Too Crooked to be Good? Trade-offs in the Electoral Punishment of Malfeasance and Corruption

Abstract

While elections are an instrument to hold politicians accountable, corrupt politicians are often re-elected. A potential explanation for this paradox is that citizens trade-off integrity for competence. Voters may forgive corruption if corrupt politicians manage to deliver desirable outcomes. While previous studies have examined whether politicians’ competence moderates the negative effect of corruption, this paper focuses on voters’ priorities and directly assesses what citizens value more: integrity or favourable outcomes. Using a survey experiment, we assess citizens’ support for politicians who violate the law in order to improve the welfare of their community and, in some cases, benefit personally from these violations. The results indicate that citizens prefer a politician who follows the law, even if this leads to a suboptimal outcome. However, voters are more likely to overlook violations of the law that benefit the community if these do not result in a personal gain for politicians (i.e., in the absence of corruption). These findings suggest that the mild electoral punishment of corruption may be due to the public’s unawareness of private gains from malfeasance, or to the delay in these private benefits becoming apparent by election day.

Introduction

Elections should serve as an instrument to hold politicians accountable (Fearon, 1999). However, while some studies indicate that corruption is punished (Ferraz and Finan, 2007; Wood and Grose, 2022), many others indicate that voters only mildly punish malfeasant politicians (Peters and Welch, 1980; Dimock and Jacobson, 1995; Chang, Golden, and Hill, 2010). One could think that the reason behind this mild electoral punishment is that citizens fail to recognise malfeasant actions, or they do not disapprove of them. However, most citizens are capable of identifying malfeasant actions: they express an outright rejection of corruption, and they claim they would not vote for crooked politicians.

According to a logic of retrospective voting, informed citizens would, therefore, not support malfeasant politicians. There are different factors that deter the electoral punishment of crooked politicians, though (for recent reviews see De Vries and Solaz, 2017; Ares et al., 2019). One strand of the corruption literature has focused on assessing whether citizens condone corruption when politicians manage to deliver desirable public goods and outcomes such as economic growth. In line with this hypothesis, studies indicate that a good management record can overshadow the impact of corruption. For instance, corruption is less severely punished at the ballot box during favourable economic times (Klašnja and Tucker, 2013; Zechmeister and Zizumbo-Colunga, 2013). This leniency may also manifest itself when a politician has a commendable management record (Esaiasson and Muñoz, 2014; Muñoz, Anduiza, and Gallego, 2016), or has implemented sound economic policies (Konstantinidis and Xezonakis, 2013). Winters and Weitz-Shapiro (2013) explicitly propose that voters may face a trade-off at the ballot box when they believe that the benefits provided by a corrupt politician outweigh the costs of corruption. However, their findings indicate that voters generally do not trade-off corruption for a good management record. If politicians are corrupt, they are punished at the ballot box. However, Zhu and Yang (2023) find that Chinese citizens make this trade-off (but see Vera, 2020). Thus, the literature does not provide a univocal answer about the trade-off between competence and integrity.

We propose that some of these conflicting findings may be related to the fact that most studies analyse this trade-off indirectly and do not account for the mechanisms that may deter or drive the punishment of corruption in these scenarios. Most of this prior literature focuses on politicians who, despite being corrupt, have a good management record. This we would argue is unlikely to reveal citizens’ true preferences about the trade-off between competence and integrity. In our study, we contribute to this literature by analysing citizens’ likelihood to vote for politicians who manage to deliver desirable outcomes by being corrupt, while also investigating the mechanisms that may drive or deter the punishment of corruption in these cases. As such, we evaluate whether individuals put more weight on politicians’ ability to get the job done than on their integrity.

We test this argument through a survey experiment fielded in Spain, in which a politician is faced with the dilemma of either adhering to the law or solving an emergency more quickly (i.e., a direct trade-off between competence and integrity). Through this experimental design, we examine: (i) whether citizens infer different character traits depending on how politicians react to this dilemma; (ii) how these traits affect voting decisions; and (iii) whether the impact of procedural violations increases in cases of outright corruption. To do so, we distinguish between two degrees of malfeasance: breaking the law with and without obtaining a private benefit. Since only the latter can be considered corruption, distinguishing between these two degrees of malfeasance allows us to study citizens’ reactions to actions that, while entailing a procedural breach, may not be considered corruption. Previous studies suggest that citizens react differently to different forms of corruption (Weschle, 2016; Bauhr, 2017; Botero et al., 2021). Our research tests whether citizens make a distinction between unlawful decisions and instances of corruption, even when both contribute to welfare enhancement and yield the same positive societal outcomes.

The results reveal that voters punish malfeasant politicians even when their actions increase the welfare of their communities. Respondents prefer a law-abiding politician, even if this leads to a suboptimal outcome. The analysis of the mechanisms reveals that the electoral punishment of malfeasant politicians is driven by the loss of trust generated by the unlawful decisions taken to achieve an optimal outcome, which is not offset by the gains in the perceived efficiency of that politician. This is especially apparent when the illegal decision involves a direct private gain for the politician (i.e., in cases of outright corruption). Even if they lead to superior outcomes, procedural violations (not following the law) have negative electoral consequences, and this is exacerbated in cases of outright corruption.

These findings have implications for the study of corruption accountability. By distinguishing between different degrees of malfeasance, our results indicate that citizens may tolerate procedural (legality) violations that involve a side benefit only when politicians do not obtain a private gain (i.e., when there is no corruption involved). Our design, combining a direct trade-off and multiple degrees of malfeasance allows us to complement existing studies on this topic (e.g., Vera, 2020; Botero et al., 2021). We would argue that the fact that citizens are more willing to tolerate unlawful decisions when they do not bring private gains for politicians can help us explain the moderate punishment of corruption. Our analyses of the mechanisms driving this relationship suggest that the loss of trust associated with procedural violations, especially corruption, tends to offset the perceived gains in efficiency, and this explains the electoral punishment of politicians who break the rules. This calls for greater attention to the study of mechanisms based on politicians’ trait evaluations when analysing why corruption is not always being punished.

Theoretical framework

Outcome accountability and procedural accountability

In assessing the circumstances under which citizens support dishonest candidates, it is important to bear in mind that while voting decisions may be driven by integrity considerations, other factors may also play a role in deciding who to vote for. Corruption is only one out of many factors to evaluate a candidate’s performance (Breitenstein, 2019). Voters may implicitly trade-off politicians’ integrity for other more valued qualities.

In line with this reasoning, recent studies have examined whether citizens forego malfeasance when candidates have a good management record and deliver desirable outcomes such as economic growth. For example, evidence from 19 presidential systems indicates that corruption does not reduce support for presidential candidates in good economic contexts, while it can reduce their vote share by up to 10 per cent in bad economic contexts (Zechmeister and Zizumbo-Colunga, 2013). These findings are corroborated by experiments showing that citizens condone corruption in good economic contexts (Klašnja and Tucker, 2013), when the politician has a good management record (Muñoz, Anduiza, and Gallego, 2016), or when they implement favourable economic policies (Konstantinidis and Xezonakis, 2013). In fact, experimental evidence from Sweden and Spain, indicates that citizens would rather vote for a corrupt but competent politician than for an honest but incompetent one (Esaiasson and Muñoz, 2014). Taken together, these findings suggest that voters base their voting decisions on the outcomes produced by politicians’ decisions rather than on the procedural integrity of their actions while in office.

These findings contrast sharply with the views expressed by citizens in public opinion polls. These reveal a clear preference for honest politicians, even if this means that these politicians are less efficient at delivering public goods (see e.g., Allen and Birch, 2011). When asked to choose between the two characteristics, European citizens consistently prioritise politicians who are honest over those who get the job done (Allen, Birch, and Sarmiento-Mirwaldt, 2016).

To address this puzzle, we examine why voters might support crooked politicians who manage to get the job done. In contrast to most previous studies, though, we assess citizens’ likelihood of supporting politicians who manage to deliver superior outcomes precisely by circumventing established legal procedures. As we argue in the introduction, some of the conflicting findings in the literature may be related to the fact that most studies analyse this trade-off indirectly, by focusing on candidates who are corrupt but, at the same time, competent. This is why in this paper we focus on citizens’ reactions to politicians that are competent because they are corrupt (or thanks to being corrupt), rather than reactions to politicians that are competent yet corrupt. This direct confrontation with a trade-off will make the tensions between the two desirable outcomes more apparent and will be more conducive to revealing the underlying preferences of citizens (see Ares et al., 2024).

Assessing citizens’ reactions to situations that pose an immediate dilemma between outcome and procedural accountability seems pertinent, given that while the side benefits of political misconduct (e.g., faster decision-making) are realised immediately, its negative societal consequences may take more time to materialise. To begin with, burdensome regulations may lead citizens to perceive that circumventing them could be a more effective and efficient approach to policymaking. Second, most of the negative consequences of malfeasance take several years to materialise. For example, the negative consequences of building a school with materials of inferior quality to those budgeted for are not immediately apparent, even though the long-term consequences could be catastrophic (Fisman and Golden, 2017). However, the positive impact of a new school in a community is unequivocal and directly perceptible in the short run. This highlights the need to assess this trade-off directly if one aims to analyse citizens’ reaction to it. When casting a vote, the information about specific wrongdoings, if available, does not usually include details about the negative societal consequences of these actions. Hence, it might be unfeasible to directly assess the negative impact of a specific act of misconduct when deciding who to vote for. However, this does not necessarily imply that citizens condone malfeasance and corruption in light of positive outcomes. Examining whether they do, and the reasons why this may be the case, is better achieved by focusing on situations where there is a direct trade-off between competence and integrity.

Existing studies have assessed citizens’ reactions to politicians that are competent because they are corrupt. For example, Fernández-Vázquez et al. (2013) show that mayors’ misdeeds that result in positive outcomes for the community are not met with negative electoral consequences. Similarly, mentioning that corruption has brought construction jobs to the community reduces the electoral consequences of corruption (Klašnja, Lupu, and Tucker, 2021; Vera, 2020). However, these studies, like most of the previous literature, assess the mitigating effects of competence, or these side benefits of corruption. This is what Breitenstein (2019) coined as the conditional punishment argument. This could be considered a softer test of the competence-corruption trade-off hypothesis (Zhu and Yang, 2023). Instead, in our study we test the relative weight argument and assess whether citizens prioritise outcomes over procedural integrity. As such, we evaluate whether individuals put more weight on politicians’ ability to get the job done than on their integrity in following the established legal process. Specifically, we focus on scenarios in which politicians can either follow the established legal procedure and achieve a suboptimal outcome, or bypass that procedure to achieve a superior or optimal outcome. In such scenarios, competence is achieved by defying the established legal procedures, which means that a politician can only be either competent or honest.

This trade-off framework, which confronts politicians and citizens with a direct tension between two desirable outcomes that must be prioritised, has several advantages. First, the trade-off scenario is more appropriate for assessing respondents’ priorities regarding integrity and competence, as only one quality (integrity) can be obtained at the expense of the other (competence) (Ares et al., 2024). Second, it reduces social desirability bias, as neither of the two scenarios can be identified, a priori, as more acceptable than the other. Third, it describes the dilemmas that politicians sometimes face, thereby increasing the external validity of our experimental design. For example, during the COVID-19 pandemic, many governments needed to urgently procure medical equipment. This led some governments to relax their standards on procurement rules (World Justice Project, 2020). Moreover, when caught red-handed, politicians sometimes justify their wrongdoings by arguing that their actions led to an optimal societal outcome. For example, in recent cases in Spain, politicians have justified the direct and discretionary award of a public contract to a company without holding a legally required call for tenders because it was faster and cheaper (e.g., Rejón, 2018; Cornejo, 2019).

Our initial expectation is that politicians’ decisions in situations involving a trade-off between competence and integrity will affect voting choices. However, the magnitude and direction of these effects depend on how these decisions shape citizens’ perceptions of politicians’ character and attributes. We elaborate on these arguments and our hypotheses in the following subsection.

The mechanisms: malfeasance, politicians’ traits, and voting decisions

Our discussion implies that voters intrinsically value the procedural integrity of politicians, as well as the outcomes they deliver, and that this shapes their voting decisions. This argument follows closely the retrospective voting literature. This strand has primarily focused on the importance of economic performance for voting (see Lewis-Beck and Stegmaier, 2007). The literature on corruption voting has closely followed this perspective (see e.g. Ecker, Glinitzer, and Meyer, 2016). This is reflected in the common use of expressions such as “punishing malfeasant politicians” or “hold corrupt politicians to account”, which clearly refer to retrospection. Following this perspective, whether or not voters punish malfeasant behaviour is a matter of what they prioritise: past outcome accountability over past procedural accountability. However, while citizens do take politicians’ past performance into consideration and use elections as a sanctioning tool, this may not be the only mechanism at play.

Empirical studies indicate that voters consider elections more as opportunities to select “good types” of political leaders rather than as accountability (“reward-punishment”) mechanisms (see Fearon, 1999). While the sanctioning or reward-punishment model is based on retrospective considerations, the selection model is based on prospective considerations about the “good types” of politicians (Fearon, 1999). In the latter model, these considerations are based to a great extent on the character evaluations of candidates, which provide cues as to whether candidates are likely to act on behalf of citizens in the future (independently of re-election incentives).

The relevance of the evaluations of the character, personality and traits of candidates has been widely recognised. These character evaluations are an important determinant of voting choices because they provide cues about the expected future behaviour of candidates. Pancer et al. (1999) identified a limited set of trait-dimensions that summarise the basis for political judgments and voting decisions across multiple contexts: competence, integrity, and warmth/empathy. In this sense, the “good politician” that citizens will vote for is often portrayed as principled, honest, skilful, competent and efficient (Fearon, 1999: 59).

People form their impressions of political leaders in the same way as they do of all kinds of people: by making inferences based on the information available (Funk, 1996). Thus, politicians’ decisions are consequential for the character traits that individuals associate with them. This should be especially the case when politicians address a situation that involves a trade-off or dilemma. Studies in social psychology suggests that reactions to situations that involve a dilemma between following moral principles or achieving a superior outcome are consequential for the trait inferences that witnesses make about decision makers (Uhlmann, Zhu, and Tannenbaum, 2013).

We therefore expect that the way politicians respond to a dilemma will influence citizens’ evaluations of their character and traits. Following a “selection model”, these character evaluations should, in turn, affect the likelihood that citizens will vote for that politician. Figure 1 summarises this logic. A pivotal aspect of our argument is that the electoral repercussions faced by politicians who circumvent laws and regulations in pursuit of a greater social good depend fundamentally on the electorate’s perception of their character and personal qualities. We focus on two specific traits that voters value in politicians: their perceived trustworthiness (one of the classical manifestations of integrity) and their perceived efficiency (a paramount manifestation of their competence). These traits capture the essence of the competence-integrity trade-off. This allows us to provide relevant insights into the reasons why citizens may (or may not) support corrupt politicians.

Figure 1. Analytical framework.

When analysing citizens’ reactions to the competence-integrity trade-off, it is important to distinguish between different degrees of malfeasance, though. Previous research has shown that citizens’ reactions to corruption scandals are shaped by the specificities of each type of corruption (Bauhr, 2017; Botero et al., 2021). In our case, we adopt a different strategy by distinguishing between two levels of malfeasance: breaking the law with and without deriving a personal gain. Since only the former qualifies as corruption, this distinction allows us to examine how citizens react to actions that involve procedural violations but are not necessarily perceived as corrupt. This distinction between these degrees of malfeasance is crucial for analysing the electoral consequences of the competence-integrity trade-off, because we expect that each of these degrees of malfeasance will lead to a different character evaluation of that politician and will, therefore, have different implications for voting behaviour.

Our basic expectation is that a politician who chooses to follow the law, even if bypassing it, would lead to a better social outcome will be perceived as more trustworthy (for simplicity, we refer to such a politician as a lawful politician). Everett et al. (2016) suggest that individuals who make these types of deontological decisions are perceived as more trustworthy. ¹ The decision to obey the law provides a relevant indication of the reliability of the politician’s future behaviour, a key aspect of the selection model. If a politician decides to follow the law, even when they have clear incentives not to do so, this may signal that they will follow the same principles in the future. This is relevant for political trust, as individuals entrust power to authorities without being able to directly monitor the actions of those authorities (Levi and Stoker, 2000).

Following the same reasoning, we expect that a politician who circumvents the law, even if they do so to achieve a superior societal outcome, will be seen as less trustworthy than a lawful politician. In this case, the behaviour of the political authority provides a negative cue about the reliability of her future behaviour. An authority who breaks the law once can be expected to do so at other times, even when that behaviour would not benefit society as a whole. In fact, people who use this type of consequentialist reasoning tend to be associated with negative traits such as, for example, being selfish or self-interested (Kreps and Monin, 2014).

The extent to which bypassing the law has a negative impact on political trust will depend, though, on whether the politician who breaks the law benefits privately or not from her actions (i.e., the degree of malfeasance). Studies in social psychology indicate that consequentialist actions can be considered more appropriate when the negative consequences they entail are only a side effect of the action (Cushman, 2013), or when they are not the main intent of the actor (Waldmann and Dieterich, 2007). This leads us to expect that a politician who breaks the law without obtaining any private benefit will be more trusted than one who privately benefits from it, even if both achieve the same (optimal) outcome through their actions (for simplicity, we refer to the former as the unlawful politician and to the latter as the corrupt politician). If a politician receives a private benefit as a result of breaking the law, one might question the motives that led the politician to forgo the legal procedure. Since this politician privately benefits from her action, one could think that breaking the law to obtain a personal gain was her main intention. This should provide an even more negative cue about the reliability of the corrupt politician’s future behaviour, and should reduce her trustworthiness even further.

Focusing now on competence, we expect that politicians who circumvent the law in order to achieve a better social outcome will be perceived as the most efficient. Someone who bends some laws when necessary may be perceived as more rational than someone who blindly follows legal procedures. The cost-benefit calculations made by pragmatic politicians who circumvent bureaucratic procedures to achieve a better outcome signal that they are actors guided by a sense of rationality that dominates over other considerations such as the legality or morality of their actions. They are likely to make decisions that maximise social welfare and avoid the loss of utility that might result from following burdensome bureaucratic procedures. This should send a strong signal about the efficiency that can be expected from these politicians, who might be seen as more competent to solve society’s problems. Conversely, politicians who choose to follow the law at the expense of public goods may be seen as less likely to weigh up costs and benefits in their decision-making processes, and more likely to be guided by a sense of morality and upright values. As a result, these politicians will be perceived as less efficient.

These theoretical arguments lead us to the following hypotheses:

H1: A lawful politician, who follows the law even when circumventing it would lead to a better societal outcome, will be more trusted than an unlawful politician, who circumvents the law in order to achieve a better societal outcome.

H2: A corrupt politician, who circumvents the law to achieve a better societal outcome and obtains a private gain, will be less trusted than an unlawful politician, who circumvents the law to achieve a better societal outcome but obtains no private gain.

H3: Unlawful and corrupt politicians, who circumvent the law to achieve a better social outcome, will be considered more efficient than lawful politicians, who follow the law even when circumventing it would lead to a better social outcome.

As we argue above, we expect that these changes in politicians’ character evaluations, triggered by their response to situations involving a trade-off between competence and integrity, will affect voting behaviour. Indeed, politicians who are judged to be more trustworthy or efficient enjoy greater support (Costa and Ferreira da Silva, 2015). In other words, changes in politicians’ character evaluation will mediate the impact of their reactions to the trade-off on voting behaviour (see Figure 1). The final (total) impact on voting behaviour will therefore depend on the extent to which each type of behaviour (legal, illegal, corrupt) affects the perceived trustworthiness and efficiency of politicians. If the gains in efficiency derived from circumventing the law to attain a better social outcome (H3) offset the loss of trust caused by this decision (H1 and H2), unlawful and even corrupt politicians may enjoy greater support than lawful politicians. However, if this is not the case, lawful politicians may enjoy the highest support. Overall, though, since the final effect of politicians’ responses to the trade-off on voting behaviour depends on the magnitude of the impact of politicians’ reactions to the trade-off on the evaluation of each of their traits (the mediators), we do not formulate a specific hypothesis about this final (total) effect. With the insights provided by the existing literature, we can hypothesise about the direction of the effects of politicians’ reactions on the mediators, and the effects of the mediators on voting behaviour, which can coexist and operate simultaneously. However, we are unable to hypothesise about the strength of these effects. Whether the unlawful or even the corrupt politicians will enjoy greater support than the lawful politician remains an open (theoretical) question that we address empirically.

Empirical strategy

Our analyses draw on a vignette experiment fielded in an online survey in Spain in December 2018. The sample of 1200 respondents was drawn from the Netquest online panel using quota sampling, including representative quotas for gender, age and education. ²

We would argue that Spain is a suitable case to conduct this experiment. It is a democratic country that regularly holds free elections and has an average level of corruption. According to Transparency International’s 2017 Corruption Perceptions Index, ³ it scored 57 on a scale of zero (highly corrupt) to 100 (very clean), ranking 42 out of 180 countries. Moreover, corruption scandals are frequent in Spanish politics (see e.g. Ares and Hernández, 2017). This makes Spain a good case to conduct our experiment, as we confront respondents with situations of malfeasance and corruption with which they are likely to be familiar. Moreover, in Spain, citizens report a very negative view of corruption in surveys, but often do not punish corrupt politicians in elections (Costas-Pérez, Solé-Ollé, and Sorribas-Navarro, 2012), making it a good case to research why voters disapprove of corruption but continue to vote for corrupt candidates. Moreover, the level of corruption in Spain is very similar to other European countries such as Portugal (63), Slovenia (61), Poland (60), Latvia (58), Czech Republic (57), Italy (50), Slovakia (50) and Greece (48). Spain also shares similarities with non-European countries such as Israel (62), Botswana (61), Costa Rica (59), Rwanda (56) and South Korea (51). Spain can therefore be seen as representative of many democracies with moderate levels of corruption.

The vignette of our experiment describes a hypothetical situation in which heavy rain has flooded a town and has destroyed its sewage system (see Table 1) ⁵ . The law stipulates that in order to repair the sewage system, the mayor of the town should hold a public bidding process (call for tenders) and award the repair to the best bidder; however, this would delay the repairs. The mayor foresees that bypassing the legal process (i.e., the call for tenders) and directly hiring an experienced company to carry out the repairs would speed up the resolution of problem, but it would be against the law.

Table 1 Wording of the experimental vignettes ⁴

After introducing participants to the situation, respondents were randomly assigned to one of the three different treatments. In the first treatment, the mayor follows the law but, as a result, there is a delay in solving the problem (lawful decision). In the second treatment, the mayor bypasses the legal procedure but prioritises the outcome (unlawful decision). Finally, in the third treatment, the mayor also prioritises the outcome and does not follow the legal procedure but, at the same time, this decision leads to a private gain for the mayor’s political party (corrupt decision). This third treatment closely matches most definitions of corruption. While in the second treatment the mayor simply breaks the law, in the third treatment the mayor breaks the law by favouring a company that in the past has provided a private gain for her political party in the past. A manipulation check confirms that our manipulation altered the perceived malfeasance and corruption of the mayor in the expected direction (see Figures A1 and A2, in the online Appendix). Respondents in the third treatment group are much more likely to think that the mayor is corrupt (perception of corruption: corrupt decision > unlawful decision > lawful decision).

We would argue that the dilemma presented in the vignette, although hypothetical, is realistic. Politicians often face a trade-off between resolving a public demand or problem quickly and following the established legal process. As we mentioned above, the need to procure medical equipment during the Covid-19 pandemic led many governments to bypass their standard procurement rules and award contracts through fast-track procedures (Transparency International, 2021). Indeed, bypassing calls for tenders to award public contracts to companies linked to a political party is a common way of engaging in corruption in Spanish politics (Leal, 2017).

In the vignette presented to respondents, we control away the effect of partisanship. We do so because previous studies have shown that partisanship can bias perceptions of corruption (Anduiza, Gallego, and Muñoz, 2013; Solaz, De Vries, and de Geus, 2018). The vignette always refers to a mayor of the respondent’s preferred party, so that the effects are comparable across all respondents. Respondents who declare no partisanship are excluded from our analyses. ⁶ Controlling for partisanship is important for two reasons. First, partisanship is a cue that is always available to voters when deciding who to vote for. Second, if we presented a fictitious mayor with no party affiliation, respondents might infer their partisanship from the mayor’s behaviour. The gender of the mayor was also randomised by manipulating his or her name. ⁷

The main dependent variable of our analyses is the propensity to vote for the mayor depicted in the vignette. This variable is measured by a direct question asking respondents how likely it is that they would vote for the mayor they have read about in the vignette, with 0 being “would never vote for this mayor” and 10 “would certainly vote for this mayor”.

To analyse the causal mechanisms, we asked respondents to evaluate if the mayor described in the vignette had certain traits. These items were asked after enquiring about the main outcome of interest (propensity to vote for the mayor). This question draws on the format commonly used in the American National Election Study to ask about candidate traits: Next, you will see a series of expressions that people can use to describe a politician. To what extent does each expression describe the mayor of the text you just read? “he/she is efficient”, “he/she can be trusted”. ⁸ For each of the statements, respondents could answer either quite well or quite badly. These questions allow us to measure the two basic dimensions discussed above: competence (efficiency) and integrity (trustworthiness).

Results

Average treatment effects

Figure 2 summarises the propensity to vote for the mayor depicted in the vignette. The average propensity to support the mayor is 5.58 on a 0-10 scale (where 10 indicates that the respondent is completely sure that they would vote for that mayor). This propensity to vote for the mayor differs significantly between the three experimental groups. The mayor who enjoys the highest average support is the one who follows the law, even if this leads to a suboptimal outcome (7.1). The mayor who circumvents the law in order to achieve a superior outcome (i.e., assigns the repair to an experienced company without holding a public tender in order to solve the problem faster) enjoys an average support of 6.18 (unlawful decision). Finally, the corrupt mayor, who assigns the repair directly to a company that solves the problem faster, but has contributed to her party’s election campaign, receives an average support of 4.75 (corrupt decision). The support for this mayor is therefore 2.35 points lower than the support enjoyed by the lawful mayor.

Figure 2. Mean propensity to vote across different treatment conditions.

Note: 95% confidence intervals around the means.

These results indicate that, on average, respondents prefer a mayor who respects the integrity of the procedure, even if this means achieving a suboptimal outcome, rather than one that achieves an optimal outcome by bypassing the legal procedure. It is worth noting that while the lawful and unlawful mayors are more likely to be supported than not (their average support is higher than the scale midpoint: 5), the corrupt mayor has an average support that is lower than the scale midpoint. Therefore, respondents are unlikely to support a mayor from the party they identify with if the politician obtains a superior societal outcome by corrupt means. Thus, respondents are much more likely to defect from their own party when violations of procedural integrity involve a private gain.

Focusing now on the personal traits that respondents infer about politicians who prioritise either superior outcomes or procedural integrity, Figure 3 summarises the proportion of respondents who believe that each of the mayors depicted in the vignette can be associated with a given trait. The lawful mayor is clearly the most trusted, the unlawful mayor follows her, and the least trusted is the corrupt mayor. It is worth noting that not only do the vast majority of respondents (around 85 percent) trust the mayor who respects procedural integrity, but also that a significant number of respondents (almost 64 percent) believe that the unlawful mayor is someone who can be trusted. The situation is quite different for a mayor who is openly corrupt, in which case only 43 percent consider her to be trustworthy. Therefore, in line with our expectations (H1 and H2), the degree of malfeasance has a clear negative impact on politicians’ trustworthiness, and trust takes the biggest hit in the presence of corruption.

Figure 3. Proportion of respondents considering the mayor trustworthy and efficient across different treatments conditions.

Note: 95% confidence intervals around the means.

Turning now to the perceived efficiency of the mayor, the politician who bypasses the established legal procedures to achieve a better societal outcome, but at the same time does not obtain any private gain is considered the most efficient of the three (suggesting that our experimental manipulation “works”). However, both the mayor who follows the legal procedures and the mayor who engages in corruption are considered less efficient. These results go against some of our expectations (H3). Since both the unlawful and the corrupt mayors prioritise outcomes and achieve a better result, we expected that both would be considered more efficient. However, the corrupt mayor receives the same efficiency rating as the lawful mayor, even though the former achieves an optimal outcome while the latter does not. One might speculate that individuals might doubt the motives that lead the corrupt mayor to opt for the course of action that achieves the optimal outcome if that decision also brings her private benefits. In this case, the inference about the present and future efficiency of that mayor might be less certain or tainted, since one does not know whether they would also prioritise the achievement of optimal societal outcomes if they had nothing to gain from that decision.

Mediation analysis

To formalise our analysis of the hypothesised causal mechanisms we now turn to mediation analysis. Traditional mediation analysis assumes the independence of the competing causal mechanisms (Imai et al., 2011). That is, we would assume that our mediators (trustworthiness and efficiency) do not influence each other. In our case, this independence assumption seems too restrictive. The inferred efficiency of the politician could affect her perceived trustworthiness and vice versa. Therefore, we turn to mediation analysis for multiple causally dependent mechanisms following the framework proposed by Imai and Yamamoto (2013), implemented through a varying-coefficient linear structural equation model. This method relaxes the independence of the mediator assumption. Moreover, within this framework, one can test the sensitivity of the results to violations of the key identifying assumptions: sequential ignorability and homogeneous interaction.

To increase the plausibility of these assumptions all the mediation models include the following pre-treatment covariates: sex, age, education, political interest, social trust, ideology, ideology squared, and a variable measuring the strength of each respondent’s party identification. In this way, we address the main criticism of mediation analysis—the lack of experimental manipulation of mediators—by controlling, first, for possible covariates that might be correlated with the mediator and the independent variable, and, second, for another possible mediator that might influence the main mediator of interest (Green, Ha, and Bullock, 2010). In addition, following the recommendation of Imai and Yamamoto (2013), we also assess the sensitivity of the results against violations of the sequential ignorability assumption (see the tests summarised in Figure A.3 in the Appendix).

First, we estimate the average causal mediation effect (ACME), which represents the causal effect of each of the treatments on the likelihood of voting for the mayor that are transmitted through the mediator of interest. This, together with the proportion of the total effect that is mediated, is the key estimate of interest for assessing the causal effect of each mediator, and, overall, our expectation of mediation. Second, we estimate the average direct effect (ADE), which represents the rest of the treatment effect that is transmitted either through the alternative mediator or through other mechanisms. Finally, the total effect, which represents the full change in the outcome variable that is attributable to the treatment, is obtained by adding the ACME and the ADE.

We focus first on the models in which we treat trustworthiness as the main mediator and efficiency as the alternative mediator. These models estimate the effects of the treatments on the likelihood of voting for the mayor that are mediated by changes in the perceived trustworthiness of the politician (H1 and H2), taking into account that there may be an additional character trait at play (efficiency) that may also mediate the effects of the treatments.

The ACME estimates, summarised in Figure 4, indicate that the relationship between the treatment and the outcome is significantly mediated by changes in the perceived trustworthiness of the mayor. 47 percent of the effect of the mayor that defies the established legal procedure on the probability to vote is mediated by the decrease in trust (taking the lawful mayor as the control baseline). Similarly, when comparing the corrupt and the unlawful mayors, a further decrease in trustworthiness accounts for 36 percent of the reduction in the likelihood of supporting the corrupt mayor (compared to the unlawful mayor). These results indicate that the decline in trustworthiness is one of the main factors explaining why the lawful mayor tends to be the one preferred by voters. Violations of procedural integrity, especially when they also involve personal gain, lead to a significant decrease in the belief that a politician is trustworthy. In turn, this belief has a sizable negative effect on the likelihood of supporting a politician. This result is in line with the idea advanced in recent accounts of the relationships between politicians’ traits and voting behaviour, which indicate that integrity-related traits are of primary importance for voting decisions (Laustsen and Bor, 2017).

Figure 4. Mediation analysis of trustworthiness with confounding alternative mechanism (efficiency).

Note: Percent mediated by trust: Unlawful vs. Lawful 46.72% / Corrupt vs. Unlawful 36.42% / Corrupt vs. Lawful 36.69%.

In the second set of models, we treat efficiency as the main mediator and trustworthiness as the alternative mediator. These models estimate the effects of the treatments on the likelihood of voting for the mayor that are mediated by changes in the perceived efficiency of that politician (H3).

The ACME estimates in Figure 5 indicates that the relationship between the treatment and the outcome is effectively mediated by efficiency only for some of our treatments. However, the extent to which efficiency mediates the treatment effects is substantially smaller than in the case of trustworthiness. Gains in the perceived efficiency of the politician only mediate the negative effect of adopting an unlawful decision on the likelihood of supporting the mayor by 14 percent (when compared to the mayor who adopts a lawful decision). This means that efficiency gains mitigate the drop in the likelihood of voting for the unlawful mayor, as respondents consider this mayor to be more efficient than the lawful one, and efficiency is positively related to the likelihood of supporting this mayor. However, while efficiency gains seem to reduce the electoral punishment attributable to the mayor’s decision, they are not sufficient to fully offset the negative effects of her decision. This is even more so in the case of the corrupt mayor, who does not actually see any gains in perceived efficiency as a result of her decision (see Figure 3 above). The lower perceived efficiency of the corrupt mayor, compared to the unlawful mayor, contributes to exacerbating the electoral punishment of the mayor who obtains a private gain through her decision. In this case, changes in efficiency mediate the effects of adopting a corrupt decision on the likelihood to vote for that mayor by 14 percent (compared to the likelihood of supporting a politician who adopts a decision that bypasses the law but does not involve a personal gain). This exacerbates the difference in the support of the unlawful and corrupt mayors. Therefore, it seems that in the case of the corrupt mayor, a loss in her perceived trustworthiness and the absence of any gains in her perceived efficiency, contribute to the severe electoral punishment of this politician. A comparison between the lawful and corrupt mayors can help us to better understand this situation.

Figure 5. Mediation analysis of efficiency with confounding alternative mechanism (trustworthiness).

Note: Percent mediated by trust: Unlawful vs. Lawful −14.42% / Corrupt vs. Unlawful 14% / Corrupt vs. Lawful .44%.

We expected that in a comparison between a politician who prioritises outcomes and one who prioritises the integrity of the procedure, the former would always increase her electoral support through perceived gains in efficiency. However, this does not always seem to be the case. The comparison between the lawful and the corrupt mayor indicates that efficiency does not deter the punishment of the corrupt mayor, since in this case the ACME value is 0. As we propose above, it is possible that citizens might doubt the reasons that motivate the corrupt mayor’s decision. This should weaken the inferences voters make about the efficiency of that politician.

These mediation analyses help us to examine the mechanisms driving the electoral punishment of corruption. However, they assume that, conditional on covariates, there is no unmeasured confounder in the relationship between the mediator and the outcome (sequential ignorability assumption) and no interaction between the two mediators (homogeneous interaction assumption). In Figure A3 in the Appendix we assess the sensitivity of our findings about the mechanisms against violations of these assumptions. The results reveal that the ACME estimates for efficiency are highly sensitive to violations of these assumptions. Hence, our experiment only provides weak support for our expectation about the mediating effects of inferring efficiency of a politician who adopts a decision that prioritises either achieving superior outcomes or respecting procedural integrity. Conversely, the sensitivity analyses reveal that our conclusions are much more robust when it comes to the mediating effects of trustworthiness.

Conclusion

This paper addresses a long-standing question in the literature on corruption voting, namely whether and to what extent citizens might tolerate corruption when politicians get the job done. Specifically, we ask whether and why citizens might trade procedural integrity for better societal outcomes. We argue that the best way to do so is through a design that directly pits procedural integrity against the capacity (competence) of politicians to deliver public goods. The experiment implemented in this study allowed us to assess the effect of a political decision that achieves an optimal outcome but violates procedural integrity. The results indicate that the way in which politicians respond to situations that pose a dilemma between competence and integrity affects citizens’ voting behaviour.

In line with observational evidence on citizens’ preferences (Allen, Birch, and Sarmiento-Mirwaldt, 2016), respondents prefer a politician who follows the law, even if that implies achieving a suboptimal outcome. Contrary to Esaiasson and Muñoz (2014), our results confirm that, all else equal, citizens prefer “honest incompetent politicians” to “competent corrupt politicians”. However, our results also reveal that the politician who breaks the law in order to achieve an optimal outcome without obtaining a private benefit for herself is more likely to be supported than not (her average support is higher than the midpoint of the propensity to vote scale). Yet, support for a politician falls sharply if they break the law to achieve an optimal outcome and at the same time they obtain a private gain from it. When politicians break the law to achieve a better societal outcome and they do not privately benefit from it, citizens are much more likely to support them than when their decision involves a direct private gain. By distinguishing between two degrees of malfeasance (breaking the law with and without a private gain), our analyses add some nuance and complement previous studies that have analysed whether corruption is punished when politicians also manage to get the job done. According to our results, a corrupt incumbent is punished even if they are efficient and deliver. However, if breaking the law is not linked to corruption (i.e., there is no private benefit for the politician), the punishment for malfeasance is significantly milder.

To shed more light on these effects, we analysed why corruption is punished more severely than just breaking the law. The results indicate that some of the inferences that respondents make about the character and attributes of decision-makers are the key mechanisms. Breaking the law undermines trust in decision-makers, even if they manage to achieve a better social outcome by circumventing the legal process. However, this reduction in trust is much more pronounced when a politician breaks the law while at the same time benefiting privately. That is, in cases of outright corruption. We would argue that citizens may doubt the true motives of someone who breaks the law in order to achieve a superior societal outcome, but also obtains a private benefit as a result. While politicians who break the law are always less trusted, we hypothesised that if, as a result of bypassing the legal procedure, those politicians are better able to get the job done, they should also be considered more efficient. Our results indicate that the loss of trust associated with breaking the law is partially offset by the perception that such politicians are more efficient. However, there are some caveats to this claim. First, the gains in efficiency are not sufficient to fully offset the negative impact that the decline in trust brings for the electoral support of those politicians. Second, only those politicians who do not privately benefit from breaking the law are perceived to be more efficient.

In light of this evidence, we can conclude that a loss of trust is one of the main mechanisms driving the electoral punishment of corruption. While in some particular cases politicians may be able to counter the negative impact of corruption on their electoral fortunes by pointing to their track record and efficiency, this is unlikely to work. In cases of blatant corruption, citizens will not perceive corrupt politicians as more efficient, regardless of their performance. And if they do, the gains in their perceived efficiency will not outweigh the negative impact of corruption on their trustworthiness.

Some of the potential limitations of these findings are related to the experimental methodology that we used in this paper. Recent studies have raised concerns about the suitability of survey experiments to accurately assess citizens’ voting behaviour, suggesting that the effect of corruption on electoral accountability may be overestimated due to social desirability and hypothetical bias (Boas, Hidalgo, and Melo, 2018; Incerti, 2020). Social desirability bias refers to respondents willingly over-reporting socially acceptable responses, while hypothetical bias refers to respondents behaving differently in surveys than they would in real life because they are not faced with the actual costs associated with that behaviour (Incerti, 2020). However, other studies have found no evidence of overreporting of disapproval of corruption due to social desirability bias (Pavao, 2015; Breitenstein, Anduiza, and Muñoz, 2022). In any case, our paper addresses these potential limitations in several ways. First, we specifically designed the experiment in the form of a direct trade-off between honesty and competence. This, like any other trade-off scenario, is likely to reduce social desirability and hypothetical bias (see e.g., Ares et al., 2024). Our design mitigates concerns about social desirability bias, as neither scenario presented to respondents can be definitively identified as more acceptable. It also mitigates concerns about hypothetical bias by better replicating the trade-offs and costs faced by citizens in real-world scenarios. Second, we increase the external validity of our survey experiment by closely replicating actual dilemmas faced by politicians. Third, by implementing a gradual treatment of malfeasance, this paper is better able to capture respondents’ true preferences about procedural integrity. Even so, we should keep in mind that in real-world settings, where the short-term benefits of breaking the law are more tangible and the negative consequences are imperceptible, voters may be more willing to condone corruption (as shown in Fernández-Vázquez, Barberá, and Rivero, 2013). This is a relevant limitation of our design.

Our results should therefore be interpreted with some caution, as they reflect the impact of the integrity-competence trade-off in the specific context of our study. Our reading of these results, in conjunction with the findings of other recent studies designed with the aim of reducing social desirability (Winters and Weitz-Shapiro, 2013; Breitenstein, 2019; Vera, 2020; Klašnja, Lupu, and Tucker, 2021), is that, all else equal, voters generally prefer non-corrupt candidates. However, in the actual process of selecting a candidate during a specific election, various other considerations often come into play, and the all-else-being-equal condition is often not satisfied (Breitenstein, Anduiza, and Muñoz, 2022). In any case, our experiment contributes to the literature by showing that voters do care about integrity and that, ideally, they would like to be governed by law-abiding politicians. We would argue that this is a relevant result that contributes to the expansion of our cumulative knowledge about corruption accountability, even if, in some cases, voters may end up voting for a corrupt politician due to lack of credible information (Botero et al., 2015; Weitz-Shapiro and Winters, 2017), lack of clean alternatives (Charron and Bågenholm, 2016; Agerberg, 2020), partisan allegiances (Anduiza, Gallego, and Muñoz, 2013; Solaz, De Vries, and de Geus, 2019), or the need to coordinate with other voters to throw the rascals out.

In this study, we have shown how respondents react to a dilemma between achieving an optimal social outcome and following the law. The experiment intends to realistically portray a scenario in which politicians face a dilemma between urgently solving a public problem and following the legal procedures. It remains for future research to test respondents’ reactions to variations of this dilemma, involving different public goods and with different levels of urgency. Moreover, the dynamics and reactions to how this trade-off is resolved may vary considerably depending on the contextual levels of corruption (Klašnja and Tucker, 2013). While we argue that Spain is an ideal setting to test our theoretical claims, and it is representative of other countries with a moderate level of corruption, we expect that these findings will inspire other researchers who may want to test the theoretical claims that we develop in our paper in other contexts and scenarios.

Finally, in this paper we have focused on mediation effects because this has allowed us to understand the mechanisms that drive the punishment of corruption. However, further research could build on these findings and examine moderation instead of mediation. First, future research could analyse how characteristics such as the gender, education or sophistication of respondents influences how citizens navigate the integrity-competence trade-off. Second, researchers could analyse how changes in the evaluation of politicians’ traits and their impact on voting behaviours may differ depending on respondents’ characteristics (i.e., moderated mediation).