There are such people: the role of corruption in the 2021 parliamentary elections in Bulgaria

ABSTRACT

Voters often express a dislike for corruption, yet on election day, they still vote for corrupt politicians. While existing research highlights the impact of information and concern about corruption on voter behavior, our novel theoretical approach integrates three elements–dislike, knowledge, and care–to better understand corruption's role in parliamentary elections. We test this framework on Bulgaria, a nation grappling with pervasive corruption. Using a commissioned survey from Alpha Research, we discover that voters who have a dislike for corruption, can identify a party as corrupt, and consider corruption when voting are significantly less likely to support a corrupt party.

KEYWORDS:

• Corruption
• elections
• voter behaviour
• Bulgaria

Introduction

The c word seems to matter in the politics of Central and Eastern Europe (CEE). Turn on the television news, read a newspaper or eavesdrop on a conversation between voters about politics and the word “corruption” is frequently invoked. Political scientists studying the region have long suggested that corruption has been a central theme of elections (Birch 2019; Charron and Bågenholm 2016; Engler 2016; Enyedi and Deegan-Krause 2018; Kostadinova 2009), but the role corruption plays has been invoked more often than explored systematically. How does corruption shape voting behaviour? Why do some parties make hay from anti-corruption appeals? Why do other parties tainted by corruption manage to hang on to their support? And why do anti-corruption appeals only seem to have an impact for some voters who express their dislike for corruption, but not for others?

Building on the insights of De Vries and Solaz (2017), we develop and test a new theoretical argument highlighting the role corruption plays in elections and in voters’ choices. In the retrospective voting model, voters acquire information about corruption, they then attribute blame and that produces a behavioural response. Much of the existing literature tends to see anti-corruption voting in black or white terms: either corruption matters to voters or it does not and in the latter case is mainly due to some form of ideological or economic compensation. While some voters are mobilized by binary distinctions, for many others their views are less categorical and lie on a sliding scale. The effect of corruption in elections is not simply a black and white story. Our novel contribution to the academic debate is to maintain that the question is not so much whether corruption matters, but how much it matters to the voter. Furthermore, we argue that understanding the impact of corruption on voting behaviour goes beyond just knowledge of corruption but is the product of a combination of factors: dislike, knowledge and taking corruption into account when voting. The extent to which corruption is central to an individual voter’s decision at the ballot box is shaped by the salience of the issue to the voter and the extent to which the party he/she chooses places the anti-corruption appeal at the heart of its message. This underlines the combinatorial aspect of anti-corruption voting that lies at the heart of our argument, but to that combination, we need to add one component to the retrospective voting model: the am I bothered? element. Voters need to care about corruption for it to play a role in their decision-making on election day.

To test our theoretical model, we draw on the case of Bulgaria. It is a suitable case study for several reasons linked to widespread and enduring corruption, active media reporting, awareness of the topic among the population, and the use of anti-corruption rhetoric by parties and protest organizers. First, the country’s position at 72 out of 180 countries on Transparency International’s (2022) corruption rankings indicates that corruption is perceived to be prevalent in the country and hence we might expect it to play a role in electoral behaviour. Second, the theme of corruption in Bulgaria has been prominent in both the domestic and international press.¹ Third, corruption is not so endemic the country is beyond the pale where voters might consider any efforts futile. Indeed, it is striking that not only has corruption been a salient theme of Bulgarian politics in the past three decades provoking a number of large protests over the years (e.g. Engelbrekt and Kostadinova 2020), but much of the coverage of Bulgaria’s three parliamentary elections in 2021 stressed the importance of anti-corruption appeals (e.g. Zankina, Lin, and Haughton 2021). In those elections, parties that won large slices of the vote such as “There Is Such a People” and “Change Continues’ trumpeted such appeals and their anti-corruption credentials.

Many of the most illuminating studies of the impact of corruption on voting behaviour have focused on individual countries, especially in Europe and Latin America (Bågenholm and Charron 2020; Ferraz and Finan 2008; Mares and Young 2018; Winters and Weitz-Shapiro 2020). While acknowledging the limitations of a detailed study of one country, in line with Pepinsky (2019), we maintain an in-depth study of Bulgaria (which has not been the subject of such an analysis) accords us both an opportunity to examine a phenomenon systematically and generate insights that can then be tested on a larger scale in further studies. In this regard Bulgaria can be treated as a typical case (Gerring 2008); the findings of which would likely illuminate dynamics at work in a wide variety of countries.

Although some publicly available databases (Comparative Study of Electoral Systems and Chapel Hill expert survey) offer useable data on party positioning and aspects of electoral (GESIS 2022; Jolly et al. 2022), there is a lack of large cross-country comparative election data examining the role of corruption in electoral behaviour in CEE in general and Bulgaria in particular. Given the dearth of data, we commissioned a survey from (one of) the leading polling agencies in Bulgaria, Alpha Research which allowed us to explore more systematically how and why corruption matters at the information acquisition, blame attribution and behavioural response stages of De Vries and Solaz (2017) model.

We demonstrate not only that voters may perceive some parties differently to expert opinion, but that voters weigh up their options at the ballot box: the more corruption matters to voters the more they take it into account when voting and hence the more likely they are to vote for clean parties and the less likely they are to vote for corrupt parties. These findings not only help explain the outcome of the 2021 elections, but also illuminate the need to examine more closely the complex picture of voter preferences in party systems with significant fluidity such as those in CEE.

We begin by drawing on the scholarship of De Vries and Solaz (2017) highlighting the need to examine the stages of information acquisition, blame attribution and behavioural response. After introducing the landscape of Bulgarian party politics, we outline the parameters and merits of our survey. We then examine the data in detail highlighting how voters’ gradations of salience explain why some parties were beneficiaries of the support of anti-corruption voters more than others. Moreover, we use decision tree analysis to illuminate the conditions present when voters make their choices. We conclude by exploring how far and wide the findings will travel and outlining potential fertile furrows for future research.

Corruption and voting behaviour

Existing scholarship on the role of corruption² in shaping voter behaviour tends to fall into one of four camps. Firstly, there is an almost fatalistic, “that’s just the way it is”. Voters may simply see corruption as endemic or ingrained in the country’s politics and wider society (e.g. Pavão 2018). Despite the fact that some voters clearly do express such fatalistic sentiments, the use of anti-corruption appeals by parties that are successful in elections in CEE indicates not all voters share such fatalism.

Secondly, voters do consider corruption to be an important issue, but its importance is overshadowed by other issues, most notably dealing with economic challenges, political instability or security threats. The salience of corruption is impacted by the form of political malfeasance (Mares and Visconti 2020), ideological positioning and (group) identity politics (Burlacu 2020), the place of parties on the political spectrum (Jiménez and García 2019), and the proximity of cleaner alternatives (Riera et al. 2013). This body of literature clearly points to the need to weigh up how much corruption matters and how its importance to voters interacts with, and impacts on, other factors that shape choices at the ballot box.

Thirdly, corruption voting is impacted by the clarity of responsibility. The ability and ease of identifying who is responsible shapes how much corruption calculations are placed at the heart of voters’ decision-making (Schwindt-Bayer and Tavits 2016). The clarity of responsibility can be muddied in multi-party systems with coalition governments where the finger of blame becomes difficult to point and hence may not translate easily into vote choice. More significantly, clarity of responsibility only explains the punishment part of electoral behaviour linked to whom a voter would not cast his/her vote rather than his/her actual choice. Indeed, on the reward side of voting behaviour we see that many parties in CEE that win the anti-corruption vote are new parties (Engler 2016; Hanley and Sikk 2016; Haughton and Deegan-Krause 2020).

Fourthly, even when corruption is recognised as an important problem for society and alternatives do exist, voters’ decisions may be made by cost–benefit calculations (Klašnja, Tucker, and Deegan-Krause 2016). Voters may ignore corruption if it brings home the bacon to the local community or ethnic group. As with the clarity of responsibility that may help explain some voting on the punishing side, but it does not explain why voters vote for specific clean parties in a multi-party system. Moreover, much of the corruption–in terms of scandals–indicates that corruption tends to benefit a few members of the elite and their bank balances rather than ordinary voters, indicating at best a cost–benefit miscalculation.

This overview of the literature highlights a number of elements that make up anti-corruption voting linked to punishment and reward. Recognizing the various components that make up anti-corruption voting De Vries and Solaz (2017) developed a three-stage model (Figure A1). In the initial stage voters need to acquire information about corruption. Politicians may be very adept at hiding information on scandals or leaning on media outlets to supress reporting. But information can emerge through media reporting or personal experience (e.g. Klašnja, Tucker, and Deegan-Krause 2016). We might assume here that an increase in the level of public information regarding corruption would provoke voters armed with such information to punish corrupt politicians, but the fact that voters continue to vote for corrupt politicians suggests the information acquisition stage may be a necessary, but far from a sufficient condition for anti-corruption voting (Incerti 2020). Following the acquisition of information, the second stage in De Vries and Solaz’s model is the attribution of blame with voters identifying who is responsible for the corrupt behaviour. This attribution may be impacted by the lack of clarity of responsibility mentioned above. Acquisition and attribution, however, need a third element to become anti-corruption voting: a behavioural response. De Vries and Solaz (2017) specify three behavioural responses voters can take when they identify corruption: switch to a clean party, stick with the corrupt party or abstain from elections. Voters weigh up competing factors that influence their vote choice. In the behavioural response stage, voters might take into account whether the corrupt party is ideologically close to them or whether the corruption the party commits is profitable for themselves. De Vries and Solaz (2017) model offers three questions central to a decision tree for anti-corruption voting: what corrupt activities were committed; who was responsible; and how (if at all) should that impact my vote?

De Vries and Solaz (2017) model enhances considerably our understanding of corruption in electoral behaviour as it maps out the different decision stages a voter has to go through from experiencing or perceiving corruption until this perception or experience is manifested in a quantifiable behavioural voting response. Nonetheless, we suggest the need for three significant changes when modelling anti-corruption voting (see Figure 1 for a graphical overview). Firstly, we argue the decision-making process is captured better by a four-element model. The new additional element, am I bothered?, can be regarded as a preconditioning element. It focuses our attention first and foremost on the voter. Ultimately voting behaviour begins and ends with the voter who may, or may not, be concerned about corruption. Not all voters will be concerned about corruption (some may fall into the fatalistic group mentioned above) so the information acquisition stage only matters for those voters for whom corruption is a matter of concern. Voters that do not bother about corruption will not even start acquiring information about corruption. They have already “checked out” before the three-stage process has even started.

Figure 1. Theoretical decision tree on corruption and elections (Stages 1–3 based on De Vries and Solaz 2017).

Secondly, in the information acquisition phase we need to examine not just the corruption messages sent by parties, but also how voters receive those messages. Much of the research on anti-corruption appeals in CEE has tended to focus on parties’ pitches to the electorate and has used expert survey evaluations of parties’ anti-corruption appeals (e.g. Engler 2016). But in order to understand voting behaviour, we need to focus on the demand rather than the supply side of party politics: at the voters themselves and at how much the voter perceives a party to be corrupt. A party seen by experts to be corrupt may be viewed distinctly differently by voters and hence we may get a distorted perspective of the role corruption plays in elections if we look through a party or expert lens rather than through the eyes of the voter. What matters in terms of producing an anti-corruption vote is a combination of the am I bothered stage with the “is the party corrupt?” Both need to be present for anti-corruption voting. Some voters may perceive particular activities as evidence of corruption, but not care; others may care, but not deem actions to be corrupt.³

Overall, we suggest that that the four elements—dislike, corruption information, blame attribution, and taking corruption into account when voting–in combination explain when, how and why corruption matters in elections. Although we suggest that it is the combination of the four that matters and not so much the sequencing, it is most likely for the four elements to play out in the logical sequence of Figure 1, but we recognise that the elements do not have to play out in this sequence to end up with this outcome.

Critics might point out that how voters perceive corruption might be different from what scholars define as corruption. In our work this is less of a concern as we shift all concepts and effects of corruption to the voters’ side of the electoral equation. In our survey design and the regression analysis we focus strictly on the voter and not our own understanding of corruption. We ask the voter whether they dislike corruption, whether they perceive the party they voted for as corrupt and whether they took corruption into account when voting. We are not imposing our scholarly understanding of whether the voter perceives corruption “correctly” but allow the voter to interpret what they mean by the term corruption all by themselves. As such the concept of corruption stays constant for the voter for the whole time of the process.

Thirdly, the binary distinction between caring and not caring about corruption is helpful at the start of the theoretical decision tree, but when considering the behavioural response stage/element we argue it is necessary to consider corruption as a potential intensifier. Whereas previously scholars argued that the behavioural response to corruption might be influenced by whether the party is the one the voter usually votes for or is ideologically close to (Burlacu 2020) or if the perceived corruption is actually benefitting the voter (Klašnja, Tucker, and Deegan-Krause 2016), we argue for the importance of something more fundamental: we need to consider how much corruption matters to voters. As they weigh up their options, depending on the intensity of their attitudes towards corrupt behaviour, a voter may or may not choose to punish a corrupt party and choose the party with the strongest anti-corruption appeal. Our argument would suggest that only with the combination of disliking corruption, identifying a party to be clean, and caring sufficiently about corruption is there a significantly higher likelihood of voting for a clean party.

Given the combination of factors required for anti-corruption voting, we argue there are many scenarios involving a corrupt party where a voter would still choose to cast a ballot for that party. A voter may dislike corruption and may be willing to base his/her voting decision on corruption, but does not perceive the party to be corrupt, so does not punish that party. Moreover, a voter might dislike corruption, perceive the party to be corrupt, but does not care enough about corruption to take it into account when making a voting choice. Furthermore, there are scenarios where for voter only one of those factors is present. Finally, there is a least worst scenario in which a voter does care about corruption, perceives a party he/she votes for as corrupt and does actually take corruption into account when making a choice, but still votes for the corrupt party (maybe as a lesser evil).

Our discussion highlights that voting choice based on corruption is less likely than one might think. There is a sequential dimension to our argument, but our explanation for voting behaviour is primarily a combinatorial one. At a minimum, there are several necessary conditions required for corruption to impact on electoral behaviour at all. If the chain of decision-making of the four stages breaks down at one point in the process, voting behaviour will be affected. And yet it is not only the different components of “care/dislike”, “knowledge” and “decision to act” that fully explain the role corruption plays in election, but the combination of the elements that explain it in the most useful way.

There are eight possible scenarios on the relationship of corruption and an individual’s voting decision. In only one scenario would a voter not vote for a corrupt party: when all three conditions are met i.e. when the voter dislikes corruption, perceives the party to be corrupt and takes corruption into account when voting. And yet, in the case of Bulgaria, as discussed below, a large share of the population falls into this one category which made corruption a salient issue in elections. Furthermore, Figure 2 also illustrates that dislike alone is a worse explanation in understanding how corruption affects voting decisions than the combination of the three.

Figure 2. Theoretical voting decision tree.

Although we would suggest that the relationship between corruption and voting decision or election outcomes is best explained when looking at a combination of the three factors—dislike of corruption, knowledge about corruption and taking corruption into account when voting—there is a possibility that this might not be the case: when a single effect can equally well explain voting decision. Studies of Argentina and Brazil (Weitz-Shapiro and Winters 2017; Winters and Weitz-Shapiro 2020), for instance, suggest that credible information (i.e. knowledge) about corruption affects voting decisions. Hence, we provide single effect hypotheses in this instance to make sure that the sum of the three parts is really more useful in explaining voting decision than the three parts on their own. Furthermore, whereas other scholarship focuses on the sending of information (e.g. Winters and Weitz-Shapiro 2020) i.e. the credibility of the information source, our work focus on the receiving of information and how that impacts voting behaviour.

Given the preceding discussion we derive the following hypotheses:

Hypothesis 1 – Corruption Dislike: The more a respondent dislikes corruption the less likely the respondent is to vote for a party an expert would rank as corrupt (1a) and the more likely to vote for a party an expert would rank as a clean party (1b).

Hypothesis 2 – Corruption Knowledge: The more corrupt a respondent perceives a party to be, the less likely his/she is to vote for a corrupt party.

Hypothesis 3 – Corruption as Basis for Voting Behaviour: The more a voter bases their voting decision on issues around corruption, the more likely he/she is to vote for an expert rated clean party (3a) and the less likely he/she is to vote for an expert rated corrupt (3b) party.⁴

Hypothesis 4 – Conditional Effects of Dislike, Information and Voting decision: Voters who dislike corruption, perceive a party to be corrupt, and take corruption into account when voting, are less likely to vote for a corruption party (i.e. more likely to vote for a clean party) than voters who do not dislike corruption (all other variables become moot).

There are such a people: anti-Corruption appeals in Bulgaria

Labelled the poorest and most corrupt country in the EU (The Guardian 2017)⁵, it is no surprise that corruption has been a perennial theme of politics in Bulgaria over the past three decades (Engelbrekt and Kostadinova 2020; Noutcheva and Bechev 2008).

Corruption and wider questions over the poor quality of governance have long been barbs thrown at one of the perennials of Bulgarian politics, the Bulgarian Socialist Party (BSP) (the successor to the Bulgarian Communist Party), that led governments in all three decades since 1989. The BSP-led 2005–2009 coalition government was severely criticized by the European Commission for widespread corruption, organized crime, the embezzlement of EU funds and an inefficient judiciary. Allegations of corruption, combined with poor performance in government and economic woes fuelled the fires of widespread protests at several points during Bulgaria’s post-communist experience (Engelbrekt and Kostadinova 2020), most notably following the tin-eared decision of the 2013–2014 BSP-led coalition government to appoint Delyan Peeveski, an elected deputy and media businessman accused of having connections with organized crime, as head of the State Agency for National Security.

More broadly, corruption has been a prominent theme at election time. Scandals, for instance, contributed to the slump in support for the centre-right Union of Democratic Forces (SDS) that had stormed the boards with over half of the vote in the 1997 election but could only muster just 18% four years later. Those elections in 2001 were won by a party launched just weeks before polling day: the New Simeon II Movement. It garnered over 40% of the vote, galvanizing voters’ support with a message that a new broom would help sweep away corruption and allow the country to make a fresh start (Gurov and Zankina 2013).

The dominant figure of Bulgarian politics in the twelve years running up to the 2021 elections was Boyko Borissov. For those dozen years he held power almost uninterruptedly, serving as prime minister three times (Spirova and Sharenkova-Toshkova 2021). Anti-corruption appeals were central to his initial pitch to voters in 2009, although mixed into a larger call to remove the BSP and another of Bulgaria’s perennial parties, the Movement for Rights and Freedoms (MRF), from power. By 2021, the appeal of the party he founded and led, GERB (Citizens for the European Development of Bulgaria),⁶ was a combination of his personal appeal and delivery in office. In April’s election his social media feeds were replete with posts of the prime minister travelling around the country visiting roads, school and churches, built during GERB’s time in government. But surrounding many of the infrastructure projects was more than the whiff of corruption. Many tenders associated with their projects seemed to benefit GERB’s friends and associates more than the Bulgarian population, provoking widespread demonstrations in the autumn and the rebuke of the European Anti-Fraud Office. Just as in 2001 and 2009 allegations of graft and cronyism fuelled the fires of new parties in the 2021 elections.

The year 2021 was an unprecedented year for Bulgarian electoral politics. The country held not one but three parliamentary (see Table 1) and one presidential election. The plethora of parliamentary elections owed much to the difficulty of forming a government and the specificities of the Bulgarian constitution, but that in turn stemmed from the appeals and motivations of politicians running in those elections. The experience of three of Bulgaria’s established parties was mixed. GERB came out on top in the April 2021 elections, but saw its vote fall. GERB was outperformed in the subsequent 2021 elections, registering an ever-declining number and percentage of votes. The once mighty machine of Bulgarian politics, BSP, registered historically low results in all three 2021 elections, downgrading the party to a second- if not third-rank player. In contrast, MRF, which draws much of its support from the ethnic Turkish voters (but also from other minorities such as Roma and Pomaks), saw its vote share and total vote increase in 2021. Corruption allegations have long swirled around MRF, both in terms of local level politics where some areas are seen as the party’s fiefdoms, but also in national level scandals.

Table 1. Parliamentary election results 2017 and 2021.

		2017			April 21			July 21			November 21
Type	Party	Votes	%	MP	Votes	%	MP	Votes	%	MP	Votes	%	MP
Established/Main Stream	GERB	1,147,292	33.54	95	837,707	26.18	75	642,165	23.51	63	596,456	22.74	59
	BSP	955,490	27.93	80	480,146	15.01	43	365,695	13.39	36	267,817	10.21	26
	MRF	315,976	9.24	26	336,306	10.51	30	292,514	10.71	29	341,000	13	34
Other/Hybrid	Volia	145,637	4.25	17	75,926*	2.13*	0	85,795*	3.14*	0	7,067	0.27	0
	RB+/DB	107,407	3.14	0	302280	9.45	27	345331	12.64	34	166968	6.37	16
	DaBG/DG	101,177	2.96	0
New Parties/Protest Parties	ITN				565,014	17.66	51	657,829	24.08	65	249,743	9.52	25
	ISMV				150,940	4.72	14	136,885	5.01	13	60,055	2.29	0
	PP										673,170	25.67	67
Nationalist/PRR	IMRO	318,513	9.7	27	116,434	3.64	0				28,322	1.08	0
	NFSB				75,926*	2.13*	0	85,795*	3.14*	0	8,584	0.33	0
	ATAKA				15,659	0.49	0	12,585	0.46	0	12,153	0.46	0
	Revival	37,896	1.11	0	78,414	2.45	0	82,147	3.01	0	127,586	4.86	13

Source: National Electoral Commission cik.bg.

*Running in a coalition together

The most striking aspects of all elections in 2021, however, was the success of new parties which cashed in on public discontent and on anti-government and anti-corruption appeals. Some of those parties were born out of the 2020 anti-Government protests which demanded the resignation of Borissov and the Prosecutor-General, Ivan Geshev. The protests engendered several new political formations, including a movement led by former Ombudswoman Maya Manolova (former BSP MP), Izpravi se! Mutri vŭn! (ISMV-Stand Up! Mafia, Get Out!). Another formation, Democratic Bulgaria (DB), emerged as a coalition of already existing parties with predominantly urban, but limited support.

But electorally the most significant party to emerge and achieve success in spring 2021 was founded and led by the prominent singer and long-standing television presenter Slavi Trifonov: Ima Takav Narod (ITN—There is such a People). Although his party was new Trifonov was not a complete novice to politics, most notably he had been a driving force behind the 2016 referendum on reform of the electoral system. His appeal owed much to his pop-folk music, which struck a chord with voters attracted by its homeland undertones. Trifonov used his media appearances, including his own TV station, and virtual concerts to spread his anti-corruption and anti-status quo message to voters at home and abroad alike.

Despite the strong anti-GERB sentiment among most parties in the new legislature, there was no consensus on a viable coalition. Consequently, the new National Assembly was dissolved by President Radev after just four weeks. Radev installed an interim government, and further elections were scheduled for 11 July. On that day ITN, garnered 24.1% of the votes and 65 seats, and succeeded in narrowly defeating GERB who only managed 23.5% and 63 seats, marking the first time since 2007 that GERB or a GERB-led coalition had not been placed first in elections. The BSP obtained 13.4%, Democratic Bulgaria secured 12.6%, the MRF won 10.7% and 29 seats, while Izpravi se! Mutri vŭn! (now renamed Izpravi se! Nie Idvame!, Stand Up! We are Coming!) won 5.0%. ITN was given a mandate to form a government but failed to reach a consensus among protest and established parties. Bulgarians hence went to the polls for a third parliamentary election in November 2021, which was combined with the presidential elections.

The winner was yet another new party with a very appropriate name given the churn of Bulgarian politics: “Change Continues” (PP – Provalzhavame Promyanata). Formed by entrepreneurs and Harvard graduates, Kiril Petkov and Assen Vassilev, “Change Continues” reaped the popularity they had gained as ministers in the caretaker government formed in May. Winning the November election with 25.7% of the vote, PP undercut the support for other new protest parties. ITN and Democratic Bulgaria lost half the votes they had won in July, while Stand Up BG! We are Coming! lost even more and failed to cross the electoral threshold. Following November’s election, the apparent never-ending churn of Bulgarian politics continued with elections in October 2022 and April 2023. In the latter case, ITN experienced a modest revival returning to parliament and a far right party that had spent its first seven years outside parliament won a seventh of the vote.

No account of Bulgarian politics would be complete without mention of the nationalist parties. In the decade and a half running up to 2021's elections around a tenth of voters had cast their ballots for nationalist parties. Since 2005, various configurations have been represented in parliament and, between 2017-2021, even in government. Those include Ataka, the Internal Macedonian Revolutionary Organization (IMRO), the National Front for Salvation of Bulgaria (NSFB), and as of more recently, Revival. In 2021's three elections different configurations ran in all three of the elections. In July’s election, for instance, IMRO-Bulgarian National Movement, the Volya Movement and the National Front for the Salvation of Bulgaria came together under the umbrella of Bulgarian Patriots. But by November’s election all three of those parties and Ataka ran separately, yielding no seats for any of them (see Table 1), at the expense of Revival which increased its support tenfold from March 2017 to April 2023.

This overview of party politics in Bulgaria indicates that corruption mattered in helping explain the rise and fall of parties and their respective successes at the ballot box. Moreover, a large proportion of respondents (91 percent) in our survey (to which we return below) expressed their dislike of corruption, with 75 percent greatly disliking corruption; figures that are in line with other studies of citizens’ attitudes across the globe (Fisman and Golden 2017). Nonetheless, it still leaves open the question of how much corruption really does impact on voting behaviour.

Research design: examining the motivations of voters

In order to examine more systematically the role of corruption in electoral behaviour we opted to collect voting data in a traditional survey design as we argue that the benefits of a traditional survey design outweigh the advantages of using other methods such as an experimental design.⁷ In particular, voting decisions are hard to examine through experiments. Not only are the “costs” of changing one’s vote “lower and more abstract in hypothetical environments” (Incerti 2020, 761), but there is invariably a gap between the set-up of the experiment and real-world voting, both in terms of the context and in terms of the choices voters are offered.

We recognize that surveys are not without their limitations. There may be a social desirability bias that would encourage survey respondents to overreport their distaste for corruption and underreport how corrupt a party is (although the survey agency Alpha Research considers this to be less of a problem in Bulgaria as corruption features very prominently in society and politics in Bulgaria). Moreover, from the general debate around corruption measurement, we know that citizens tend to underreport their engagement/participation in corrupt interactions in surveys (Oliveros and Gingerich 2020). More broadly, participants might want to be seen as part of the winning team or respondents may try to give the answer they think the interviewer wants to hear and hence in post-election surveys respondents might report voting for the party that won even if they did not. We are cognizant of the risks, but we know that the bias of reporting needs to be very substantial to translate into a biased estimation. Given these concerns, we asked the survey agency to compare the survey results with the overall election results to make sure we have reliable data with respect to voting decisions. From the actual election results and the voting decision reported in the survey, we conclude that the bias in the response category for voting is minimal.

Despite the imperfections of traditional survey design, they offer clear advantages. Voting in elections is driven by choices made by individuals. It is impossible to understand fully why voters cast their ballots in the way they did without asking them. Moreover, conducting a survey around the time of the election accords us the opportunity to shine a light on the process of decision-making and the extent to which factors like corruption appeals played a role in their decision-making process. Therefore, to test our hypothesis, we commissioned the highly respected survey agency Alpha Research to carry out the data collection. The fieldwork for the survey of 1000 participants was carried out 22–31 July 2021, immediately after the election. The sample is a two-stage sample, stratified by region and type of settlement, randomly recruited respondents. The data was collected using F2F tablet-assisted personal interviews (TAPI) at the respondents’ homes. More details on survey and method choice are provided in the online appendix.

To test all components of our theoretical argument, we asked three questions specifically on corruption and the role corruption played in a respondent’s voting decision (see Table 2): first, how corrupt the respondents perceive the party they voted for to be; second, how much the voters generally dislike corruption; and third, how much voters take corruption into account when voting.⁸

Table 2. Questions asked in the survey.

Variable Name	Variable Description
Dependent Variables
Voteapril	Which party/coalition did you vote for in the parliamentary election on 04 April 2021?
Votejuly	Which party/coalition did you vote for in the last parliamentary election on 11 July 2021?
Vote	If the next general election for Parliament was next Sunday, how would you vote?
Corruption Variables
Q1	Corruption is often considered to be present in Bulgarian politics these days. On a scale from 1 to 5 how clean or corrupt would you perceive party [XX], the party you voted for, to behave in the next Bulgarian parliament?
Q2	How much do you dislike corruption on a scale from 1 to 5 (1 being I greatly dislike corruption to 5 - I don't mind corruption). Transformed: 1 don’t mind corruption and 5 greatly dislike corruption.
Q3	How much did your perception of corruption of party [XX] play a role in your decision to vote for the party [XX]? My perception of corruption for party [XX] 1 - affect my voting decision a lot to 5 - My perception of corruption for party [XX] did not affect my voting decision at all. Transformed: 1 did not take corruption into account when voting to 5 corruption affected my voting decision a lot.
Control Variables
age	Age of respondent varies between 18-87.
edu	Education: “University/college, High school, Elementary school and lower”
sex	Gender: “Male, Female”
income	Income: “Monthly income per HH member”

The sequence in which we ask questions about corruption is important in order to reduce potential social desirability response biases. For instance, if we were to ask voters first how much they dislike corruption, they might underreport how corrupt they perceive the party they voted for to be.⁹ Hence, we started by asking respondents first how corrupt they perceive the party they voted for to be and only ask afterwards how much they dislike corruption and how much they base their voting decision on corruption.

We recognize the threats of endogeneity—the explanatory variable correlated with the error term—in our work which can be caused by omitted variables, measurement error or simultaneity. As it is often the case in social science research, endogeneity is not something we can fully rule out, but we have tried to mitigate the threats as much as possible. More broadly, we recognize that the design of single country studies by definition “lack the cross-nationally comparative data that allow us to deal with endogeneity concerns” (De Vries 2017, 966) One common methodological approach to account for endogeneity caused by simultaneity would be to use instrumental variable models or a difference-in-difference approach or a lagged explanatory variable (often considered the least adequate approach). However, the data we have does not allow to run these specific models. Yet, we would also suggest that given we look at the voting intention data in the future, we are to a limited extent modelling the “dislike of corruption”, “perception of corruption” and “corruption taken into account when voting” for future voting intentions and thereby looking at a time sequenced decision-making process. Furthermore, we provide case study evidence as well as theoretical arguments to outline the direction of causality. As for omitted variables bias, we included standard voting control variables. Measurement error remains a possibility, but we worked with a highly trusted polling agency with decades of experience. Our results are a hoop test (see Van Evera 1997) for the relationship between dislike, perception and corruption taken into account in voting decisions.

To test our hypotheses, we used both regression and decision tree analysis as using both methods provide us with the best possible test for our interaction hypotheses. For the interaction hypotheses the decision tree adds some additional support for our proposition that the combination of the three variables is important. Although decision trees are known to outperform regression analysis when it comes to accurate predictions (see Cranmer and Desmarais 2017), they are usually inferior when determining the strength of the relationship of two variables as well as being unable to test for statistical significance. Hence, a combination of decision tree and regression analysis will give us the best of both worlds by providing us with empirical evidence of what successfully predicts election outcome (decision tree) and how strongly is the effect on election outcomes (regression). In line with other studies of corruption and voting we control for common voting variables such as age, gender, level of education, income, type of settlement (e.g. Agerberg 2020; Burlacu 2020; Charron and Bågenholm 2016; Riera et al. 2013) as well as survey weights.¹⁰

Empirical results

Our hypotheses predict first, the more voters dislike corruption the less likely they are to vote for a corrupt party, second, the more corrupt a party is perceived to be by a voter the less likely a voter is to vote for that party, and third, the more a voter takes corruption into account when voting the less likely they are to vote for a corrupt party. Our theoretical model further predicts that the combination of factors is important. If a voter is not bothered about corruption, all the other factors such as information acquisition, blame attribution and behavioural response no longer play a role. Only if the voter is bothered about corruption do information acquisition and taking corruption into account matter for a voter’s decision whether to vote for a corrupt party or not. Before we formally test our hypothesis 1-4, we provide some descriptive statistics for our corruption variables and voting decision for clean and corrupt parties. As we have data for three points in time (voting in April 2021, July 2021 and for the future election in autumn 2021), we can track where voters moved from one party to the next in Figure 3. The figure illustrates how voters change or plan to change their votes in the election. The left part of the figure illustrates the vote share for the different parties running in the election in April 2021, whereas the middle part illustrates who the respondents voted for in the July 2021 parliamentary election, and the right part illustrates their voting intention in the next election.

Figure 3. Voting decisions in April, July and future based on survey data collected¹².

Descriptive analysis – corrupt or clean?

We asked voters how corrupt they perceive the party they vote for to be. In the category “very clean” we see some stark differences. Whereas 70% voters of Stand Up! and 65% percent of “There is Such a People” considered the parties they voted for to be clean, only just over half of BSP voters (56%) and Democratic Bulgaria’s voters (52%) placed their chosen party in the same category. Strikingly fewer than half the voters of Borissov’s party GERB (44%) and fewer than a third of MRF voters (30%) placed the parties they voted for in the “very clean” category.

While our primary aim of this study is to shift the emphasis from expert evaluations of corrupt parties towards the perceptions of voters (independent of expert opinion), Hypothesis 1 necessitates relying on expert assessments to distinguish parties categorized as clean or corrupt by experts. To achieve that we rely on expert opinions on Bulgarian parties (e.g. Engler 2016; Engler, Armingeon, and Deegan-Krause 2021 or Zankina 2022) which have, as one can see from our data and analysis, notable differences with those of voters’ views. The differences come less in the ordering of the parties and more in the headline figures with voters’ less likely to see the party they voted for as clean. Whereas experts consider GERB and MRF to be parties with clear, evidential and longstanding corrupt cases in the past (Zankina 2022), however, many MRF and GERB’s voters disagree with the expert opinion by perceiving the party to be clean (see above). In contrast, the expert view of BSP sees the longstanding party that has been frequently in government as one guilty of corrupt activities, but their level of corruption in recent times makes it look like a “kindergarden” in comparison to MRF and GERB. Voters somewhat share this view: around half of the voters of BSP consider the party to be corrupt and half consider the party to be clean. As for the parties experts would consider to be clean, two of them “There is such people”, Stand Up! Mafia Out! and Democratic Bulgaria, offer no surprises: the vast majority of voters perceive them to be clean. Nonetheless, for Democratic Bulgaria there is a slight difference between the perception of experts and voters with experts rating it as cleaner than the voters.

The above disparities between the perception of voters and experts underline that to understand fully the role corruption plays in elections, we need to move away from experts’ perceptions towards a focus on citizens’ perception of corruption of parties. This is not, however, a general call to abolish expert ratings on corruption. On the contrary, corruption rating of experts, such as the party data set by Engler, Armingeon, and Deegan-Krause (2021) is very valuable to understand parties, their behaviour and strategies. Nonetheless, if we want to understand how corruption affects voters, we need to understand how voters perceive how corrupt parties are and focus on those perceptions rather than the view of experts. The descriptive data in our empirical analysis underlines that the vast majority of the respondents in our survey dislike corruption and yet, despite similar proportions of dislike spread across the political spectrum, we see that this dislike translates into very different voting behaviour. The following paragraphs will outline our empirical results in more detail.

Traditional regression analysis – inferential analysis

To test our first three hypotheses, we run ordinary multinomial probit regression model (Long and Freese 2014) with robust standard errors and survey weights for the data we have on the April 2021, July 2021 and future 2021 voting intentions of the voters.¹¹ We present the results for July below and refer to the other results for April and future vote 2021 in the online appendix.

The results in Table 3 and Figures 4(a) and 4(b) bear out our first hypothesis that the more a voter dislikes corruption the less likely the voter is to vote for corrupt party (1a) and more likely to vote for a clean party (1b). Both GERB and MRF, rated by experts as corrupt, shows the voters with a low dislike of corruption are more likely to vote for those parties (Figure 4(b)). Moreover, the results for the new parties show if a voter has a high dislike corruption, the more likely they are to vote for ITN but the results are less clear for Stand Up! Mafia Out! and Democratic Bulgaria (Figure 4(a)). For the latter two, the likelihood to vote for the parties is lower than for the corrupt parties. This weakens the support for hypothesis 1. It suggests that dislike of corruption alone is not enough to explain the voting support of the parties.

Figure 4. (a) Voters that have a high dislike of corruption and their voting intention for parties. (b) Voters that have a low dislike of corruption and their voting intention for parties.

Table 3. Main model: single effect - July 2021.

	Other Parties and No Voters	GERB	There is Such People Reference	BSP for Bulgaria	MRF	Democratic Bulgaria	Stand Up! Mafia Out!
	0	1	2	3	4
Party perceived to be corrupt	0.14	0.01		−0.10	0.35**	0.11	−0.06
	(0.12)	(0.10)		(0.12)	(0.13)	(0.11)	(0.15)
Voters dislike corruption	−0.29+	−0.44**		−0.17	−0.38*	−0.08	−0.05
	(0.16)	(0.13)		(0.15)	(0.15)	(0.15)	(0.18)
Perception of corruption played a	−0.33**	−0.41**		−0.24**	−0.47**	−0.16*	0.08
role in voting decision	(0.07)	(0.07)		(0.07)	(0.09)	(0.07)	(0.08)
Income distributed on each family	−0.04	0.25**		0.02	0.05	0.19*	0.12
member	(0.09)	(0.08)		(0.08)	(0.10)	(0.08)	(0.09)
Age	0.03**	0.03**		0.07**	0.02**	0.02**	0.04**
	(0.01)	(0.01)		(0.01)	(0.01)	(0.01)	(0.01)
Sex	−0.30	0.17		0.50**	0.10	0.14	0.01
	(0.20)	(0.17)		(0.18)	(0.22)	(0.18)	(0.22)
Education	−0.48**	0.02		0.01	0.77**	−0.58**	−0.32+
	(0.15)	(0.14)		(0.14)	(0.19)	(0.15)	(0.17)
Constant	1.56	0.37		−2.73**	−1.10	−0.78	−2.64*
	(1.05)	(0.95)		(1.04)	(1.17)	(1.05)	(1.26)
Observations	773
Count R^2	.
Adj. Count R^2	.
AIC	2360
BIC	2611

Note:+significant at 10; *significant at 5; **significant at 1. Robust standard errors are in parentheses. Estimations performed using Stata 15.

Our second hypothesis suggested that the more corrupt a respondent perceives a party to be, the less likely he/she is to vote for a corrupt party. While the results for GERB (Figure 5(a)) and BSP (Figure 5(b)) offer some limited support for this hypothesis, the results for ITN (Figure 5(c)) and MRF (Figure 5(d)) challenge it. The results show the more corrupt a respondent perceives MRF to be the more likely they are to vote for MRF. This result may be linked to MRF’s ethnic and regional appeal. It may be simply that ethnicity trumps all other factors for MRF voters or, perhaps uncharitably, that less than clean activities may be linked to bringing home the bacon to ethnic Turks and regions with a high proportion of ethnic Turkish voters.

Figure 5. (a). Perception of Corruption and voting for GERB. (b) Perception of corruption and voting for BSP. (c) Perception of corruption and voting for “There is such People”. (d) Perception of corruption and voting for MRF.

Our third hypothesis suggested the more voters base their voting decision on issues around corruption, the more likely they are to vote for a clean (3a) and the less likely they are to vote for a corrupt (3b) party. Respondents were asked to rate the impact of corruption on their voting on a 1–5 scale, where 1 = corruption affected my voting decision very little to 5 = corruption affected my voting decision a lot. We find very clear support for the parties experts perceive as corrupt i.e. MRF (Figure 6(a)) and GERB (Figure 6(c)) as well as for the parties experts perceive to be clean ITN (Figure 6(b)) and “Stand up!” (Figure 6(d)). In other words, if corruption is taken into account for the voting decision this has a more consistent effect in explaining how corruption affects electoral outcomes than whether a voter perceives a party to be corrupt or their dislike for corruption indicating this is the only corruption variable of the three that seems to matter.

Figure 6. (a) Perception of corruption played a role in voting decision for GERB. (b) Perception of corruption played a role in voting decision for “There is such People”. (c) Perception of corruption played a role in voting decision for MRF. (d) Perception of corruption played a role in voting decision for “Stand up!”.

Our fourth hypothesis suggests that whether a voter is bothered about corruption (i.e. dislikes corruption) as well as how corrupt a voter perceives the party condition the effect of whether a voter votes for a corrupt or clean party. Figure 2’s theoretical decision tree summarises our theoretical predictions and Table 4 shows the estimation of our multinomial regression analysis.

Table 4. Main model: interaction effect - July 2021.

	Other Parties and No Voters	GERB	There is Such People	BSP for Bulgaria	MRF	Democratic Bulgaria	Stand Up! Mafia Out!
	0	1	2	3	4	5	6
Party perceived to be corrupt	1.34*	0.40		0.49	0.94	1.40*	0.13
	(0.60)	(0.53)		(0.61)	(0.66)	(0.57)	(0.63)
Voters dislike corruption	0.39	1.69*		0.06	1.54	2.31*	−0.12
	(1.22)	(0.77)		(0.99)	(1.11)	(0.92)	(0.89)
Party perceived to be corrupt *	−0.67	−0.36		−0.17	−0.51	−1.19**	−0.07
Voters dislike corruption	(0.48)	(0.33)		(0.43)	(0.45)	(0.44)	(0.35)
Perception of corruption played a	0.39	0.71*		0.11	0.43	0.46	−0.29
role in voting decision	(0.35)	(0.27)		(0.34)	(0.36)	(0.30)	(0.35)
Party perceived to be corrupt * Perception of	−0.28+	−0.08		−0.07	0.01	−0.20	0.01
corruption played a role in voting decision	(0.16)	(0.14)		(0.18)	(0.16)	(0.14)	(0.20)
Voters dislike corruption * Perception of	0.10	−0.23		0.24	−0.03	−0.33	0.20
corruption played a role in voting decision	(0.29)	(0.19)		(0.25)	(0.26)	(0.23)	(0.21)
Party perceived to be corrupt * Voters dislike	0.12	0.06		−0.04	0.03	0.20*	−0.03
corruption * Perception of corruption played a role in voting decision	(0.11)	(0.07)		(0.11)	(0.10)	(0.10)	(0.09)
Income distributed on each family member	−0.05	0.27**		0.00	0.02	0.18*	0.12
	(0.09)	(0.08)		(0.08)	(0.10)	(0.08)	(0.09)
Age	0.03**	0.03**		0.07**	0.02**	0.02**	0.04**
	(0.01)	(0.01)		(0.01)	(0.01)	(0.01)	(0.01)
Sex	−0.36+	0.15		0.49**	0.14	0.12	0.01
	(0.19)	(0.17)		(0.18)	(0.22)	(0.18)	(0.22)
Education	−0.48**	0.06		0.01	0.80**	−0.57**	−0.32+
	(0.15)	(0.14)		(0.14)	(0.19)	(0.15)	(0.17)
Constant	−2.62+	−6.31**		−5.52**	−7.50**	−4.51**	−2.36
	(1.54)	(1.21)		(1.41)	(1.62)	(1.31)	(1.56)
Observations	773
Count R^2	.
Adj. Count R^2	.
AIC	2405
BIC	2740

Note:+significant at 10; *significant at 5; **significant at 1. Robust standard errors are in parentheses. Estimations performed using Stata 15.

The results support hypothesis 4. If voters are not bothered about corruption, neither corruption perception nor whether voters take corruption into account affects the voting behaviour for corrupt (and clean parties) as illustrated in Figure 8. The situation is different if voters dislike corruption (Figure 7). If voters dislike corruption / are bothered about corruption, they are less likely to vote for parties they perceive to be corrupt (GERB, BSP) and more likely to vote for clean parties (ITN). The more a voter then takes corruption into account when voting, the less likely the voter is to vote for a party he/she perceives to be corrupt (GERB, MRF) and the more likely the voter is to vote for a party he/she perceives to be clean (ITN). Whether voters are bothered greatly affects the role corruption plays in their voting decision and ultimately affects the outcome of a parliamentary election.

Figure 7. Voters that dislike/bothered about corruption; (a): voting for GERB (b): voting for there is such people (c): voting for BSP (d): voting for MRF (e): voting for Democratic Bulgaria.

Figure 8. Voters that do not dislike/not bothered corruption; (a): voting for GERB (b): voting for there is such people (c): voting for BSP (d): voting for MRF (e): voting for Democratic Bulgaria.

Robustness

We reran the analysis for the voting decision in April 2021 as well as future voting intension. The results are available in the online appendix (Table A1.a-b for single effect, A2.a-b for interaction effects). We also did a matching analysis (Table A3). Observations matched on whether voters perceive the party to be corrupt or not (binary 0 corruption perception 1–2 and 1 corruption perception 3-4). Perception of corruption of the party is not relevant for an explanation of whether a voter votes for a corrupt party or not. This finding further supports our contention that voters’ perceptions of corruption of parties are less relevant than one might think. Our results remain robust for these estimations lending further support to our conclusion.

To explore the importance of this “botheredness” further as well as to see how corruption plays out in the voting decision with respect to the control variables, we ran a decision tree analysis (see the online appendix). Four variables are identified: age; perception played a role in voting decision; dislike of corruption; and education. Older respondents of 66 and above are more likely to vote for a corrupt party and this applies to 22 percent of the respondents (Figure A14). Respondents from the older than 66 cohort seem to be most likely to vote for one of the corrupt parties (GERB, MRF, and BSP). When interpreting the results, we recommend a dose of caution. We suspect that part of our results are driven by the long tradition BSP has in Bulgarian politics (stretching back into communist times) and the older age electoral cohort the party has tended to attract over the years.

To test the robustness of our results further, we move all abstainers into a separate category and rerun the analysis (Tables A5–A7, Figures A15-A20). With this altered coding, our findings not only maintain their robustness but also exhibit increased and more consistent statistical significance in certain areas. Figure A19 in the online appendix, for instance, shows results for ITN now achieve statistical significance even in the context of future voting data. None of our visual representations would suggest or support the expectation that voters would abstain more often when they perceive heightened corruption levels. (see Figure A15–A20).

In fact, the results even imply that individuals perceiving lower corruption levels are more inclined to abstain (see Figure A17). We leave it to future research to explore these interesting findings further.

Conclusion

Corruption is frequently invoked as an explanation for electoral outcomes in Bulgaria, across Central and Eastern Europe, and indeed more widely. But hitherto most focus on the role of corruption in elections has been on the object of choice. This article has sought to contribute to redressing the balance by focusing on voter choice. We have argued theoretically, and shown empirically, that the role corruption plays in elections is best understood by breaking the voting decision into four elements i.e. it is not as simple as voters disliking corruption and ergo holding politicians accountable for their corrupt actions at election time. Moreover, only with a specific combination of answers at these decision points will a voter choose not to vote for the corrupt party. Indeed, our empirical results highlight the merits of examining the decision-making stages a voter goes through when casting a ballot against a corrupt party, and underlines that corruption impacting actual voting decisions may be far less likely theoretically and empirically than one might think. Nonetheless, it does happen. No fewer than 31% of our respondents met all of the conditions in our model.

Our research does provoke a series of further questions that could be the focus of future studies not least why respondents take corruption into account to a varying degree. Two avenues may be particularly fruitful. Firstly, from the literature we know that voters sometimes perceive a party to be the lesser evil (Agerberg 2020) and may turn a blind eye towards the substantial corruption the party they support is committing, especially if corruption is perceived to be rampant. It may be the case that stability and experience even with a large slice of corruption may be considered a price worth paying even for voters for whom corruption is a major concern. We do not have the data to say for certain, but we would suggest there is at least some circumstantial evidence to say at least a portion of voters in Bulgaria think along those lines. Secondly, the reasons why some voters weigh corruption more strongly in their voting decision may be linked to the role of cross pressures on voting choice (Dassonneville 2023). In order to test that in a compelling manner requires a dataset covering a series of elections over many years. By not defining corruption in our survey, we cannot determine what type of corruption voters take into account whether that be “petty” or “grand”. Future research could specify different forms of corruption and how they affect voting decisions. Finally, the literature presents potential explanations that remain untested due to limitations in our dataset. One such area pertains to voting abstention and in particular the extent to which voters are already refraining from elections due to a fatalistic perspective on corruption. The role of corruption perceptions in voting abstention is clearly a potentially fruitful research furrow to plough.

Despite the limitations of an in-depth study of one country, our framework and approach have analytical purchase beyond the confines of a relatively small country in South-East Europe. Firstly, by breaking down the process of decision-making we suggest scholars can better understand both why corruption does matter in elections, but also why it does not when we might have expected it to. Moreover, by looking at the elements of decision-making may help tease out what factors may be driving particular electoral behaviour. Our decision tree analysis, for instance, points to the importance of age, indicating the merits of exploring further whether this is an age (or cohort) effect.

Secondly, our results could indicate that some findings generated from experiments may be a little misleading. Our decision tree analysis, for instance, could be interpreted to suggest that the focus on revealing information about corruption in those experimental studies may be less important than other stages (or components) when a voter makes a decision in an actual election. We would encourage future studies to take on board our combinatorial framework when devising research to examine the role of corruption and information about corruption in electoral politics. Clearly corruption does matter to some voters. In Bulgaria, across CEE, and indeed more widely, there are such people.

Ethical review

Alpha Research carried out the data collection according to the standard ethical practices of opinion polling in Bulgaria.

Acknowledgements

We are grateful to Alpha Research for conducting the poll for us and to Temple University and the University of Birmingham’s School of Government Research Fund for providing funds for the survey. We also extend our thanks to members of the POLSIS writing group for providing a friendly and supportive environment that helped us carve out time to write this article when confronted with challenges associated with the coronavirus pandemic. We would like to thank the editor and two anonymous reviewers for valuable feedback on earlier versions of the article.

Disclosure statement

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

Data availability statement

Data set and replication files are available on the corresponding author’s university website.

Additional information

Funding

This research was supported by the University of Birmingham's School of Government Research Fund and Temple University Rome.

Notes on contributors

Natascha S. Neudorfer

Natascha S. Neudorfer is Professor of Political Economy, Heinrich-Heine-Universität Düsseldorf (2023-), Associate Professor University of Birmingham (2020-2023).

Tim Haughton

Tim Haughton is Professor of Comparative and European Politics and a founding co-director of the Centre for Elections, Democracy, Accountability and Representation (CEDAR) at the University of Birmingham.

Emilia Zankina

Emilia Zankina is an Associate Professor in Poliical Science, Vice Provost for Global Engagement at Temple University and Dean of Temple University Rome.

Notes

1 See here https://www.euractiv.com/section/justice-home-affairs/news/twenty-people-dominate-corruption-in-bulgaria-pm-tells-meps/ or here https://www.euractiv.com/section/politics/news/greco-calls-out-bulgaria-for-lack-of-fight-against-corruption/ (accessed 4 March 2023).

2 Following Transparency International (2020), “[w]e define corruption as the abuse of entrusted power for private gain.”

3 In countries like Bulgaria, corruption is widely discussed both amongst ordinary citizens and in the media, hence we assume voters have access to information about corruption. The processing of information has been studied by other scholars (Winters and Weitz-Shapiro 2020; Weitz-Shapiro and Winters 2017). In contrast, our aim is to address overlooked aspects using observational data, amending the debate theoretically and empirically by examining the dislike of corruption, the actual conclusions voters draw about the corruption of a party (i.e., whether the party is corrupt or clean), and whether voters take corruption into account when voting.

4 Although this formulation may sound tautological, we argue that corruption is but one of the many salient issues a voter considers. When corruption is high in a voter’s hierarchy of preferences, then it plays a greater role in the decision making process of the voter and ultimately the voter's choice. As demonstrated throughout the article, many voters who care about corruption still vote for corrupt parties.

5 https://www.theguardian.com/world/2017/dec/28/bulgaria-corruption-eu-presidency-far-right-minority-parties-concern

6 In the 2021 elections GERB ran formally as GERB-SDS reflecting the alliance of GERB with the Union of Democratic Forces. GERB, however, was the overwhelmingly dominant part of the alliance so for the remainder of this article we use the label GERB

7 Further reasoning can be found in the online appendix.

8 We have not included Party Identity (PID) into our model in part because the party landscape in CEE is so volatile with many parties with very short shelf lives (see Haughton and Deegan-Krause 2020).

9 Another concern could be that not admitting the party one voted for is corrupt could be due to social desirability bias. We would argue for a social desirability bias to exist the topic needs to be sensitive or carry a stigma. Corruption does not carry a high stigma in Bulgaria as it maybe does in some other countries. The feeling of corrupt actors might be “everybody does it so why shouldn’t I?”. In the context of Bulgaria, we are not talking about one specific corruption scandal and an environment where corruption is the except, but about a country with a continuous flow of corruption scandals and an environment where corruption is the norm.

10 See the online appendix for details of how we dealt with abstentions.

11 To test further the soundness of our statistical model, we ran additional tests: changed the estimation of a multinomial probit to a multinomial logit model; took out the robust standard errors; and ran an Independence of Irrelevant Alternatives test. None of the tests were statistically significant.

12 Graph created using this website: https://app.rawgraphs.io/ recommended through this website: https://www.azavea.com/blog/2017/08/09/six-sankey-diagram-tool/