedded in the survey described above. In the experiment, we presented the respondents with two hypothetical politician profiles and varied two central aspects in the description of these profiles: the party affiliations of the politicians and whether the politicians are involved in a scandal or not. The experiment is designed in such a way that one politician is always a member of the party that the respondent would vote for in the General Election (‘co-partisan’).Footnote 2 The second politician has a randomly chosen party affiliation, i.e., she or he is a member of one of the six major parties in Germany. Further, the profile descriptions also varied regarding whether a candidate has been involved in a corruption scandal or not. Crucially, the experiment is designed in such a way that the politician who shares the party affiliation with the respondent is involved in the scandal, while the other politician—the candidate with the randomly chosen party affiliation—is ‘clean.’ This design allows us to examine the impact of a clean alternative when competing with a corrupt politician. The respondents essentially have to make a decision between a corrupt politician who is from their preferred party and a clean politician who might represent a party that they are less likely to support. It should be noted, however, that our experiment is not designed to fully resemble a vote choice in a German election. Rather, the description of the experiment and vote decision between the two candidates was given so that respondents should indicate which candidate they perceived generally more suitable for being a politician. In this regard, our experiment rather measures the intensity of an ‘anti-corruption norm’ in Germany than replicating an actual vote choice.Footnote 3

All other factors of the politicians’ profiles were kept constant. The corruption scandal is always the same. We did not aim to test for the effect of different types of scandals but were interested in the overall effect of misbehavior on voters’ candidate choice. Specifically, we use an economic and office-related scandal in which a politician has accepted illegal party donations from a company. To avoid biases due to other desirable attributes, we tried to make both descriptions of the politicians quite similar. Moreover, we randomly varied which of the two profiles is the corrupt or clean candidate. Thus, the other attributes do not affect our treatment. Both vignettes read as follows:

Vignette 1:

“Politician A is 55 years old, married, and has two adult children. He is a member of the [PARTY]. Before starting his political career, he has worked as lawyer. His issue focus is on economic and finance policy. [He is currently involved in a scandal, as he has accepted illegal party donations from companies.]”

Vignette 2:

“Politician B is 47 years old, in his second marriage and has one child. He is a member of the [PARTY]. Before starting his political career, he has worked as a teacher in grammar school. His issue focus is on education and science policy. [He is currently involved in a scandal, as he has accepted illegal party donations from companies.]”

This design has, of course, several potential limitations. First, we do not allow for a situation in which both politicians are corrupt. As Agerberg (2020, pp. 259–260) points out: “it is better to have a corrupt politician in office with views similar to yours, than a politician that is both corrupt and politically distant to you.” Thus, in such a situation it is unlikely that voters punish their preferred party’s candidate as they have no advantage from switching their vote to the alternative. Therefore, and because corruption is quite rare in the German context, we did not make use of this option.

Second, as sketched above, we did not vary the type of corruption scandal. It is well known that scandals vary strongly with regard to the specific type of misbehavior (Kumlin and Esaiasson 2012), but also that certain candidate traits are important when voters evaluate scandals (Funk 1996; Rajan and Pao 2022). Previous empirical evidence on the effect of different types of scandals is mixed. Some studies suggest that politicians are punished more severely if they engage in any form of financial misbehavior (Funk 1996; Carlson et al. 2000), e.g., tax evasion or bribery. In contrast, the consequences seem to be less serious for moral misbehavior, such as adultery (Funk 1996; Doherty et al. 2011; Bhatti et al. 2013; Doherty et al. 2014), but this depends on whether the scandal included the abuse of political power. Altogether, previous studies have identified various types of scandals that possibly affect candidate evaluation to varying degrees. In our study, we opted for a scandal which is likely to be evaluated negatively, but we cannot rule out that other types of scandals might lead to other findings because they would be evaluated as more or less severe. Further, we decided to not vary certain individual traits of the candidates. Previous studies show that candidate traits are important predictors for being punished after a corruption scandal. It is shown that especially a candidates’ gender (Funk 1996; Barnes and Beaulieu 2014; Barnes et al. 2020) or being part of any minority group in general (Rajan and Pao 2022; Berinsky et al. 2011) can lead to more punishment after a scandal. Thus, we decided to keep the candidate profiles in our experiment very similar and only vary some characteristics.

Third, we did not offer the option to not support any of the two candidates. This means that in certain cases voters might would neither support the corrupt politician nor the clean alternative. Thus, in such cases, voters might select a politician without having a clear preference for them. However, by forcing respondents to choose which of the two politicians they would rather support, we retrieve their evaluation of the trade-off between corruption and partisan preferences. As previous research has demonstrated, such an enforcement can lead to more accurate results (Hainmueller et al. 2015). Moreover, this is especially important as we are mainly interested in the electoral behavior of voters from the political fringes. Voters of radical right and left parties are probably most likely to abstain from the ballot box when they face a choice between a corrupt politician of their party and another party’s clean alternative, which is due to their ideologically distant position to other candidates in the election. Hence, forcing them to decide in such situations allows us to get a sufficient number of cases and to gain insight in radical voters’ behavior. However, this decision comes at the cost of potentially inflated treatment effects.

Fourth, the general critique of using survey experiments for studying voter responses to corrupt behavior applies to our experiment. As Incerti (2020) demonstrates, survey experiments usually overestimate treatment effects due to social desirability bias. It is much easier for voters to say in a survey that they would punish corrupt politicians than actually doing so in an election. To address this problem, Incerti (2020, p. 772) recommends to use fully randomized conjoint experiments and then to compare “the probability of voting for a realistic candidate with outlier characteristics (e.g., corruption) to the probability of voting for the same realistic candidate without this characteristic, rather than examining differences in AMCEs across fully randomized candidate profiles”. While we do not use a conjoint experiment to focus specifically on the evaluation of the party of the alternative candidate, our experiment is particularly designed to provide two very realistic candidates to the respondents. Finally, it is possible in the experiment that the ‘clean alternative’ has the same party affiliation as the corrupt politician. We deliberately opted for allowing this situation because it provides a ‘manipulation check,’ i.e., when both candidates share the same party affiliation and voters are aware of the treatment and want to punish corruption, then we should observe treatment effects at least in this ‘most likely’ situation.

Despite these limitations, there are also some notable advantages of the experimental design that allow us to test our theoretical expectations. As stated above, our design essentially forces the respondents to make a choice between a corrupt and clean politician. To account for our first hypothesis (i.e., radical party supporters are more willing to support a corrupt co-partisan), this expectation can be tested by regressing whether the corrupt candidate was selected on the party preference of the respondent. Notice that this part of the analysis is purely observational as it only asks the question if radical party supporters are more willing to support a corrupt candidate, all else being equal. The deliberate oversampling of radical voters is particularly helpful in this context as it allows us to get precise estimates for the behavior of radical voters.

Another advantage of our design is that we can analyze in two different ways the impact of the party of the clean alternative on respondents’ willingness to support a corrupt candidate. First, we can analyze the impact of the party label of the clean alternative conditional on the party a respondent supports. For example, based on our theoretical expectations, we would expect that a voter of the Green party is more likely to support a clean alternative when the clean candidate comes from the SPD (social democrats) and not from the AfD (radical right). However, this approach ignores that there might be variation among party supporters in how strongly they like or dislike certain other parties. For example, for some Green voters the CDU/CSU might be a viable alternative while some other Green voters might never consider voting for this party. To account for this aspect, we have asked all respondents to evaluate all German parties on a five-point scale from 1 to 5 (1 = ‘very bad’ to 5 = ‘very good’). By replacing the party label of the clean alternative with this evaluation of the party by the respondent, we obtain a measure of how (un)favorable a respondent considers the party of the clean alternative. This measure is comparable between radical and non-radical voters and, thus, allows us to compare radical and non-radical voters without relying on the specific party affiliation of the clean alternative. In this aspect, our study is different from existing research. Compared to previous studies, we are able to estimate how viable an alternative candidate is for a specific respondent. This estimation is not based on the party affiliation of the alternative in the experiment that either matched those of the respondent or not as in existing studies (Agerberg 2020), but on the evaluation of all parties by a respondent.

Results

We begin with testing our first hypothesis, i.e., whether radical voters are more willing to vote for corrupt politicians. It is important to note that this hypothesis is not directly tested based on our experimental treatments. Instead, the hypothesis simply suggests that, all else being equal, more radical respondents are more likely to select the co-partisan but corrupt candidate in the experiment. Thus, we test this hypothesis by regressing the outcome variable (was the corrupted candidate selected?) on the actual vote choice of a respondent. As we deliberately oversampled voters of radical parties and undersampled mainstream voters, we group all voters of mainstream parties into a single category. We also include a number of control variables, such as the level of political trust, populist attitudes, political interest, age, gender, and education.

Figure 1 displays the results of the analysis based on predicted probabilities. As can be seen, voters of the radical right are substantially more willing than voters of the radical left and mainstream parties to select the corrupt candidate. In contrast, radical left voters are not more likely to vote for corrupt politicians compared to mainstream voters. These findings provide some support for our first hypothesis as radical right voters are clearly more willing to support a corrupt co-partisan candidate. However, radical left voters do not follow this pattern. To assess the robustness of these findings, we replaced vote choice with the self-placement of voters on the left–right spectrum. Following the radicalness argument, we should observe that more left- and right-wing voters are more willing to support corrupt politicians, while more mainstream voters are less likely to do so. We therefore run another regression now using the self-placement of respondents on an 11-point left–right scale as independent variable in addition to the other control variables. As we expect to find a U-shaped relationship, we also include a squared term of left–right self-placement. The results are displayed in Fig. 2 and confirm the expectation. Indeed, mainstream voters are the least likely to support corrupt candidates. Radical left respondents are more likely to support corrupt politicians, but the by far highest probability of supporting corrupt politicians is observed for radical right respondents. Overall, these findings support our Hypothesis 1 with the important specification that especially radical right voters are willing to turn a blind eye on corruption. A potential mechanism why the radical left voters are not different from the mainstream party voters might be found in the specific case of the Left Party in Germany. While being clearly the radical left party in the German political system, the Left Party appears not as isolated as the AfD in the party system, especially from the perspective of voters. Our data support this suspicion. In Fig. 5 in the appendix of this paper, we demonstrate that mainstream voters and Left Party voters evaluate each other rather favorably. In contrast, the AfD is evaluated as extremely negative by all voters except for their own supporters.

Fig. 1
figure 1

Predicted probabilities of supporting a corrupt politician conditional on party preference. Note X-axis displays probability to support the corrupt politician who is a member of the party a respondent wants to vote for. Y-axis displays the party preference of a respondent with voters of mainstream parties grouped together. Horizontal lines are 95% confidence intervals. Visualization is based on Model 1 in Table 1

Full size image
Fig. 2
figure 2

Predicted probabilities of supporting a corrupt politician conditional on left–right self-placement. Note Y-axis displays probability to support the corrupt politician who is a member of the party a respondent wants to vote for. X-axis displays the self-placement of a respondent on the left–right scale. Vertical lines are 95% confidence intervals. Visualization is based on Model 2 in Table 1

Full size image

Effect of viable alternatives

We now assess our second hypothesis, i.e., whether the effect demonstrated above depends on the existence of a viable alternative. To do so, we include the party affiliation of the competing candidate in the analysis. It is important that we do not directly include the party label of the candidate against which corrupt candidate competes against in the analysis.Footnote 4 Instead, we replace the party label of the clean alternative with the evaluation of this party by a respondent. To do so, we asked all respondents at the beginning of the survey to rate all six parties on a five-point scale ranging from ‘very negative’ to ‘very positive.’ We use this evaluation to replace the party affiliation of the clean alternative in the analysis. For example, when a respondent has evaluated the SPD as ‘rather negative’ and the clean politician in the experiment comes from the SPD, we use this value as the treatment. Our treatment variable, thus, has a five-point range from ‘very negative’ to ‘very positive.’ The expectation is that voters are generally more likely to continue to support the corrupt candidate the worse the evaluation of the party of the clean alternative. Crucially, to test H2, we interact this treatment variable with the party preference of the respondents. If radical party voters are equally critical of corruption than mainstream party voters, we should see no differences in the predicted probabilities to support the corrupt candidate conditional on the evaluation of the clean alternative’s party.

The results are displayed in Fig. 3 and demonstrate that there are no systematic differences between radical party supporters and supporters of mainstream parties, when we use the evaluation of the party of the clean alternative as the treatment variable. This finding supports our expectation that radical voters are similarly likely to switch away from their preferred party in case they evaluate the alternative as favorable as mainstream voters (Hypothesis 2). This strongly suggests that the previous findings—radical party supporters are more willing to vote for a corrupt candidate—are driven by the fact that radical party supporters evaluate other parties more often negatively. Overall, the findings from this alternative model specification suggest that radical party supporters are no different from non-radical party supporters when they have the same evaluation of the clean alternative. Both groups of voters are likely to vote for a clean alternative when they positively evaluate the alternative party, and they are very likely to vote for a corrupt politician when they evaluate the clean alternative’s party as very negative.

Fig. 3
figure 3

Predicted probabilities of supporting a corrupt politician conditional on Evaluation of Party of the ‘Clean Alternative’ (y-axis) and Party Choice. Note Horizontal lines are 95% confidence intervals. Visualization is based on Model 4 in Table 1

Full size image

Effect of strength of partisanship

Finally, we conduct additional analyses for a better understanding of the relationship between party preferences and support for corrupt politicians. In particular, what is slightly surprising about the previous results is that quite a large amount of voters tends to support the corrupt politician even when they evaluate a different party as quite favorable. According to Fig. 3, more than 12.5% of voters still support the corrupt politician despite evaluating the clean alternative’s party as ‘very positive.’ This is even more surprising as survey experiments notoriously overestimate the amount of voters who say that they would not vote for a corrupt politician due to social desirability bias (Incerti 2020). Thus, we assume that there might be unobserved heterogeneity in the data which could explain this finding, at least, partially.

Specifically, we assume that the strength of support for the party that a respondent would vote for can moderate the effect as suggested by our Hypothesis 3. The costs of switching votes might be higher for a respondent who strongly supports a certain party compared to a voter who is less partisan. Hence, as argued in the theoretical section, we can expect that the absolute strength of support for the party a respondent would vote for has an effect on the probability of voting for a corrupt politician.

We put this expectation to an empirical test by creating three groups of voters: (1) strong partisans, i.e., respondents who evaluate the party they would vote for as ‘very positive,’ (2) partisans, i.e., respondents who evaluate the party as ‘mostly positive,’ and (3) weak partisans, i.e., respondents who evaluate the party they would vote for as ‘neither positive nor negative’ or even worse. The majority of respondents fall into the second category of ‘partisans’ (N = 530) but there is also a reasonable amount of respondents in the two other groups (N = 280 for ‘strong partisans’ and N = 191 for ‘weak partisans’).

The results are displayed in Fig. 4. They demonstrate clear variation between respondents with different strengths of partisanship. Strong partisans are indeed more likely to support a corrupt politician of ‘their’ party than voters with weaker levels of partisanship. Thus, these findings further explain the relationship between partisanship and the support for corrupt politicians.

Fig. 4
figure 4

Predicted probabilities of supporting a corrupt politician conditional on Evaluation of Party of the ‘Clean Alternative’ (y-axis) and Strength of Partisanship. Note Horizontal lines are 95% confidence intervals. Visualization is based on Model 5 in Table 1

Full size image

Conclusion

In this paper, we have analyzed how the availability of ‘clean alternatives’ affects a voter’s probability to support corrupt politicians. Based on the idea that voters face a trade-off when confronted with a corrupt politician of the party they intent to vote for and a ‘clean alternative’ from a potentially different party, we argued that the ideological proximity of the alternative politicians matters for voters’ decision-making. Following Charron and Bågenholm (2016) and Agerberg (2020), we have further argued that the ‘ideological isolation’ of radical voters might decrease their willingness to support clean politicians from other parties, simply because they are more likely to lack viable alternatives. Our survey experiment conducted in Germany finds evidence for these mechanisms.

Our paper thus contributes to recent literature, which highlights the important role of analyzing the political competitors when analyzing why voters tend to support corrupt politicians (Agerberg 2020). Our results corroborate the conclusion of Incerti (2020, p. 770) that voters “find it costly to abandon their preferences even if it forces them to select a corrupt candidate.” In fact, we find that some voters still support the corrupt candidate even though they evaluate the party of the alternative candidate quite positively. Based on these outcomes, it can be concluded that it is less costly for politicians of radical parties to be involved in a scandal compared to politicians of established mainstream parties. To an extent, our study also provides an empirical explanation of the fact that radical and scandal-ridden politicians, such as Donald Trump, remain unhurt within their electorate. Voters of mainstream parties often evaluate several parties as potential alternatives, and our findings indeed show that voters of these parties are more likely to switch votes in favor of a clean alternative compared to voters of radical parties. However, this finding is driven by the fact that radical party voters evaluate other parties more negatively. Once it is controlled for this evaluation of the alternative party, we find that radical voters behave similar to non-radical voters and are equally likely to switch away from the corrupt candidate. It is thus important to notice that supporters of radical parties are not necessarily less likely to evaluate corruption negatively, but they are often less likely to switch votes due to a corruption scandal. Our findings further demonstrate that also the strength of the partisan attachment of a voter is important. Strong partisans are less willing to support another candidate than their preferred one, even if this candidate has conducted misbehavior. Voters with weaker partisan attachments, in contrast, are more likely to support a non-corrupt candidate from another party. In this regard, our findings can speak to the literature on partisan polarization. With higher levels of polarization, which increase hostility toward other parties and strengthens identification with the preferred party, corruption might be less punished by voters.

Of course, several limitations of this study should be acknowledged. First, the experimental setup considers only two candidates and provides a limited amount of information about the politicians. Further, we decided to not give the respondents the possibility to abstain from their vote. While this was chosen deliberately, future studies could use more complex designs such as conjoint experiments in which more information is provided. Second, we only used one scandal—the illegal acceptance of party donations—but scandals strongly vary with regard to their topic and severity. A problem here is that there is rather mixed evidence on which types of scandals are evaluated as less or more harmful by voters; therefore, we cannot say whether similar results would be found for other scandals, such as adultery or drug abuse. More research on what voters perceive as a particularly severe case of misbehavior thus seems desirable. Further, as our findings rely on a case study it is important to mention that Germany is generally considered to be a low corruption country. The candidate choice of voters in countries where corruption scandals are more common thus might differ from our findings. Particularly in contexts where corruption is widespread voters are likely to continue support for corrupt politicians, as it is also demonstrated in previous studies (Bauhr and Charron 2018; Klašnja and Tucker 2013; Klašnja et al. 2021). Under such circumstances voters are so frequently confronted with corrupt behavior that it does not affect their candidate evaluation anymore. Instead, they rely on other candidate traits (Pavão 2018). However, as German politicians often resign from office when they are accused of corrupt behavior, it is particularly interesting to see which impact corruption can have in such cases. Analyzing the behavior of voters in low corruption countries thus is important to find out whether the distaste for corruption is stronger in such cases. However, to assess the generalizability of our findings, replications in other contexts and with different methods are highly desirable. The limitations that are inherent to using survey experiments for studying corruption (Incerti 2020) should also be kept in mind.

These limitations notwithstanding, our study also provides the ground for follow-up questions and future research on this topic. Our paper calls for an extension of the experimental design. More specifically, future studies can use a more complex experimental design to test the impact of radicalness and viable alternatives on candidate choice related to corruption. For example, varying the gender of the candidates or including further candidate traits, such as their political experience or offices, can further enhance our understanding of voting behavior in the aftermath of corruption scandals. Further, our study also raises the question whether voters’ reactions to scandals might depend on additional factors than partisanship. For example, populist attitudes are often characterized by seeing established parties as corrupt. Based on such a perspective, populist voters might be more sensitive to corruption. However, populist voters might also be strongly polarized and therefore not punish their favorite politician for corrupt behavior. Comparable arguments could be made for political trust. Analyzing such factors in more detail and based on experimental approaches seems to be a valuable avenue for future research.