Introduction
Between 2017 and 2019, citizens throughout Romania poured into the streets to protest the government on nearly 40 different occasions, a rate not seen since the fall of the communist regime. The protests were initially a response to a government ordinance, passed in the middle of the night, that raised the bar on what constitutes corruption as a crime, and which significantly reduced the punishment for those involved; a reform, opponents argued, that in effect decriminalized corruption. While in this case some high-level politicians initially paid a price via the legal system (including the sentencing of ex-prime minister Liviu Dragnea), in general, most politicians indicted in political corruption crimes do not pay a price at the ballot box, despite sustained high levels of publicly expressed indignation against corruption (Bågenholm and Charron Reference Bågenholm and Charron2020; Klašnja Reference Klašnja2015; Mungiu-Pippidi, Reference Mungiu-Pippidi2018). This clear incongruity, known in the literature as the ‘corruption-voting paradox’, has received much scholarly attention during the last decade (for an overview, see, for example, Bågenholm Reference Bågenholm2021; Incerti Reference Incerti2020). Despite the fact that survey after survey demonstrates that the overwhelming majority of the public detest corruption and perceive it as a major problem in their countries, incumbents who have been involved in corruption scandals nonetheless get re-elected in democratic elections at a surprisingly high rate. While there is a consensus among scholars that corrupt politicians generally lose support, the loss of a subsequent election is rare. Thus, despite the opportunity to ‘throw the rascals out’ in democratic elections, citizens oftentimes collectively fail to do so when it comes to punishing corruption.
The existing literature has put forth several plausible explanations as to why this is the case, and scholars have demonstrated relatively consistent findings, highlighting the importance of credible information, partisanship, and trade-offs, among others (De Sousa and Moriconi Reference De Sousa and Moriconi2013; De Vries and Solaz Reference De Vries and Solaz2017). Quite naturally, most studies have focused on voters, ie those individuals who actually turn out to vote on election day, and whether and why they cast their ballot for the corrupt incumbent or an opposition candidate. Less attention, however, has been paid to the importance of turnout and the potential role that nonvoters can play by their decision to withhold support for either the incumbent or opposition candidate, thereby indirectly helping one of the two sides to win. In this study, we center our focus on electoral contexts where the turnout is low, and in which, we argue, abstention plays a particularly crucial role for accountability (or lack thereof) in relation to political corruption. Effective mobilization of a rather small group of loyal partisans can thus be enough to secure re-election in plurality systems. For example, in the case of Romania, in the 2016 and 2020 local elections, on average, just 16.7 per cent and 16.3 per cent, respectively, of the total number of eligible voters in a municipality decided the winner (Marin Reference Marin2021).
In this paper, we first show that despite a strong notional anti-corruption norm among Romanian voters, knowledge about political corruption of mayors does not affect incumbent voting in the 2020 municipal election, which is completely in line with the empirical literature (Incerti Reference Incerti2020). As in past years, this lack of reaction to corruption led to a break-down in accountability, as many indicted incumbents were re-elected (Bågenholm and Charron Reference Bågenholm and Charron2020). While Boas et al. (Reference Boas, Hidalgo and Melo2019) have aptly demonstrated this feature in a single study, the context of Romania is considerably different from their Brazilian case, with mandatory voting and high turnout, and thus perhaps suggesting alternative theoretical mechanisms to explain this disparity.
We explain this by applying a collective action logic with the focus on low turnout and hence on nonvoters, who previously have been overlooked in the corruption-voting literature. In line with the collective action logic, we argue that people who believe that most others do not care about punishing corruption have a strong tendency to abstain on election day, despite their detestation of corruption. By doing so, they allow for a relatively small number of incumbent loyalists, who care little about corruption and willingly continue to support their preferred candidate, to determine the outcome of the election. In order to make electoral accountability work in this respect, it is crucial to overcome the collective action problem by convincing voters that most others are willing to punish corruption, as they say they will when asked in surveys. The low turnout, however, is a catch-22. Until there is clear evidence that turnout is increasing, most people, particularly non-incumbent supporters and previous nonvoters – will assume that staying at home is rational, and if they do so, turnout will not increase, and in turn, corrupt incumbents will remain in power. Thus, to fully address the corruption-voting paradox, we cannot only ask why certain voters continue to vote for incumbents they know to be corrupt, but we must also ask why others choose to abstain altogether.
In addition to focusing on nonvoters, our study makes several novel contributions. First, empirically, we fielded a novel two-round panel survey in Romania just before and after the municipal elections in the fall of 2020. Thus, in contrast to much of the research on voting behavior, which is based on hypothetical scenarios, our study is embedded in a real-world election.
Second, we establish a clear contrast between stated norms of anti-corruption and actual behavior. Given that recent empirical work has demonstrated that indicted mayors have been quite successful in getting re-elected in this context (Bågenholm and Charron Reference Bågenholm and Charron2020), we first establish whether there is an anti-corruption norm in Romania, which we see as a necessary condition for electoral accountability in the face of corruption. In other words, if there is a passive acceptance of political corruption in this context, we would then not expect ours or any other of the theoretical explanations from the literature to carry any weight. Building on Boas et al. (Reference Boas, Hidalgo and Melo2019), we embedded a conjoint experiment in the survey to test the degree to which voters reject corruption, while also asking about the most pressing issues in the upcoming elections. We found, via our survey experiment, that there is an overwhelming disdain for corruption, and the direct question about issue salience asked in our survey shows that corruption is the most salient issue among a plurality of respondents.
Third, we then tested the degree to which hearing of one’s mayor being indicted on corruption charges affects people’s vote choice, accounting for respondents’ past voting behavior. Prior to the survey, we collected data from the Romanian Anti-Corruption Body, DNA (Direct,ia Nat,ională Anticorupt,ie) on all municipalities with an indicted mayor for the period 2016 to 2020. Our results show that hearing about corruption indictments did not affect respondent vote choice in the 2020 election, despite the overwhelmingly strong norms against corruption in Romania. Less surprisingly, we also found that loyal incumbent voters tend to support their candidates, regardless of corruption indictments. This implies that in order to ‘throw the rascals out’, a sufficient number of opposition supporters, together with nonvoters, need to be mobilized.
Our fourth main contribution is theoretical. Given the observed break-down of the accountability mechanism despite a strong anti-corruption norm among Romanian voters, we posit a novel argument within the corruption-voting literature, focusing on nonvoters. Our theoretical expectations are based on the logic of collective action, which, in short, implies that people act according to their anticipation of how others will act (Mungiu-Pippidi Reference Mungiu-Pippidi2011; Persson et al. Reference Persson, Rothstein and Teorell2013; Tucker Reference Tucker2007). While previous literature has highlighted that a lack of suitable alternatives can lead to higher abstention (Agerberg Reference Agerberg2020), our research design relies on real-world election data rather than a hypothetical scenario, thus enhancing external validity.
Finally, and perhaps most importantly, we build on the literature of political anti-corruption messaging to assess our collective action hypothesis. The field of anti-corruption messaging, which is backed by large NGO’s and civil society organizations, aims to sway citizens away from engaging in petty corruption and to influence the public to demand cleaner governance in general. While the effects of anti-corruption messaging on reducing citizen participation in bribery is mixed, or even contrary to the intended goals (Cheeseman and Peiffer Reference Cheeseman and Peiffer2022; Corbacho et al. Reference Corbacho, Gingerich, Oliveros and Ruiz-Vega2016; Köbis et al. Reference Köbis, Troost, Brandt and Soraperra2019), few studies to date has investigated how messaging could influence actual electoral responses to political corruption (see Chong et al. Reference Chong, De La O, Karlan and Wantchekon2015) as an example). Using an experimental design, we provide a short anti-corruption message based on the logic of ‘nudging’, which seeks to alter behavior, without forbidding any alternative behavior or altering any economic incentives (Thaler and Sunstein (Reference Thaler and Sunstein2009). In addition, we build on previous work that highlights the conditional effects of the nudging information (Bicchieri and Dimant Reference Bicchieri and Dimant2019) in that the effects of such anti-corruption information on voting behavior are conditional based on citizens’ underlying beliefs of how they believe ‘others will act’.
In line with our hypothesis, the more pessimistic voters, ie those who believe that others will not come out to vote against corrupt incumbents, tend to abstain on election day. These findings thus shed further light on how the accountability chain breaks down in retrospective corruption voting, as well as why corruption levels are sticky and hard to change. Our more optimistic finding is that the aforementioned respondents tend to increase their participation rate when provided with nudging information regarding the intention of other citizens to vote against corrupt incumbents. The main policy implication from our findings is that information about corruption per se is not enough for people to engage and come out and vote to hold corrupt politicians accountable. They also need to be persuaded that most other people will do so too. To break this negative impasse, we thus need to overcome the collective action problem among the (majority) group of non-incumbent leaning supporters.
The paper is organized as follows. The next section gives a brief review of the most commonly suggested explanations as to why corruption voting occurs. Thereafter, we discuss the collective action problems associated with both corruption in general and voting and non-voting and present our novel argument on how this relates to the corruption voting puzzle and electoral accountability. We then turn to data and methods, followed by analysis, and finally our discussion and conclusions.
Explaining corruption voting
Although the by now large body of studies on corruption voting has become more diverse, it still consistently shows that politicians and parties implicated in corruption scandals are punished to a far lesser extent than one would think, given voters’ general detestation of corrupt practices. Oftentimes, such politicians and parties get re-elected. This so-called ‘corruption-voting puzzle’ has generated an increasing amount of research that tries to discern why this is the case (see Bågenholm Reference Bågenholm2021; De Sousa and Moriconi Reference De Sousa and Moriconi2013; and De Vries and Solaz Reference De Vries and Solaz2017 for extensive overviews). It has been argued that credible information about corruption cases is a necessary condition in order for voters to correctly identify who is to blame for the scandal and who should be punished for it (De Vries and Solaz Reference De Vries and Solaz2017; Jiménez and Cainzos Reference Jiménez and Cainzos2006: 194). Yet, there is a strong tendency among voters to disregard information discrediting their favorite party, the so called ‘home team’ or ‘partisan’ effect (see De Sousa and Moriconi Reference De Sousa and Moriconi2013: 486; Jankowski et al. Reference Jankowski, Juen and Lewandowsky2022) or, as de Figueiredo et al. (Reference de Figueiredo, Hidalgo and Kasahara2023) demonstrate, that anti-corruption opposition parties are punished harder if engaged in corruption by their own supporters, than are incumbents. In cases where corrupt behavior is too obvious to deny, there is a tendency to think that rival parties or candidates are equally corrupt (Pavão Reference Pavão2018).
Another factor that may undermine electoral accountability in the face of corruption is that bad behavior may be mollified or even superseded by incumbent performance. The economic voting literature consistently finds that governments that manage the economy well, or are perceived to do so, have a better chance of re-election (see, for example, Lewis-Beck and Stegmaier Reference Lewis-Beck and Stegmaier2000) and in turn, the corruption voting literature specifically shows that voters tend to turn a blind eye to corrupt behavior under such circumstances. Conversely, a government will be negatively affected by corruption when it is perceived to mismanage the economy (Choi and Woo Reference Choi and Woo2010; Zechmeister and Zizumbo-Colunga Reference Zechmeister and Zizumbo-Colunga2013). There is, thus, some kind of rationality in the electorate, insofar as people who think that they will personally benefit from government policies will also continue to support it, regardless of corruption, if they find it to be to their advantage (Fernández-Vázquez et al. Reference Fernández-Vázquez, Barberá and Rivero2016; Manzetti and Wilson Reference Manzetti and Wilson2007). A recent study moreover shows that voter perceptions of the incumbents being both incompetent and corrupt double the electoral punishment in relation to just being perceived as corrupt (Rienks Reference Rienks2023).
Finally, a third strand of the literature highlights the importance to voters of an alternative ‘clean’ and reasonably ideologically close party. Without any viable alternatives, voters will let ideology trump corruption as the sacrifice will be too big. Recent studies have shown that party systems with a higher number of effective parties, which increase the likelihood that voters will find a reasonably close alternative, actually enable voters to switch from their old preferred party to a clean one to a higher extent compared to systems with few and ideologically polarized parties (Charron and Bågenholm Reference Charron and Bågenholm2016; Elia and Schwindt-Bayer Reference Elia and Schwindt-Bayer2022; Vera Reference Vera2022). The same results are found in experimental studies that let the participants choose between two candidates with randomly assigned credentials (Agerberg Reference Agerberg2020). Moreover, voters may ignore incumbency corruption, even when they are perfectly aware of it and when there are plenty of alternatives available. For example, as Pavão (Reference Pavão2018) has shown, voters in highly corrupt countries tend to believe that all candidates are equally corrupt, making corruption a meaningless issue on which to base one’s selection (Pavão Reference Pavão2018). In short, confronting a party involved in corruption is not as straightforward as it may seem, as there are many hurdles to overcome.
We do not take issue with the above explanations. Rather, we identify previously overlooked aspects that can contribute to our understanding as to why corrupt incumbents often get re-elected, and we thus address several gaps in this literature. First, previous research has primarily focused on whether voters support the corrupt incumbent or the ‘clean’ challenger. Far less attention has been paid to nonvoters and the effect their decision to stay at home on election day may have. In short, by being inactive, citizens are also making an impact on the effectiveness of electoral accountability. In low-turnout electoral contexts such as Romania, we find this argument to be highly relevantFootnote 1 . Second, we put forth a novel perspective within this field of research, namely a collective action approach, implying that most people’s actions will be dependent on the anticipation of how others will act. Third, the literature has mostly tested the effects that information about corrupt incumbents has on voters’ electoral behavior, whereas we test the effects of information about the intent of other voters in response to political corruption. Finally, the experimental literature in particular mainly relies on hypothetical survey experiments that contrast the findings of field experiments (Incerti Reference Incerti2020), while, by way of contrast, we embed our study within a ‘real-world’ election.
Corruption, voting, and abstention as collective action problems
It has been argued that fighting corruption in general is a collective action problem (Mungiu-Pippidi Reference Mungiu-Pippidi2013; Persson et al. Reference Persson, Rothstein and Teorell2013; Tucker (Reference Tucker2007). For example, as Warren states, ‘Corruption, it is increasingly noted, breaks the link between collective decision-making and people’s powers to influence collective decisions through speaking and voting, the very link that defines democracy’ (Warren Reference Warren2004: 328). As mentioned above, surveys consistently show that most citizens across the globe express abhorrence for corruption, but their willingness to avoid engaging in corrupt activities is contingent on how they perceive that others will act (Persson et al. Reference Persson, Rothstein and Teorell2013). This implies that in contexts where corruption is more endemic, most people will take part in corrupt transactions simply because they perceive that others will, and that by not participating in the corruption game, there is a fear of the ‘sucker’s payoff’ (Karklins Reference Karklins2016). By focusing on the difficulties of changing the status quo and the existing stable sub-optimal equilibria, the collective action approach to analyzing corruption thus helps us to understand why controlling corruption is so problematic.
Similarly, we argue that the corruption-voting puzzle can be understood in collective action terms. There is much written on why rational citizens turn out to vote, given the infinitesimal chance that their individual vote could be decisive (Downs Reference Downs1957, Olson Reference Olson2009). It has been argued that ‘in a polity comprised of rational citizens who economize on the resources they devote to politics, the logic of collective action dictates inaction unless special inducements are offered’. (Green and Smith Reference Green and Smith2003). The so-called voting paradox (Ferejohn and Fiorina (Reference Ferejohn and Fiorina1974), ie that voters actually turn out to vote, has been much debated, and is mostly beyond the scope of this paper. We instead advance the argument that voters faced with corrupt incumbents will be subject to a similar social trap as described above. From a rational standpoint, it makes little sense to expend energy to vote against a corrupt incumbent if you believe that very few others will do the same. It is only with a collective effort that this can happen, and the likelihood that such a concerted effort will be made is smaller in countries with low electoral turnout, where there is a track record of turning a blind eye to corrupt incumbents and, finally, where the incumbent advantage and vote margins are large, ie precisely the conditions that are present in the context of Romanian municipal elections.
More broadly, from the literature on social movements, several explanations on mechanisms for why people act collectively have been put forth. Most prevalent is arguably the notion of efficacy, ie the perception that the efforts will lead to a successful outcome (Bandura Reference Bandura2000; Klandermans and Oegema Reference Klandermans and Oegema1987), which in our study is the removal of corrupt incumbents through elections. Another strand that has challenged this explanation is social identification, ie the extent to which people identify with the movement, organization, or the cause in general (Hornsey et al. 2006; Stürmer et al. Reference Stürmer, Simon, Loewy and Jörger2003). However, social movements have a different dynamic than voting, as the participating in demonstrations or meetings are more direct and visible activities, where the level of engagement and thus, the potential efficacy and identification becomes easier to assess, compared to the decision whether to vote or not. Yet what both types of collective action share is the uncertainty regarding the intention of other members, in that they can only observe what others do in public, reducing the likelihood of collective action (Tankard and Paluck Reference Tankard and Paluck2016). In terms of identification, we believe that unless anti-corruption manifests itself in observable protests, the sense of belonging to a movement is a less salient factor when deciding to vote or not. We thus argue that the potential efficacy is tightly connected to the perceptions of whether others would vote, which could be assessed by protests against corruption, but more likely by experiences from previous elections, which, as noted above, returned corrupt incumbents with low electoral turnout.
In this context, we build on a set of studies that focus on institutional (dis)satisfaction and participation in democracies. In general, at both the aggregate and micro-levels, there is a consensus around the idea that negative voter assessment of political institutions, such as trust, perceptions of corruption/fairness, satisfaction with democracy, etc., generates less democratic participation (for example, Birch Reference Birch2010; Carlin Reference Carlin2011; Carreras and Vera Reference Carreras and Vera2018; Ceka Reference Ceka2013; Chong et al. Reference Chong, De La O, Karlan and Wantchekon2015; Sundström and Stockemer Reference Sundström and Stockemer2015). The reasons behind the idea that corruption demobilizes voters rather than engages them are notably increased signals of illegitimacy, cynicism, and decreased political efficacy, which in turn leads to voter resignation, in particular in contexts with relatively poorer performing institutions (Dahlberg and Solevid Reference Dahlberg and Solevid2016; Snegovaya Reference Snegovaya2020)Footnote 2 .
The corruption-voting literature in general tends to treat voters as homogeneous, without consideration for past electoral support, with partisanship being a notable exception (Anduiza et al. Reference Anduiza, Gallego and Muñoz2013). In line with this qualification, incumbent supporters could even be mobilized by some aspects of machine-type politics (e.g. Scott Reference Scott1969), as they may have a material incentive to get the incumbent re-elected, given that clientelism, patronage, selective non-enforcement of the law, and other tools at local leaders’ disposal are likely to benefit such voters; a practice named the logic of ‘rewarding loyalists’ (Gans-Morse et al. Reference Gans-Morse, Mazzuca and Nichter2014; Kitschelt and Wilkinson Reference Kitschelt and Wilkinson2007). In a context such as Romania, Mares and Young (Reference Mares and Young2019) show that incumbents continue to receive support from loyal supporters in exchange for clientalist goods or services, rendering certain voters dependant on sitting politicians, irrespective of corruption, with resources at politicians dispense coming from public coffers or even indirectly from private campaign donors benefiting from favorable procurement (Gherghina and Volintiru Reference Gherghina and Volintiru2017)Footnote 3 . Based on previous election outcomes, they can thus confidently assume that many of their fellow loyalists will turn out to vote and that by doing so, they will most likely secure victory for their preferred candidate. Thus, of most salience for electoral accountability in such contexts is the group of non-incumbent supporters, i.e., either those who voted for the opposition in the previous election or those who abstained. These voters, who constitute the overwhelming majority in most Romanian elections (Marin Reference Marin2021)Footnote 4 , have arguably less incentives to turnout, since they will most likely not succeed in getting their preferred candidate elected and are less likely to receive clientelistic goods, and moreover, their success in terms of reducing corruption is non-excludable.
We argue that the story about the limits of electoral accountability for substantial indications of corruption runs along the following lines: (1) Loyal incumbent voters are unaffected by information and knowledge about corruption indictments against their preferred candidates and they continue to support him/her. (2) In a context where turnout is generally low and where incumbents usually succeed in re-election efforts, regardless of corruption allegations or indictments, non-incumbent supporters will be less likely to vote, as they perceive that most other people will not vote, thus making a strategic voting rational calculation (Cox Reference Cox1997: 3), even though they express distaste for political corruption. Consequently, (3) the turnout will remain low and the corrupt incumbent will be reelected with a sufficient plurality vote. This, in turn, creates a vicious cycle, whereby non-incumbent (and incumbent) supporters make the same assumption and the same decision in the next election, thereby making a reduced level of corruption in general a less likely outcome. This brings us to our first empirical hypotheses:
H1: Despite voters expressing strong anti-corruption norms in the abstract, knowledge of corruption will not have an effect on voters’ behavior in ‘real world’ elections, ceteris paribus.
H2: Non-incumbent supporters who believe that others will be less likely to turn out to vote against corrupt politicians will be more likely to abstain from voting, compared with more optimistic voters.
In addition, we argue that while broad institutional assessments, such as low political trust, dissatisfaction with democracy, or higher perceptions of corruption, etc., are important, there are more specific sets of perceptions relevant to corruption-voting, namely the perceived action of other citizens. To overcome some of these collective action problems in engaging citizens in anti-corruption efforts in general, prominent NGOs and anti-corruption agencies have promoted messaging campaignsFootnote 5 . Based on the logic of ‘nudging’, such anti-corruption messages are intended to encourage citizen behavior via information about the behavior of other citizens (Manning Reference Manning2009; Tankard and Paluck Reference Tankard and Paluck2016). As people in general want to follow the social norms of their group, nudging can potentially be effective, as long as one can identify with messenger and the message is plausible. Being already in favor of the norm in question and aware that others receive the same message increases the perception of shared norms within a community (Tankard and Paluck Reference Tankard and Paluck2016).
However, the literature has demonstrated that such messaging has little effect or even increases the propensity to engage in petty corruption and fosters populist attitudes (Cheeseman and Peiffer Reference Cheeseman and Peiffer2022; Corbacho et al. Reference Corbacho, Gingerich, Oliveros and Ruiz-Vega2016; Köbis et al. Reference Köbis, Troost, Brandt and Soraperra2019). Yet when it comes to voter turnout, Gerber and Rogers (Reference Gerber and Rogers2009) find that people were less likely to turn out when getting information emphasizing low expected turnout, yet only among citizens who already vote infrequently. Yet anti-corruption information has not been applied to voting in response to political corruption in elections, and we thus address this lacuna in our study. In line with Gerber and Rogers (Reference Gerber and Rogers2009), we propose that nudging with positive information about the electoral intentions of others may mobilize non-incumbent supporters, a step that is necessary to overcome the problem of collective action. In particular, we expect the effects of the nudging messaging to be most salient among those voters with more pessimistic a priori expectations about the electoral behavior of other votersFootnote 6 . We therefore test the following premise:
H3: Positive information about other voters’ electoral intention to vote out corrupt politicians will increase the likelihood of opposition voting, particularly among more pessimistic non-incumbent supporters/past nonvoters.
Research design
The case of Romanian mayoral elections
In general, corruption has been an issue at the forefront of Romanian politics since its accession into the European Union (Ristei Reference Ristei2010), with several key leaders, including ex-prime minister Victor Ponta, being indicted for crimes related to corruption. Comparatively, the country’s score is 46 on a 0 to 100 scale on Transparency International’s Corruption Perceptions Index, which puts it in 63rd place out of a total of 198 countries. Among European Union member countries, Romania is ranked 26th out of 27 member states, ahead of only BulgariaFootnote 7 . For example, Mungiu-Pippidi (Reference Mungiu-Pippidi2018, p. 105) describes the country as a case of ‘competitive particularism’, in which public goods are frequently distributed on a non-universalistic basis, favoring loyalists and insiders. The Romanian case is thus particularly suitable for the purpose of corruption voting studies, as the corruption level is relatively high throughout the country in terms of both perceptions and experiences with petty corruption (Charron et al. Reference Charron, Lapuente and Annoni2019), making it a visible and salient issue.
Countries in which there is a direct, executive (mayoral) election and where such executives have substantive powers present voters with greater clarity regarding institutional accountability (Tavits Reference Tavits2007). Romania fits this description as its mayors run in first past the post elections every four years. This is specifically important when it comes to corruption allegations and indictments, as blame assignment is an important part of the electoral accountability chain (De Vries and Solaz Reference De Vries and Solaz2017). Finally, there is a considerable number of mayors who stand accused of corruption by the state authorities on election day. In many cases, this is more than just an allegation. Between 2016 and 2020, incumbent mayors were formally indicted on corruption charges by the well-respected Romanian anti-corruption body, DNA, in 65 municipalities, which lends credibility to the accusationsFootnote 8 . Many voters thus cast their ballot while aware of this situationFootnote 9 .
In addition, there are at least two other conditions that must be met in order to validity test our hypotheses via this case. First, as per H1, there needs to be a sufficiently strong anti-corruption norm among the population, which could plausibly suggest that electoral accountability is possible in this context. Aside from the noted mass anti-corruption protests, international survey data from Eurobarometer (Reference Eurobarometer2020) demonstrate that 64 percent of Romanian citizens claim that ‘corruption personally affects their daily lives’. Comparatively, this sentiment is the highest in the EU, where the average is 26 percent. Moreover, 83 percent claim that corruption is ‘widespread’ in their country. To test the anti-corruption norm among the sample in our survey, we included a direct question about the most important issue in the municipal election to respondents In addition, building on the research design of Boas et al. (Reference Boas, Hidalgo and Melo2019), we embedded a hypothetical voting conjoint experiment into our survey, testing the effects of political corruption relative to other candidate characteristics in explaining (hypothetical) electoral behavior.
Second, as regards H2 and H3, in order to convincingly argue that a collective action problem exists in Romanian elections, one has to demonstrate that voting actually requires some effort from the electorate. Romanian elections require in-person voting on election day (Sunday) (ie no digital voting or mail in ballots, save for those living abroad). Moreover, although a 2019 law ensured that voters waiting in line when the polls for national elections closed would be assured of the right to vote, no such protection yet exists for local elections. This implies that voters arriving late after work face the risk of being denied the opportunity to vote if the lines are too longFootnote 10 . In particular, for the 2020 elections, which took place during the COVID-19 pandemic, extra precautions were taken, whereby strict limits on the number of people (15) allowed to enter the polling stations led to long waiting times in line. These conditions thus create a sufficient time and resource-burden for potential voters to suggest that collective action can play a role.
Sample and data
To investigate the effects of corruption on voting behavior, our study relies on original panel data collected just before and after the municipal elections in Romania, which took place on September 27, 2020. Round one was collected between 1 and 14 days prior to the election, while round two was collected within 12 days following the election. Round one contained 7,070 respondents, of which 2,333 took part again in round two, a re-response participation rate of roughly 33 percentFootnote 11 . All survey administration took place online, where participation was voluntary, and respondents were assured of their anonymity. Since online administration cannot ensure random sampling of the population due to the lack of Internet access for some citizens, standard quota sampling was employed prior to data collection, and post-stratification weights, based on age, gender, education, county of residence (Judet), and partisan affiliation, were constructed after data collection for purposes of generalizing the findings more broadly. More on sampling and weighting can be found in the Appendix, section 1.
As our hypotheses are most concerned with how actual corruption indictments affect voters, we first identified all municipalities where the DNA had indicted a sitting mayor. We focused only on the mayor, which we view as the most high-profile office, rather than council members or other local political actors, in order to increase the likelihood that citizens have heard of the indictment. There were, in total, 65 municipalities with an indicted mayor, ranging from small, rural towns with populations under 4000 to several large urban cities. Moreover, in 59 of these municipalities (90%), the indicted mayor chooses to run for re-election. These municipalities represented roughly 2.1 million people (or 12 per cent of the total population), and thus we elected to over-sample in such municipalities so as to increase our sample for the purpose of significance and validity testing. In collaboration with our survey partner company, we were then able to obtain 3,070 and 1,165 respondents from municipalities with indicted mayors in rounds one and two, respectively.
Observational data
To test our hypotheses, it is necessary that our outcome variable includes three electoral choices, and not simply a binary choice for incumbent voting that collapses abstention and opposition voting into one reference category. Building on several previous studies (Agerberg Reference Agerberg2020; Bauhr and Charron Reference Bauhr and Charron2018), we therefore included three categories – incumbent voting, opposition voting, and abstention – and inquired about the 2020 as well as the 2016 vote. For the opposition voting category, we coded this as ‘1’ if the respondent chose any candidate that was not the incumbent. This could, of course, be a limitation if the opposition is also corrupt, leaving the voters with few or no ‘clean’ alternatives. We found that the DNA’s indictments most often apply to incumbents, however. Thus, according to our indicator, the opposition is, at least objectively, cleaner, although a caveat is that this might not match perceptions. A key advantage of our outcome measure is that we include abstention as a choice. This is seldom offered in hypothetical experiments, thus considerably altering the calculus of a respondent, and as a result, overestimating the effect of corruption on incumbent voting. This option is particularly salient in a setting in which voter turnout is relatively low, as in Romania or other Central and Eastern European countriesFootnote 12 .
We have several main explanatory variables of interest. To test H1, we asked about knowledge of political corruption by a sitting mayor. Previous field studies have relied on several different indicators of corruption, including information on red flags from government audit reports (Boas et al. Reference Boas, Hidalgo and Melo2019; Ferraz and Finan Reference Ferraz and Finan2008), or information on irregularities in public expenditures (Arias et al. Reference Arias, Larreguy, Marshall and Querubin2022; Chong et al. Reference Chong, De La O, Karlan and Wantchekon2015). Instead of budgetary information, we rely on indictments from the DNA, which serve as an objective indicator, thus removing any subjective decision-making on our part in parsing legal definitions of what constitutes corruption in this context. As a result of this clarity, our indicator is a potential improvement compared to previous field studies, in which the proxy for corruption might have been too weak to yield results (Incerti Reference Incerti2020). A potential weakness with our indicator is that while the DNA has been a relatively trusted institution in Romania, it has become more politicized in recent years, and thus voters could simply dismiss an indictment as political rather than factual, even if they are aware of it (Mendelski Reference Mendelski2021). Yet, among six questions in our survey on trust in national and local politicians, other people, the media, courts, and the DNA, the latter ranks as the most trusted (see Appendix, Figure A3). In addition, our study also departs from other related field experiments on the effect of corruption in real-world elections (Banerjee et al. Reference Banerjee, Kumar, Pande and Su2011; Boas et al. Reference Boas, Hidalgo and Melo2019; Chong et al. Reference Chong, De La O, Karlan and Wantchekon2015) in that we do not randomize information on corruption; the variable is observational. Essentially, we did this for two reasons. First, even as other field studies have relied on randomization of publicly available corruption information (e.g. audits, etc), there is still apparently a trade-off between the uncertainty of compliance with the treatment via experimental randomization and the certainty of self-reported knowledge via observation, as per weak treatment issues (Incerti Reference Incerti2020). Second, while we elected a more causal design in testing H3, we are less interested in the causal effects of corruption per se in testing H1. Our main aim is to investigate any association with incumbent voting, given corruption awareness. We therefore elected to rely on the a priori, self-reported information of voters, which also provides us with more certainty regarding the internalization of information on corruption. For rounds one and two of the survey, 26.7% and 28.7%, respectively, of respondents claimed to have heard about a DNA indictment of their sitting mayor (see Appendix section 4 for the specific question wording). To check the validity of the measures, we performed a simple t-test, comparing the proportions of ‘Corruption knowledge’ of an indictment by two groups of municipalities – those living in municipalities with an indicted mayor and those living elsewhere. As anticipated, we find a significant difference, whereby roughly 19 percent and 36 percent claim to have heard such information in municipalities with no indicted mayor and with an indicted mayor, respectively. In the former group, this could reflect either noise or that there was corruption relating to other elected candidates. Yet the clear difference (p < 0.0001) in the two sets of localities is nonetheless reassuring in terms of validity (see Appendix Table A2 for test results). In addition, we check the correlates of such corruption knowledge, and found no systematic relationship with demographic characteristics or levels of trust in institutions (Appendix Figure A2).
Next, we are interested in past voting behavior for H2 and H3, and how knowledge of corruption affects various groups of voters in different ways. In round one, we inquired for whom the respondent voted in the 2016 municipal election. We then coded respondents into three groups – (1) past supporters of the sitting mayor, (2) past supporters of opposition candidates to the sitting mayor, and (3) past nonvoters. In addition to the main variables, the survey included standard demographic information (gender, age, education, etc.) along with partisan affiliation and a battery of trust in institutions questions, inter alia. The full survey is in Appendix, section 3.
Experimental data
In complement with the observational data used to assess H1, we build on the work of Boas et al. (Reference Boas, Hidalgo and Melo2019), whereby we embedded a survey experiment in the second round of our survey. This is intended to highlight the normative reaction to political corruption and provide us with a baseline expected effect of corruption on actual voting behavior, to which we can compare our real-world results (see Appendix, section 2 for more details). As regards H3, we test the effects of the collective action hypotheses by embedding a randomized survey experiment into round one of our survey. The idea behind our collective action hypothesis is that indicted mayors are re-elected despite corruption because most voters believe that others will not turn out and vote against them. If our idea is correct, then information contrary to this belief may affect voter behavior, particularly among those who believe that others will not act collectively against a corrupt mayor in an election. Table 1 summarizes the set-up and data collection of the design.
Summary of data collection and design
To set up the experiment, we first asked respondents the following question towards the end of the first round of the survey: If a politician is proven to be corrupt while in office, how likely is it, do you believe, that other citizens in your area will stand up and vote him/her out of office?
Respondents could choose from ‘very likely’ (1) to ‘very unlikely’ (4). Roughly 61 percent of our sample responded ‘unlikely’ or ‘very unlikely’, implying that a sizeable majority perceived a problem of electoral collective-action among their fellow electorate. We highlight the observational correlates of the ‘pessimistic respondent’ in Table A8 in the appendix. We then provided 50 percent of our sample, randomly assigned, with a pie chart visual (see Appendix, Figure A1) and the following information taken from a recently published survey in Romania: ‘In fact, a recent poll showed that 82% of Romanians claimed that they would vote against any politicians charged with corruption in the upcoming election’. The empirical implication from our theory posits that we should observe, from among the treated respondents, higher rates of action against the incumbent, i.e., more opposition votes in general or decreased levels of abstention among non-incumbent loyalistsFootnote 13 . We performed a balance check of treatment assignment with a list of potential confounding factors and found that the treatment was unrelated to other variables (see Appendix Figure A1).
Results
Norms versus action? The effect of corruption on hypothetical versus actual voting
H1 asserts that despite a strong anti-corruption norm in principle, in practice, political corruption will have little affect on real-world voting behavior. We evaluate H1 in two parts − e.g. ‘norms’ and ‘action’. With respect to ‘norms’, we begin by establishing the degree to which an anti-corruption norm is present among Romanian voters using both observational and experimental data. The aim is to establish a baseline expectation that knowledge of political corruption should, in fact, influence voters. First, via observational data, a direct question in our survey regarding which issue most motivated voters in the 2020 local election shows that corruption is the most salient issue among a plurality of voters (Figure A6 in the Appendix). Second, evidence from our conjoint experiment (from survey round 2) demonstrates a clear anti-corruption norm, whereby Romanians would strongly sanction incumbents tied to political corruption in a hypothetical setting. Table 2 summarizes two key results for H1 from the experimental data (see Appendix, section 2 for full results), where we find that effect of describing candidate A has the largest effect of all variables in terms of magnitude. The effect of corruption reduces the probability of choosing ‘candidate A’ by 28.9 percent, while it increases the choice of abstention by 17 per cent. Moreover, following Agerberg (Reference Agerberg2020) in conditioning results on candidate ‘B’ being ‘clean’, the model predicts a similar negative effect for choosing candidate ‘A’, while the conditional effect for abstention drops to just 0.045%, implying that a clean alternative in the race leads to higher participation.
Treatment effects of corruption on vote choice from conjoint experiment
Note: The table shows the predicted probability of voting ‘candidate A’ in the conjoint experiment based on whether candidate A is presented as corrupt or clean (unconditional), and conditional on candidate ‘B’ being clean. The difference is thus the average treatment effect given ‘B’ is non-corrupt. Number of observations is 2,333. Full results used to calculate these predicted probabilities are found in Appendix, Table A3. **p < 0.01,*p < 0.05.
Having established a strong anti-corruption norm, we then assess ‘action’ via the effect of knowledge about the mayor’s indictment on incumbent voting. Table 3 also adds controls for key theoretical expectations from the literature, as well as a number of sample specifications. Model 1 in Table 3 includes all respondents, while model 2 is restricted to only those living in a municipality with an indicted mayor (CM) from 2016 to 2020. Model 3 adds further to the restriction of only indicted mayors who chose to run for re-election. Models 4 and 5 split the samples between past ‘non-loyalists’ (2016 abstainers or opposition respondents) and past loyal voters, respectively. Our models control for gender, age, education, economic satisfaction, political interest, past voting, issue salience of corruption, residence in urban municipalities, and whether the mayor was indicted, along with the number of political parties in some models to account for partisan alternatives. Additionally, we further tested whether the presence of a candidate from the anti-corruption party, USR, in a municipality moderates the effect of corruption knowledge on incumbent support. We found no evidence that it does (see Appendix, Table A7).
Predicting incumbent voting in the 2020 election: logit estimates
Note: Dependent variable is the choice of voting for incumbent (binary) in 2020. Logged odds coefficients reported with standard errors in parentheses, clustered by municipality. ‘Corruption knowledge’ is a binary variable indicating that the respondent has heard that his/her sitting mayor was indicted by the DNA. ‘CM’ indicates only respondents from municipalities where a sitting mayor was indicted. Models include controls for gender, age, education, political interest, past voting, and whether the respondent resides in a municipality with an indicted mayor. **p < 0.01, *p < 0.05.
Our main variable of interest here, and a key factor in the literature, is, of course, that voters need to have information about the scandal itself in order to react (De Vries and Solaz Reference De Vries and Solaz2017; Weitz-Shapiro and Winters Reference Weitz-Shapiro and Winters2017; Winters and Weitz-Shapiro Reference Winters and Weitz-Shapiro2013). In our case, we assessed this with a dummy variable (‘Corruption knowledge’). We found that in none of the models does this variable significantly explain incumbent voting, given an alpha of 0.05, which lends support to H1, which predicts a discrepancy between norms and actual voting. This is particularly interesting because the corruption information is self-reported, which would, assumedly, trigger some social desirability biases, leading to underreported voting for an incumbent who the respondent claims to know has been indicted. Yet this is not the case, not even for past opposition supporters. We thus observe that information about the indictment may be a necessary condition for accountability, but it is not a sufficient condition. Thus, when comparing the effects of corruption on voting behavior in the actual municipal election to our conjoint experiment results, we found, similar to Boas et al. (Reference Boas, Hidalgo and Melo2019), that the effects of corruption knowledge on actual voting choices are negligibleFootnote 14 . This is consistent with most previous ‘real-world’ election studies (see Incerti Reference Incerti2020) and thus corroborates H1. In addition, we found that economic (dis)satisfaction, higher education, and more partisan choices reduced the likelihood of incumbent voting. The full model is found in the Appendix, Table A6.
Corruption and abstention: electoral accountability as a collective action problem
Thus far, our findings are largely in line with the literature, which says that anti-corruption norms do not translate into action at the polls (Boas et al. Reference Boas, Hidalgo and Melo2019; Incerti Reference Incerti2020). We now move to explain this discrepancy in our low-turnout context, based on the framework of collective action. Testing the degree to which problems of collective action disrupt democratic accountability at the micro-level is far from straight-forward, and mainly without precedent within this literature. We elected to evaluate our hypothesis via an empirical implication of the theory, and we employed both observational and experimental data, as explained in a previous section, to investigate this link. First, we compared the three predicted electoral outcomes via two, split-sample multinomial logit regression models – e.g. municipalities with and without an indicted mayor for the 2020 elections – and a set of controlsFootnote 15 . Our expectations from the collective action framework are that any reduction in incumbent support due to corruption happens due to higher rates of abstention rather than opposition voting.
Table 4 shows the predicted probabilities of each of the three outcomes according to the legal status of the municipal mayor. Two observations from these results corroborate our expectations. First, there is a noticeable trade-off between incumbent support and abstention between the two types of municipalities. While incumbent support is nearly 6 percentage points higher in municipalities without corruption, abstention is estimated at over 5 percentage points lower (significant at 90% level of confidence) compared with municipalities without an indicted mayor, implying a roughly 11 percentage point swing in these two outcomes. Second, the predicted rates of opposition support are nearly identical in the two groups. These two results imply that a loss of support for the incumbent, due chiefly to corruption, is a function of greater abstention in the aggregate.
Comparing electoral outcomes in municipalities with and without an indicted mayor
Note: Dependent variable is vote choice, with predicted probabilities from multinomial logit estimation with 95% confidence intervals in brackets. Models include controls for past voting, having heard of corruption, age, gender, education, and political interest. **p < 0.01, *p < 0.05.
Second, focusing on the micro-level, we test whether the more pessimistic group (i.e., the 61 percent who believe ‘others won’t act’ against political corruption in elections) is associated with higher voter abstention compared to those more optimistic voters in Table 5. This, ultimately, serves to hinder the accountability of corrupt politicians (Birch Reference Birch2010; Sundström and Stockemer Reference Sundström and Stockemer2015). In particular, for H2, we are interested in these effects for non-incumbent supporters, and to obtain a proxy for this, we run two additional models, whereby we split the 2020 non-incumbent (model 2) and incumbent supporters (model 3). To test this more directly, we regressed abstention in 2016 on the belief about others’ actions against corruption, along with a set of controls. Table 5 reveals that when controlling for gender, age, education, occupation, political interest, and municipality-type, the pessimistic belief that ‘others won’t act’ still explains a significant part of the variation in abstention in 2016. In addition, we also controlled for social trust so as to isolate the electoral aspect of collective action vis-a-vis lack of trust in fellow citizens more generally. The model predicts that voters believing that ‘others won’t act’ are roughly 4 per cent more likely to have abstained, all things being equalFootnote 16 . Moreover, we see that this effect is most pronounced among non-incumbent supporters, while these perceptions do not affect incumbent supporters’ past voting. The effect is demonstrated for both self-reported abstention in 2016 (models 1–3) and 2020 (model 4). In addition, in model 5, we also find that the ‘others won’t act’ sentiment significantly correlates with the choice of abstention in the hypothetical conjoint experiment from Table 2, all things being equal.
Attitudes of others’ behavior and abstention
Note: The dependent variable is a binary variable which equals ‘1’ for abstention and ‘0’ for having voted in local elections (self-reported). Abstention in the 2016 municipal elections used in models 1–3, and in 2020 in model 4. Dependent variable in Model 5 is the choice of ‘neither candidate A or B’ from the conjoint experiment (Table 2 ), and includes randomized candidate characteristics as controls (not shown). Coefficients are marginal effects from a linear probability model. ’IV’ stands for ‘incumbent voter’. Model 4 includes only control group respondents from the ‘nudging’ information experiment. Survey weights for education, age, and gender, and robust standard errors in parentheses. Full results found in Table A4 in the Appendix. **p < 0.01, *p < 0.05.
In sum, the results from Tables 4 and 5 demonstrate two important points. One, that the observed electoral effect of political corruption is more likely to be abstention rather than opposition voting, and two, that beliefs about others’ behavior is significantly linked to voter abstention – particularly among past non-incumbent supporters. Thus, the data demonstrate empirical support for H2.
Can collective action problems be mitigated with positive information?
The results thus far help explain that the lack of electoral accountability found in H1 can be partly explained by collective action issues raised in the test of H2, suggesting a vicious cycle. We now move to whether information via our survey experiment can mitigate such beliefs in the 2020 election. H3 posits a conditional hypothesis, and we therefore created a simple interaction between treatment assignment (binary) and a priori beliefs about others (‘strongly agree’ + ‘agree’ = ‘optimist’, whereas ‘strongly disagree’ and ‘disagree’ = ‘pessimist’). H3 anticipates a stronger treatment effect among pessimistic respondents compared with those more optimistic. In this case, we present the conditional average treatment effect ((CATE) Abrevaya et al. Reference Abrevaya, Hsu and Lieli2015)Footnote 17 .
Figure 1 highlights the treatment for ‘optimists’ and ‘pessimists’ for the three electoral outcomes. As we report the conditional effects of the treatment in two sub-groups, we followed the recommended practice of reporting marginal means (Leeper et al. Reference Leeper, Hobolt and Tilley2020). Several interesting observations can be seen in the results. First, we observed no effect of the treatment for either group for incumbent support. Second, the treatment has no effect on ‘optimistic’ people who already believe that others will act in the face of political corruption (left-hand side). Thus, similar to the findings of Gerber and Rogers (Reference Gerber and Rogers2009), we find that information about others’ intents only affects those who were initially more pessimistic about others’ voting behavior
Conditional treatment effects of information on others’ attitudes toward political corruption. Note: Treatment effects show by group – those who believe others will act (‘optimists’) and those who do not (‘pessimists’). Estimates from multinomial logit model. Control variables include gender, age, education, political interest, unemployment, awareness of corruption charges, political trust, economic satisfaction, urban municipality, and whether one’s municipality had a mayor charged with corruption. Survey weights used, and standard errors are clustered by municipality, with 95% confidence reported. Sample size is 2333. Full results are found in model 2, Table A5 in the Appendix.
Next, the model reveals two rather interesting results with respect to opposition-voting and abstention, in particular among those respondents who were pessimistic about others’ reactions to corruption (right-hand side). Table 6 provides more specific numbers on the abstention-opposition support trade-off among the two groups. Concerning the former outcome, among those who are pessimistic regarding the behavior of others in voting out corruption, the treatment group shows significantly less abstention in the 2020 elections compared with the respondents who received no information; whereby, the probability of non-voting reduces by roughly 11.1 percentage points for those who received the nudging information. Yet, among those with an a priori belief that ‘others will act’ (‘optimists’), the treatment yields a positive, yet statistically insignificant, coefficient. Further, a comparison of the CATE among these groups shows that the treatment effects for abstention are statistically different. When it comes to opposition voting, the nudging information results in a 6.7 percentage point increase among more a priori pessimistic voters, while it again has no effect among the more optimistic group. Together, these findings imply a roughly 18 percentage point swing from abstention to opposition voting among the more pessimistic voters when presented with encouraging information about the intention of their fellow citizensFootnote 18 . Table 6 provides more specific numbers on the abstention-opposition support trade-off among the two groups.
Comparison of treatment effects in both groups for non−incumbent support outcomes
Note: The table shows the conditional average treatment effects for the outcomes ‘abstention’ and ‘opposition support’ by a priori belief that ‘others will act’ and is taken from the model used to produce Figure 1 . Number of observations is 2,333. Full results found in Appendix, Table A5. **p < 0.01,*p < 0.05.
In sum, receiving information about the stated attitudes of other voters’ intentions can aid in countering resignation and motivate opposition voting among more pessimistic voters, who are, on average, less likely to vote in general. We interpret this finding as evidence for corroborating H3Footnote 19 . The implication of these results is that incumbent politicians charged with corruption benefit from the collective action of pessimistic sentiment among voters, as long as such voters are not informed about others’ intention to vote.
Discussion
This paper sets out to better understand the dynamics of retrospective voting behavior when political corruption is a factor. Specifically, our main research question investigates how and why corrupt incumbents manage to stay in power via popular elections in which voters voice their strong dislike of corruption. The literature is mainly framed along the lines of the ‘corruption-voting paradox’, i.e., why citizens continually vote for incumbents who have engaged in corruption. Using unique survey data just prior to and after the 2020 municipal elections in Romania, we find that the answer to the question is that only a relatively small minority of the total voting population does, in fact, support corrupt incumbents. These individuals are mainly those who supported the incumbent in previous elections and who perceive that they are gaining from the current situation (e.g. having positive opinions of the state of the economy, inter alia). Moreover, we find that among the group of past supporters from 2016, even self-reported knowledge of a mayor being indicted for corruption plays no role in explaining their vote in 2020.
Yet, we find that the overwhelming majority of potential voters do not in fact vote for corrupt incumbents, they either vote for an opposition party or simply abstain altogether. Further, we find a strong anti-corruption norm among voters in our conjoint experiment, as well as the opinion that corruption is the most salient issue for voters prior to the 2020 election. Thus, there is an enormous potential for electoral accountability to work effectively, i.e., as a channel through which citizens ‘vote the rascals out’. However, in practice, this is rarely the case in Romania or elsewhere (Bågenholm and Charron Reference Bågenholm and Charron2020; Incerti Reference Incerti2020). Our study provides a plausible explanation as to why, in fact, this is not the case, even though the majority do not support corrupt incumbents.
Given the stability of past incumbent-supporting voters, in order to successfully unseat a corrupt incumbent, a sufficient number of past abstainers must, in fact, vote for an opposition party candidate, while past opposition voter support must hold. Even among such non-incumbent-supporting voters, having heard of corruption indictments played no role in their vote choice, implying that the information hypothesis, which posits that voters act only when informed of political corruption (Winters and Weitz-Shapiro Reference Winters and Weitz-Shapiro2013), fails to explain corruption-voting in this context. Thus, some of the more common explanations from the literature do not hold in our case, which is one of low voter turnout, strong incumbency advantage, large winning margins, and recurrent re-elections of corrupt incumbents. We thus argue that previous research has under-played the majority of the electorate in such contexts, namely the nonvoters, and it is they who hold the key to effective electoral accountability. The question is why, despite their detestation of corruption, they decide to abstain on election day. In order to explain this, we develop a novel theoretical argument – that there is a collective action problem among non-incumbent supporters. In general terms, we posit that people will act in accordance with how they think most other people will act. In this specific case, it means that non-incumbent supporters will have little reason to turn out and vote if they believe most other voters will not punish corrupt incumbents.
The results support our expectation: Respondents who believe most others will not act against corruption are less likely to turn out to vote. This is pessimistic news for those fighting for a less corrupt society. Our findings suggest that there is a vicious circle, whereby ‘corruption voters’ are mobilized by the incentive structures in the specific context described above, while their opponents remain inert. The question is, then, how to break this vicious circle and activate non-incumbent supporters. Fortunately, our findings also provide, to some degree, optimistic news in this respect.
Perhaps our most novel empirical contribution to the literature from this study is our unique research design that included an original two-round survey data and a randomized experiment that gave respondents information about the willingness of others to engage in the fight against corruption. We show that when those respondents who had a negative view of others acting against corruption were provided with information saying that a vast majority of voters would throw the rascals out, they abstained significantly less and chose opposition candidates at higher rates. Thus, we in fact add additional nuance to the information hypothesis; it is not sufficient for voters who are most pessimistic about collective action against corruption to simply be informed about political corruption, they also need information on the intent of other electors’ reactions. Such findings show support for our theoretical mechanism and have clear implications for anti-corruption policy, which could be actionable in subsequent elections.
Our study relies on within-country variation, and thus generalizability becomes a question. Yet Romania is far from a global outlier in terms of level of corruption, turnout, and experience with democracy, and thus we speculate that the case we present is most applicable in countries/elections with relatively low turnout, where partisan loyalty is relatively weak (Tatar Reference Tatar2013). Moreover, these features in the Romanian context, in particular the low turnout and large margins of victory for incumbents (Bågenholm and Charron Reference Bågenholm and Charron2020), make a positive turn-around quite unlikely in the short-term perspective. And while our findings offer some degree of hope, it is worth remembering that collective action problems are, by nature, difficult to solve, as the approach primarily explains why the status quo, which in this case is corruption, prevails. Our experimental results do show, however, that some positive ‘nudging’ information about the intent of others can have a positive effect on accountability. Our contribution with this study, by focusing on the disincentives for nonincumbent supporters to vote, is thus to increase understanding as to why not only corruption but also corruption voting is such a sticky phenomenon in this context and in others similar to Romania.
Speaking to a broader literature, our findings also corroborate previous work that suggests that transparency and providing information about specific incidences of political corruption can have adverse effects for electoral accountability, in particular in contexts with relatively moderate to high corruption (Bauhr and Grimes Reference Bauhr and Grimes2014; Carreras and Vera Reference Carreras and Vera2018). It is precisely among the voters one needs for accountability (mainly past nonvoters and opposition voters) that knowing about a political corruption scandal or indictment is insufficient. However, our findings show that providing more fine-grained information about the intentions of others to act against corruption, in particular to those who are, a priori, most likely to believe that others will simply abstain in the face of corruption, leads to less abstention and more opposition voting. However, whether the source of such information should be ‘top down’ from media or international organizations or more ‘bottom up’ through social networks (Arias et al. Reference Arias, Balan, Larreguy, Marshall and Querubin2019) is a question for future research. Yet given that electoral competition may be one of the last and most viable mechanisms for rooting out systematic political corruption in environments such as Romania (Klašnja and Pop-Eleches Reference Klašnja and Pop-Eleches2022), we view our findings as highly relevant for researchers and policy-makers alike.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1475676525100352
Data available from the authors, please contact Nicholas.charron@pol.gu.se for the data used in this study.
Acknowledgements
The authors would like to thank Marko Klašnja, Jana Schwenk, Dominik Schraff, and the participants of the research group Political Sociology at Aalborg University, Vlad Perju and the participants of the weekly seminars at Harvard University, Minda de Gunzburg Center for European Studies, and colleagues at the QoG Institute at the University of Gothenburg.
Funding
Charron acknowledges partial funding by RJ sabbatical (SAB20-0064), Wenner-Grenstiftelserna (SSv2020-0004), Vetenskapsrådet (no. 2019-02636), and the European Commission, DG Regio [no. 2019CE160AT022].
Open access
