832230 research-article20192019 RAP0010.1177/2053168019832230Research & PoliticsBreitenstein Research Article Research and Politics Choosing the crook: A conjoint January-March 2019: 1­–8 © The Author(s) 2019 Article reuse guidelines: experiment on voting for sagepub.com/journals-permissions https://doi.org/10.1177/2053168019832230 DOI: 10.1177/2053168019832230 corrupt politicians journals.sagepub.com/home/rap Sofia Breitenstein Abstract The coexistence of harsh disapproval of corruption and the limited electoral consequences of malfeasant behavior remains a conundrum in social sciences. While elections should be used to hold politicians accountable, evidence shows that voters only mildly punish corrupt politicians. This paper assesses the trade-off hypothesis, which suggests that voters forgive corrupt candidates when these candidates provide other valued outcomes. It distinguishes two possible factors against which integrity can be traded—partisanship and economic performance—and tests them in a multidimensional experiment. With the results of an original conjoint analysis, this paper provides compelling evidence for the relative importance of corruption when casting a vote and the mitigating effects of other valued candidate characteristics. Even when obtaining highly credible information, partisanship determines the vote to the same extent as corruption. Additionally, co-partisanship and a strong economic performance moderate the negative effect corruption has on the vote. Keywords Corruption, voting, experiment, partisanship, economic performance Corruption has severe negative consequences for societies. According to both the sanctioning and the selecting model It hinders economic development (e.g. Mauro, 1995), of voting behavior, it is theoretically unconceivable why an increases economic inequality (e.g. Fisman and Golden, informed and free citizen would vote for a corrupt incumbent 2017), decreases trust in government and other political (for a review on voting models, see Healy and Malhotra, institutions (e.g. Ares and Hernández, 2017), and obstructs 2013). Corruption is a clear signal of a harmful government the fundamental democratic principle of citizens’ impartial that will not act in the voters’ best interests, and also a dis- access to political institutions (Rothstein and Varraich, tinct motive to sanction the ruling government (Fearon, 2017). Despite its damaging side effects, corruption is still 1999). Yet, empirical evidence drawn from observational widely spread and is present in all types of political systems data, field and survey experiments, shows that voters around (Shleifer and Vishny, 1993). Even though research on the the world only mildly punish corrupt politicians (e.g. Chang causes and remedies of corruption has exponentially grown et al., 2010; Dimock and Jacobson, 1995; Peters and Welch, since the 1990s, it is still unclear as to what the best strate- 1980; Reed, 1999). However, when interviewed in surveys, gies are to curb corruption; some anti-corruption policies citizens were able to identify malfeasant activities, express have failed dramatically (Fisman and Golden, 2017). One their clear rejection of corruption (Afrobarometer Round 31; of the key queries that must be further addressed is the pres- World Values Survey, Wave 6, V2022), and their low inten- ence of corruption in democratic countries. From a theo- tions to support corrupt politicians (e.g. Muñoz et al., 2016). retical point of view, it is puzzling how corrupt governments survive in democratic societies, as one of the main func- Universitat Autònoma de Barcelona, Spain tions of free elections, a fundamental to every democracy (Dahl, 1971), is to hold governments accountable. Schmitter Corresponding author: Sofia Breitenstein, Campus de la Universitat Autònoma de Barcelona, and Karl (1991) define modern political democracy as “a Departament de Ciència Política i Dret Públic, 08193 Bellaterra, system of governance in which rulers are held accountable Spain. for their actions in the public realm by citizens…” (p.76). Email: sofia.breitenstein@uab.cat Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 Research and Politics  Different arguments have been used to explain the coex- set their attention on a trade-off between integrity and the istence of a harsh disapproval of corruption and the low competence of candidates in other areas. According to this electoral punishment of corrupt politicians (for a review, literature, voters disregard corruption when candidates see De Vries and Solaz, 2017). This paper tests the trade-off deliver other benefits, such as economic growth or other hypothesis, which proposes that voters condone corruption public goods. Several prior studies on voting for corrupt when politicians offer other valued characteristics. While politicians provide evidence that corruption is less pun- some studies have looked at the moderating effects of (co) ished in good economic contexts (Klašnja and Tucker, partisanship, others have looked at the conditioning effects 2013; Zechmeister and Zizumbo-Colunga, 2013), when the of the candidate’s economic performance. Through a multi- politician has a good management record (Esaiasson and dimensional survey experiment, this article tests both trade- Muñoz, 2014; Muñoz et al., 2016), or when he/she has off hypotheses in one experimental setting. It consequently implemented favorable economic policies (Konstantinidis increases the external validity of the study, as it provides a and Xezonakis, 2013). Nevertheless, other research did not more accurate account of what is happening in real elec- obtain the same results. According to Winters and Weitz- tions where voters are confronted with multiple trade-offs Shapiro (2013), when voters learn about corruption, they when casting their vote. In addition, the paper discusses punish those candidates even if they performed well. two means of understanding the trade-off hypothesis and Therefore, it is essential to reassess this hypothesis and the differential impact these could have on anti-corruption verify if the economic performance of a candidate actually policies. determines whether voters condone corruption at the polls. Likewise, this study poses a hard test for this hypothesis as respondents also received information on the candidates’ Theoretical framework partisan affiliations. Regarding the reasons that informed and free citizens vote This paper provides at least two contributions to the lit- for corrupt politicians, the literature has highlighted the erature. It is one of the first studies to test both trade-offs in relative importance of corruption when selecting a candi- one survey experimental setting. By using the data of an date. According to Rundquist et al. (1977), voters seem to original conjoint analysis, this study evaluates how respond- care about corruption, but they also have other concerns ents make decisions in a multidimensional scenario where and may trade integrity for other valued characteristics of they have to take several trade-offs into consideration. the candidate. These authors pointed to an implicit exchange Whereas most previous studies have tested the moderating between the integrity of the candidate and her position on effect of partisanship and a good economic performance certain policies or issues. Voters might forgive corruption separately, this paper reassesses their moderating effect in a when malfeasant candidates take a position on issues that multidimensional scenario. Respondents received informa- are more important to them. tion on various candidates’ characteristics, thus posing a A number of previous studies have shown that partisan- more difficult test for both hypotheses. ship moderates the perception of corruption (e.g. Anduiza Currently, other researchers are assessing corruption et al., 2013; De Vries and Solaz, 2017; Eggers, 2014) and accountability with a multidimensional perspective. electoral fraud (Beaulieu, 2014). In a survey experiment However, the main aim of these studies differs substantially conducted in Spain, Anduiza et al. (2013) found that from this article, both in the hypotheses they are trying to respondents consider corruption less severe when it affects assess and in their research design. While in this paper I the party they feel closer to. Nevertheless, Konstantinidis assess the moderating effect of candidates’ characteristics and Xezonakis (2013) did not corroborate these results in on the electoral punishment of corruption, Visconti and a study conducted in Greece. The authors of the latter Mares (2018) and Klašnja et al. (2017) implemented con- study argued that the different results might be due to the joint experiments to mainly assess whether certain corrup- strong partisan alliances in place in the political context tion characteristics moderate the negative effect of where the survey experiment by Anduiza et al. (2013) was malfeasant activities. Perhaps the study by Franchino and launched. In this paper, I reassess these results in the same Zucchini (2014) is most closely related to this article; how- country, but during a time when new parties have recently ever, this also differs considerably from this paper in terms entered the political arena and, therefore, party loyalties of the literature it engages with and the candidate character- might not yet be as strong. istics that it assesses (refer to online Appendix B for a more Moreover, this paper poses a harder test for the partisan specific comparison). trade-off. Respondents also received varying information In addition to assessing the conditional effect of co-par- regarding other candidates’ attributes that have been found tisanship and the economic performance of a candidate, to be important determinants of the vote, such as the eco- this study also estimates and compares their relative weight nomic performance of the candidate (for a review, see on the likelihood of voting for a politician. Participants Lewis-Beck and Stegmaier, 2007). Building on the same were presented with profiles of two mayors with randomly idea of the relative importance of corruption, recent studies assigned information on the candidates’ party affiliation, Breitenstein 3 economic performance, integrity (corruption), educational implications for anti-corruption campaigns. If voters are and managerial qualities, and gender, and were asked to rational and vote according to the importance they attach report their probability of voting for each candidate. This to each candidate’s characteristics, anti-corruption cam- allows for determining the relative causal effect of each paigns simply have to inform about politicians’ malfeas- candidate’s characteristics on the respondent’s probability ant behavior, insist on the negative consequences of of supporting the candidate (Hainmueller et al., 2014). corruption for society, and address the necessity for voters The distinction between the relative importance of dif- to hold politicians accountable. However, if voters are ferent factors and their moderating effect on the electoral driven by unconscious biases, the strategies to motivate punishment of corrupt candidates is directly linked with the them to hold politicians accountable at the polls might not second contribution of this paper. Previous research has not be as straightforward and successful. In that case, even if agreed on exactly what the trade-off hypothesis entails, and voters are informed about the negative consequences of is, therefore, unclear on how to measure it. Some studies malfeasant behavior and their ability to hold politicians have looked at the factors that moderate the electoral pun- accountable, psychological biases, driven by co-partisanship ishment of corruption, while others have theorized about or a strong economic performance, might affect how they the relative importance of corruption. Nevertheless, these perceive the severity of malfeasant activities of their pre- two ways of explaining the trade-off hypothesis entail a dif- ferred politicians. ferent causal mechanism and, therefore, may entail a differ- ent consequence for policies that aim to combat corruption Empirical strategy by increasing citizens’ awareness. The relative weight argument, drawn from the discus- To test the trade-off hypotheses, this study uses the data of sions in previous literature and the public arena, proposes an original conjoint experiment embedded in a representa- that corruption is just another factor voters consider when tive survey of the Spanish population. Survey experiments casting their vote and that they might give more importance have proven to be a unique technique to assess causal infer- to issues other than corruption (e.g. Fisman and Golden, ence and to overcome social desirability biases, which 2017). This explanation is in keeping with the argument of could be in place in this study as it is dealing with a sensi- Rundquist et al. (1977), which assumes a rational voter that tive issue (for a review on advantages and limitations of weighs the candidates’ characteristics and votes accord- survey experiments, see Sniderman, 2018). In conjoint ingly. According to this hypothesis, voters choose a crooked experiments, in contrast to standard survey designs, several politician when they weigh her partisan affiliation or her pieces of information are manipulated in one setting. It con- economic performance stronger than her integrity. The sec- sequently allows for varying and analyzing different dimen- ond approach, which I call the “conditional punishment sions of the studied phenomenon and so increases the argument”, proposes that the negative effect of corruption external validity of the research. on the support of a candidate may diminish when the candi- Respondents were presented the profiles of two mayors date exhibits other positive characteristics. This argument is with a set of five attributes with independently randomly in keeping with the research, showing that a good economic assigned categories (or components) and were asked to situation moderates the negative evaluation of a corrupt can- report their likelihood of voting for each candidate if they didate (e.g. Zechmeister and Zizumbo-Colunga, 2013). In were running for elections in their hometown. Each this case, the causal mechanism could be less rational, as respondent was asked to repeat the same task three times voters might vary how they weigh a candidate’s corruption with random and varying pairs of candidates. The experi- allegations depending on other candidate qualities. ment was completely randomized, so no combination of Therefore, the assumption here is not that voters rationally attributes was restricted; the sequences of the attributes choose a corrupt candidate because they give priority to were also randomized across each respondent but kept con- other characteristics. In this case, voters could be equally or stant over the three tasks. This procedure allows for the more concerned with the candidate’s integrity; nevertheless, estimation of the relative influence of each attribute and other positive candidate characteristics could unconsciously enables us to assess how the attributes interact with each influence integrity’s relevance on their decision-making. other (Hainmueller et al., 2014). Hence, the conjoint design For example, Anduiza et al. (2013) provide evidence that is ideal to evaluate what is the variable that most deter- respondents diminish the perceived severity of a co-partisan’s mines the vote (relative weight argument) and whether any corrupt activity to avoid cognitive dissonance. factors moderate the negative effects of corruption (condi- Both hypotheses have different substantive implica- tional punishment argument). tions. While the relative weight hypothesis conceives a The experiment was embedded in an online survey rational voter that simply chooses to overlook corruption, (n = 2275) in Spain in June 2016. The sample included the conditional punishment hypothesis conceives a voter quotas to achieve an accurate representation of the that might be unconsciously driven by psychological Spanish population (refer to online Appendix B for more biases. Moreover, these two conceptions have different information about the survey). Spain is an appropriate 4 Research and Politics  Table 1. Attributes and text for each component. definitely vote for”). The answers were rescaled from zero to one. I selected this dependent variable instead of a forced Sex Female choice between candidates because the probability of vot- Male Party PP ing allows respondents to not vote for any candidate and, PSOE therefore, bears a stronger resemblance to real elections.3 Podemos Furthermore, Hainmueller et al. (2015) show that paired Ciudadanos conjoint experiments with a question for each option are Qualities Has compulsory education and little better at eliciting behavior in real situations. management experience Overall, there were 12,284 evaluated profiles or 6142 Has university education and prolific pairs of candidates. Table A.1 in the online Appendix shows management experience the distribution of the vote probability across the different Economic Investments in the municipality have treatments. The average vote probability of an honest can- performance increased so that unemployment has didate is 0.49. decreased by 5% Investments in the municipality have decreased so that unemployment has Results increased by 5% In this article, the probability of respondent i voting for a Corruption Has been characterized for his/her honesty candidate k in task j is modeled as a function of the candi- Has been accused by other parties of corruption for awarding contracts in date’s (co)partisanship,4 economic performance, integrity, exchange for gifts and other characteristics: Has been accused by the judge of Pvijk = β 1* Partisanshipijk + β 2 * corruption for awarding contracts in exchange for gifts Economicperformanceijk + β 4 * Corruptionijk + β 5* Otherijk + eijk setting to conduct this experiment for various reasons. It According to the relative weight model, voters elect cor- is a democratic country that holds free elections with rupt candidates when they weigh other issues or character- average levels of corruption, scoring 58 on a scale of istics more strongly than corruption. Consequently, this zero (highly corrupt) to 100 (very clean) in 2015 accord- hypothesis is corroborated if the effect of other variables, ing to the Corruption Perception Index of Transparency such as partisanship and economic performance, is stronger International. Thus, corruption scandals are salient than the effect of corruption on the probability to vote for a issues in Spanish politics. This makes it a good setting to candidate. test our hypotheses, as respondents are confronted with The complete randomization of the experiments allows situations they have either experienced or, at least, heard us to estimate the average marginal component effects of and read about. Moreover, Spain is a typical case in (AMCEs) by fitting a linear regression and clustering for that citizens report a highly negative view of corruption respondents, as each respondent repeated the same task in surveys but often do not punish corrupt politicians in three times (Hainmueller et al., 2014).5 The AMCEs (shown elections. According to a survey conducted by the offi- in Figure 1) can be interpreted as the marginal effects of cial survey institute of the Spanish government (Centro changing a given characteristic (or category) on the popula- de Investigaciones Sociológicas, 2905) in 2011, 87% of tion’s probability to vote for a candidate averaged over all respondents considered that corruption is a problem of possible values of the other characteristics. Results show paramount importance. Consistent with these percep- that the information about corruption has a strong negative tions, only 10% acknowledged that they would vote for effect on support. The accusation of corruption by other a corrupt but efficient candidate. However, if we turn to political parties decreases the support of a candidate by the actual electoral results, electoral punishment is very 0.22 (on a scale from 0 to 1) as compared to the level of low or even nonexistent (Costas-Pérez et al., 2012; support for an honest candidate, while the accusation of a Rivero and Fernández-Vázquez, 2010). judge does so by 0.27. It is especially remarkable to observe Table 1 shows the categories for each attribute and the that even though the experiment uses a strong treatment of corresponding text that was shown in the experiment. corruption and refers explicitly to corruption, under certain Please refer to online Appendix B for a precise justification conditions, partisanship has an equally strong effect on the of the selection and operationalization of each attribute support of a candidate as corruption.6 Seeing the profile of included in the experiment. a candidate belonging to a different party decreases the sup- Concerning the dependent variable, respondents port to the same degree as an accusation of corruption by a expressed their probability of voting for each candidate on judge. These results were corroborated using a variable that a scale from zero (“would never vote for”) to 10 (“would differentiates between nonpartisans and partisans of another Breitenstein 5 party and another variable that measures party preferences a corrupt candidate. For example, respondents’ average instead of partisanship (see online Appendix A, Figures A1 probability of voting for an honest candidate from a dif- and A2). ferent party with a weak economic performance and low Regarding the economic performance of the candidate, a educational and managerial qualities is only 0.39, while weak economic performance has a significant negative the probability of voting for a co-partisan candidate with effect on the support of a candidate, but this effect is con- a strong economic performance and high educational and siderably weaker than the effect of corruption. A low edu- managerial qualities who is accused of corruption by other cation and little management experience also significantly parties is 0.61. decrease the level of support for a candidate. To test the conditional punishment argument, it is neces- Overall, the results of the experiment show that sary to determine whether the negative effect of corruption respondents not only care about corruption, but there are on the support of a candidate varies according to the parti- other candidate characteristics that determine the likeli- sanship or the economic performance of the candidate. To hood of voting for that candidate. Indeed, a certain combi- do that, I estimate the average difference in the AMCEs of nation of variables increases the probability of voting for corruption between different profiles of candidates (see Hainmueller et al., 2014: 12). The conditional punishment hypothesis is corroborated if corruption has a weaker nega- tive effect when the candidate is a co-partisan or delivered Woman a strong economic performance. Table 2 shows the percentage of change in the probabil- Different party ity of voting for a co-partisan candidate or a candidate with a strong economic performance when accused of corrup- Low qualities tion (grouping both partisan and judicial accusation of cor- ruption). The results show that an accusation of corruption Weak performance has a weaker negative impact on the likelihood of receiving a vote when the candidate belongs to the same party as the Accused parties voter. Corruption decreases the probability of voting for a candidate belonging to a different party by 52%, while the Accused judge vote probability only decreases by 40% for a co-partisan candidate. The semi-elasticities in Table 3, which provide -.3 -.2 -.1 0 .1 the proportional change in Y for a change in X, confirm that this differential impact is strong and significant at a 99% Figure 1. Average marginal component effects (AMCEs). confidence interval. As far as the economic performance of Table 2. Predicted probabilities and the relative reduction of corruption. Same party Different party Vote Confidence Reduction of Vote Confidence Reduction of probability interval vote probability probability interval vote probability Lower Higher (%) Lower Higher (%) Honest 0.746 0.724 0.768 0.445 0.431 0.459 Corrupt 0.448 0.426 0.470 39.930 0.211 0.201 0.220 52.658 Strong economic performance Weak economic performance Vote Confidence Reduction of Vote Confidence Reduction of probability interval vote probability probability interval vote probability Lower Higher (%) Lower Higher (%) Honest 0.542 0.525 0.558 0.441 0.425 0.456 Corrupt 0.278 0.266 0.289 48.695 0.216 0.206 0.227 50.931 Note: the first column of each group shows the predicted values of respondents voting for a candidate. The last column shows the relative reduction in the probability of voting for the same candidate with the added component of being accused of corruption (grouping both partisan and judicial accusation of corruption), taking always as a reference the probability of voting for an honest candidate in each group. 6 Research and Politics  Table 3. Derivatives expressed as a semi-elasticity. corruption. However, this conditioning effect is mild and does not hold across all robustness checks. In summary, ey/dx Contrast this study corroborates the importance of partisanship in Partisanship condoning corruption. While this study finds some evi- Different party –0.764 (Reference) dence in favor of the economic performance trade-off, Same party –0.512 0.251*** these results are by no means as strong as the partisanship Economic performance trade-off. Weak performance –0.763 (Reference) As shown in the analysis of this paper, the relative weight Strong performance –0.699 0.064* and the conditional punishment model can be compatible, as ***p < 0.01. partisanship determines the vote to the same extent as cor- **p < 0.05. ruption; however, partisanship also moderates the effects of *p < 0.1. corruption. Although these models are not mutually exclu- Note: the first column shows the proportional change of Y when the sive, it is important to distinguish them in future research as candidate is corrupt across the different categories of partisanship and economic performance. The contrast column shows the difference in they could entail different causal mechanisms based on dis- the change of Y between each reference category and the rest of the tinct rationality in voters’ decision-making. Ultimately, categories and the significance associated with these differences. these models could have a differential impact on anti-cor- ruption policies that aim to engage citizens in the control of the candidate is concerned, the conditional punishment corruption. Due to the properties of the conjoint design, in hypothesis only holds true to some extent. Information on this paper, I could not test the rationale behind respondents’ corruption has a weaker effect when a candidate delivered decisions. Future research should now determine the exact a strong economic performance; nevertheless, the differ- pathways that might drive each argument. ence is small (2 percentage points) and only holds at a 90% Concerning the elicitation of stated preferences with confidence interval. In summary, the results of this experi- hypothetical scenarios, Hainmueller et al. (2015) demon- ment show that the negative effect of corruption is indeed strated that paired conjoint experiments are highly success- attenuated by co-partisanship and, to a lesser extent, by the ful in replicating the decision-making that takes place in economic performance of the candidate. real settings. Furthermore, I consider the high credibility of the information provided in this experiment an advantage of this study as it poses a solid test for the trade-off hypoth- Conclusion eses. This paper shows that even when obtaining highly This paper provides compelling evidence that respondents credible information and, therefore, being certain that a care about corruption but also care about other candidate candidate is corrupt, respondents are willing to trade integ- characteristics. Analyzing the results of a conjoint experi- rity for other valued characteristics of the candidate. Hence, ment proves that in a multidimensional setting, participants I provide clear-cut evidence that even informed citizens trade out integrity for other valued characteristics. While might choose to overlook corruption. previous studies have tested the moderating effect of parti- sanship and a good performance individually, this study Acknowledgements contributes to the literature by assessing how respondents I would like to thank Eva Anduiza and Jordi Muñoz for their react when they also receive information on many other insightful comments and suggestions. I also thank the two anony- candidates’ characteristics. In addition, this study helps us mous reviewers and the editors of Research and Politics for their to understand not only the moderating effect of partisanship valuable feedback. Previous versions of this paper were presented and a strong economic performance, but also their relative at the International Meeting on Experimental and Behavioral importance on the probability of voting for a candidate. Social Sciences 2016, the ECPR General Conference 2016, and The results of this multidimensional survey experiment the EPSA General Conference 2017. Thanks to all discussants and confirm a trade-off between integrity and co-partisanship. participants for their helpful comments. In line with previous findings (Anduiza et al., 2013; Barnes and Beaulieu, 2014; Beaulieu, 2014), co-partisanship Declaration of conflicting interests strongly moderates the negative effect corruption has on The author(s) declared no potential conflicts of interest with respect the likelihood of voting for a politician. Furthermore, parti- to the research, authorship, and/or publication of this article. sanship, together with corruption, is the attribute that most determines the vote. However, the results only partly cor- Funding roborate a trade-off between the economic performance The author(s) disclosed receipt of the following financial support and the integrity of a candidate. While in this experiment for the research, authorship, and/or publication of this article: This corruption determines the vote to a much greater extent work was produced in the context of the author’s PhD in Politics, than the economic performance of a candidate, the eco- Policies and International Relations at the Universitat Autònoma nomic performance does moderate the negative effect of de Barcelona, funded through an FPI grant (BES-2015-072756) Breitenstein 7 from the Spanish Ministry of Economy and Competitivity and the Costas-Pérez E, Solé-Ollé A and Sorribas-Navarro P (2012) European Social Fund. The data were collected within the project Corruption scandals, voter information, and accountability. “Living with Hard Times: How Citizens React to Economic Crises European Journal of Political Economy 28(4): 469–484. and Their Social and Political Consequences” (LIVEWHAT). This Dahl RA (1971) Polyarchy: Participation and Opposition. New project was funded by the European Commission under the 7th Haven, CT: Yale University Press. Framework Program (grant number 613237). De Vries CE and Solaz H (2017) The electoral consequences of corruption. Annual Review of Political Science 20(1): Supplemental materials 391–408. Dimock MA and Jacobson GC (1995) Checks and choices: The The supplemental files are available at http://journals.sagepub. House Bank Scandal’s impact on voters in 1992. The Journal com/doi/suppl/10.1177/2053168019832230 of Politics 57(4): 1143–1159. The replication files are available at https://dataverse.harvard.edu/ Eggers AC (2014) Partisanship and electoral accountability: dataset.xhtml?persistentId=doi:10.7910/DVN/HOSWS7 Evidence from the UK expenses scandal. Quarterly Journal of Political Science 9(4): 441–472. Notes Esaiasson P and Muñoz J (2014) Roba pero Hace? An experimen- tal test of the competence-corruption tradeoff hypothesis in 1. A clear majority of respondents across all countries consid- Spain and Sweden. QOG Working Paper Series The Quality of ered that it is wrong and punishable if a government offi- Government Institute 2014. Available at: http://qog.pol.gu.se/ cial (i) “gives a job to someone from his family who does digitalAssets/1551/1551606_2014_02_essaiasson_jordi.pdf not have adequate qualifications”, (ii) “demands a favor or Fearon JD (1999) Electoral accountability and the control of poli- an additional payment for some service that is part of his ticians: Selecting good types versus sanctioning poor per- job”, or (iii) “decides to locate a development project in an formance. In: Przeworski A, Stokes SC and Manin B (eds) area where his friends and supporters lived” (Q58a–c) (“The Democracy, Accountability, and Representation. Cambridge: online data analysis tool | Afrobarometer,” n.d.). Cambridge University Press, 1999, pp. 55–97. 2. A clear majority of respondents across all counties consider Fisman R and Golden MA (2017) Corruption: What Everyone “someone accepting a bribe in the course of their duties” (V202) Needs to Know. New York: Oxford University Press. is never justifiable (“World Values Survey Database,” n.d.). Franchino F and Zucchini F (2014) Voting in a multi-dimensional 3. Refer to online Appendix A (Tables A15–A17) for the results space: A conjoint analysis employing valence and ideol- of the forced choice as a dependent variable. ogy attributes of candidates. Political Science Research and 4. Measured by combining the respondent’s party identifica- Methods 3(2): 221–241. tion, acquired before the experiment, and the party that is Hainmueller J, Hangartner D and Yamamoto T (2015) Validating being assessed in the experiment. vignette and conjoint survey experiments against real-world 5. Results are corroborated using a linear regression with indi- behavior. Proceedings of the National Academy of Sciences vidual fixed effects. of the United States of America 112(8): 2395–2400. 6. Refer to online Appendix A for several robustness checks. Hainmueller J, Hopkins DJ and Yamamoto T (2014) Causal infer- These results hold along all tests except for the model that ence in conjoint analysis: Understanding multidimensional uses the forced choice as a dependent variable. The relative choices via stated preference experiments. Political Analysis effect of the variables is slightly different when respondents 22(1): 1–30. were forced to choose one of the two candidates. Healy A and Malhotra N (2013) Retrospective voting reconsid- ered. Annual Review of Political Science 16(1): 285–306. ORCID iD Klašnja M and Tucker JA (2013) The economy, corruption, Sofia Breitenstein https://orcid.org/0000-0002-5591-3320 and the vote: Evidence from experiments in Sweden and Moldova. Electoral Studies 32(3): 536–543. Klašnja M, Lupu N and Tucker JA (2017) When Do Voters References Sanction Corrupt Politicians? Unpublished manuscript. Avail- Anduiza E, Gallego A and Muñoz J (2013) Turning a blind eye: able at: http://www.noamlupu.com/corruption_sanction.pdf Experimental evidence of partisan bias in attitudes toward cor- Konstantinidis I and Xezonakis G (2013) Sources of tolerance ruption. Comparative Political Studies 46(12): 1664–1692. towards corrupted politicians in Greece: The role of trade offs Ares M and Hernández E (2017) The corrosive effect of corrup- and individual benefits. Crime, Law and Social Change 60(5): tion on trust in politicians: Evidence from a natural experi- 549–563. ment. Research & Politics 4(2): 205316801771418. Lewis-Beck MS and Stegmaier M (2007) Economic models Barnes TD and Beaulieu E (2014) Gender stereotypes and cor- of voting. In: Russell JD and Klingemann H-D (eds) The ruption: How candidates affect perceptions of election fraud. Oxford Handbook of Political Behavior. Oxford University Politics & Gender 10(3): 365–391. Press. Beaulieu E (2014) From voter ID to party ID: How political par- Mauro P (1995) Corruption and growth. The Quarterly Journal of ties affect perceptions of election fraud in the U.S. Electoral Economics 110(3): 681–712. Studies 35: 24–32. Muñoz J, Anduiza E and Gallego A (2016) Why do voters forgive Chang ECC, Golden MA and Hill SJ (2010) Legislative mal- corrupt mayors? Implicit exchange, credibility of informa- feasance and political accountability. World Politics 62(2): tion and clean alternatives. Local Government Studies 42(4): 177–220. 598–615. 8 Research and Politics  Peters JG and Welch S (1980) The effects of charges of corruption Shleifer A and Vishny RW (1993) Corruption. The Quarterly on voting behavior in congressional elections. The American Journal of Economics 108(3): 599–617. Political Science Review 74(3): 697–708. Sniderman PM (2018) Some advances in the design of survey Reed SR (1999) Punishing corruption: The response of the experiments. Annual Review of Political Science 21(1): Japanese electorate to scandals. In: Feldman O (ed.) Political 259–275. Psychology in Japan: Behind the Nails Which Sometimes The online data analysis tool and Afrobarometer (n.d.) Available Stick out (and Get Hammered Down). Commack, NY: at: http://afrobarometer.org/online-data-analysis/analyse- Nova Science Publishers, pp. 131–148. (accessed 12 February 2019). Rivero G and Fernández-Vázquez P (2010) Las consecuencias Visconti G and Mares I (2018) Voting for the Lesser Evil: Evidence electorales de los escándalos de corrupción municipal, 2003- from a Conjoint Experiment in Romania. Unpublished 2007. Documento de trabajo, Estudios de Progreso. 59(201). manuscript. Available at: http://giancarlovisconti.com/pdfs/ Available at: https://www.fundacionalternativas.org/public/ Mares%20and%20Visconti.2017.Abstract_1.pdf storage/estudios_documentos_archivos/a4112d386ee4ebfa- Winters MS and Weitz-Shapiro R (2013) Lacking information 92dae911711f0de3.pdf or condoning corruption: When will voters support corrupt Rothstein B and Varraich A (2017) Making Sense of Corruption. politicians? Journal of Comparative Politics 45(4): 418–436. Cambridge: Cambridge University Press. World Values Survey Database (n.d.) Available at: http://www. Rundquist BS, Strom GS and Peters JG (1977) Corrupt politicians worldvaluessurvey.org/WVSOnline.jsp (accessed 12 February and their electoral support: Some experimental observations. 2019). The American Political Science Review 71(3): 954–963. Zechmeister EJ and Zizumbo-Colunga D (2013) The varying politi- Schmitter PC and Lynn Karl T (1991) What democracy is... and is cal toll of concerns about corruption in good versus bad eco- not. Journal of Democracy 2(3): 75–88. nomic times. Comparative Political Studies 46(10): 1190–1218.