Condoning postwar corruption: how legacies of war prevent democratic accountability in contemporary Southeast Europe

ABSTRACT Do voters in postwar societies punish corrupt politicians? Or are their electoral preferences distorted by their own or the candidates’ war pasts? We answer these questions by analysing the results of an experiment embedded in a survey of over seven thousand respondents from the countries of Southeast Europe that experienced armed conflict since the 1990s. Our findings show that voters in this region punish corruption harshly, yet are more likely to ignore it for politicians with a military service record. This tendency is, however, conditioned by voters’ partisanship and sense of war grievance.


Introduction
The detrimental effects of corruption have long been recognised by scholars and policy makers. Corruption raises effective marginal tax rates and pushes businesses toward tax evasion (Alm, Martinez-Vazquez, and McClellan 2016), as well as workers toward lower productivity (Johnson, LaFountain, and Yamarik 2011). It raises the marginal costs of public projects, making many of them unviable (Olken and Pande 2012). These negative effects of corruption are especially pronounced in post-conflict societies transitioning to stable democracies where corruptionif left uncheckedcan lead into vicious cycles of instability and low popular legitimacy, which political leaders counter with ever more corrupt practices to maintain power and a semblance of order (Rose-Ackerman 2008). Corrupt postwar politicians often do that in collaboration with networks of powerholders who owe their social status to the roles they played in the conduct of the conflicts. This perpetuates the dynamics of the violence that ostensibly ended and cripples the prospects of postwar societies for successful reconstruction and recovery (Rose-Ackerman 2008).
Considering the nature of the problem presented by corruption to societies that have experienced conflict, some may see part of the solution to be in the speedy postwar creation of a functioning democracy. If the politicians are corrupt and collaborating with a relatively narrow network of beneficiaries, then the votersif given the proper opportunitywill surely throw the rascals out at the ballot box. Unfortunately, a solid body of evidence from democracies as diverse as Brazil (Ferraz and Finan 2008), Italy (Chang and Golden 2004), Japan (Nyblade and Reed 2008), and the United States (Welch and Hibbing 1997) shows that the magnitude of the voters' punishment of corrupt politicians is at best limited.
While both the condoning of corrupt behaviour among elected officials and the electoral legacies of war have spawned growing bodies of research, the intersection between both has yet to be explored. Previous research has found that postwar societies often experience "an intensification and entrenchment of the corruption-related problems" (Lindberg and Orjuela 2014, 278; also see Andreas 2004), though the mechanisms behind this phenomenon are still poorly understood. This study thus aims to contribute to solving the puzzle of why postwar societies, even decades after the conflict's end, continue to lag behind comparable countries who did not experience war in their recent history.
As such, this article contributes to a growing body of research on the role of war memory in postwar societies (Andrews, Bagot-Jewitt, and Hunt 2011;Edkins 2003;Hashimoto 2015;Wolfrum 1999;Zubrzycki and Woźny 2020). This literature emphasises the importance of remembrance, and in particular that of a violent past, in shaping collective identities and the "imagined" national community (Anderson 2016). At the same time, this article goes beyond this idea by more explicitly exploring how war memories work to the detriment of postwar societies. Specifically, we examine a pivotal pathway from war past to contemporary corruption: how war experiences and memories of recent conflict lead to a breakdown of democratic accountability and prevent voters from throwing out corrupt officials. Specifically, how does their own and the political candidates' war past affect voters' electoral choices? Does it lead to possible cognitive or emotional biases that may help candidates with a history of active participation in conflict get away with corruption? In addition, this article seeks to explore how being a war veteran helps accused candidates. Is being a war veteran an inherently valued candidate characteristic in postwar societies? Or is it merely a proxy for political views?
These questions are important, as they provide key insights into the reasons why so many post-conflict societies struggle to move forward on the path to democratic governance. We provide answers to these questions by drawing theoretical lessons from the literatures on voter choice, corruption, and war veteran politicians. We apply those theoretical lessons to the analysis of an experiment embedded in a survey of more than seven thousand adults of all age groups from six countries of Southeast Europe, which experienced armed conflict during the past two and a half decades: Bosnia and Herzegovina, Croatia, Kosovo, Montenegro, North Macedonia, and Serbia. This region, with a recent and still politically relevant history of conflict, with high levels of political corruption, but also with a functioning system of relatively free and fair elections, offers the perfect set of cases for our study.
Our findings show that voters in Southeast Europe do generally punish corrupt politicians, and they do that rather severely. They also handily reward electoral candidates who are war veterans. More importantly, the electoral punishment of corrupt candidates is smaller when they are war veterans. This relationship between voter choice, candidates' veteran status, and their corrupt behaviour is, however, conditioned by voters' partisanship and sense of war grievance. It is the rightwing respondents and respondents who still resent other national groups over what they did during the 1990s wars who are willing to ignore potential corruption among war veteran candidates. While the condoning behaviour among rightwing voters is driven by the policy expectations of these candidates once they are in power, voters with strong war grievances have an appreciation of war veteran candidates, regardless of political views. Respondents' actual war experiences, however, do not make a difference. In other words, the mechanism of accountability does not fail among voters who experienced war first-hand and were traumatised by it. It rather fails among voters for whom the ended conflict is still politically relevant.

Corruption, war veterans, and voter preferences
We define political corruption as the diversion of public resources and the use of public office by politicians for their own private ends. How voters react to corruption is a question that has occupied scholars for decades (De Vries and Solaz 2017). The seemingly obvious and straightforward answer has been that corrupt behaviour is punished by voters when they are granted the opportunity to do so (Banerjee et al. 2014;Ferraz and Finan 2008). Studies have shown that voters care about corrupt behaviour and lose trust in institutions because of it (Chang and Chu 2006;Seligson 2002). Electoral consequences do, however, require the right circumstances. These include credible alternatives to corrupt incumbents (Muñoz, Anduiza, and Gallego 2016), a not too fragmented party system (Schleiter and Voznaya 2014), a credible source behind the accusations (Botero et al. 2015), and media outlets disseminating information about corrupt behaviour (Ferraz and Finan 2008). Voters appear to punish corrupt politicians when they overcome the problem of informational asymmetry and when there is someone to replace the incumbent.
Yet even in cases where these conditions have been met, the electoral punishment has often been smaller than expected (Chang and Golden 2004;Ferraz and Finan 2008;Nyblade and Reed 2008;Welch and Hibbing 1997). In light of this, another strand in the performance voting literature has focused on why voters would knowingly vote for corrupt officials even when offered alternatives. Rundquist, Strom, and Peters (1977) have argued that voters were willing to condone corruption if the accused candidate had a strong record on other dimensions. Specifically, corruption could be overlooked when a politician had similar ideological views or was able to provide valuable goods (Winters and Weitz-Shapiro 2013), though this relationship may cease to work on higher levels of corruption (Vuković 2020). Several studies have indeed found that corrupt candidates faced fewer electoral consequences among co-partisans either because the latter do not believe the accusations or choose to ignore them (Eggers 2014;Franchino and Zucchini 2015), a direct extrapolation of the literature on the partisan bias in political perceptions (Bartels 2002). Additionally, candidates accused of corruption could alleviate some of the electoral consequences by lowering taxes (Konstantinidis and Xezonakis 2013), bringing jobs to the area (Breitenstein 2019), or presiding over an economically prosperous time in general (Choi and Woo 2010;Klašnja and Tucker 2013). However, the mitigating effects of co-partisanship or a thriving economy do not negate the fact that voters simply do not like politicians stealing their tax money (Costas-Pérez, Solé-Ollé, and Sorribas-Navarro 2012). In studies on what voters' ideal candidates look like, integrity ranks consistently and across voters among the highest valued traits (Trent et al. 1997(Trent et al. , 2001. While the electoral consequences of corruption might be lower than expected, we maintain that, overall, the support for political candidates will decrease when they are accused of corruption. Our first hypothesis, therefore, is: H1: Corruption accusations decrease the likelihood of voting for a political candidate. Political corruption is especially important in postwar societies. The difference in corruption levels between countries plagued by conflict and those in peace is truly staggering. Perhaps unsurprisingly, the bottom four places in the global rankings based on Transparency International's Corruption Perceptions Index (CPI) are occupied by four countries embroiled in war: Yemen, South Sudan, Syria, and Somalia (Transparency International 2020). Establishing a system of accountability between voters and politicians is of essential importance for the proper functioning of any postwar democracy. The question is, however, whether politicians' or voters' recent war pasts can impede the successful functioning of such a system. Elections in postwar societies are often characterised by the emergence of veteran candidates who use their military service to boost their electoral chances (Manning and Smith 2016). While the electoral benefits of military experience are commonly assumed, the empirical evidence from postwar societies is nearly non-existent and from Western societies engaged in overseas conflicts it is rather mixed (Karsten 2012;Teigen 2018). This is not to say that information on candidates' veteran status is ignored by voters. In surveys, respondents consistently associate military leaders with traits valued by voters in political candidates such as integrity, leadership, and competence (McDermott and Panagopoulos 2015).
The principal goal of this study is to explore the interplay between political candidates' veteran status and corruption because many post-war societies have extensive experiences with both, and because it brings us one step closer to understanding the challenges these societies face. Moreover, the connection between war veterans and corruption, at least in Southeast Europe, is not purely theoretical, but corresponds to real lived experiences of the way politics and economics in the region are conducted. From the smuggling networks and the various warlords running local economiesmost often with open cooperation by the politiciansduring the conflicts (Andreas 2004) to the various military leaders evolving into postwar entrepreneurs dependent on state procurement and contracts for profit, wars have been closely intertwined with politics and corruption in the region since the 1990s.
When voters are confronted with a war veteran accused of corruption, they are presented with two pieces of information that produce inconsistent evaluations of that candidate. On the one hand, veteran status signals a number of positive character traits. As mentioned above, being a war veteran is generally associated by desirable traits in a candidate (McDermott and Panagopoulos 2015). On the other hand, corruption is disliked by voters across the board (Ferraz and Finan 2008). This puts voters in a position where they hold two relatively important, yet incompatible cognitions about a candidate. This undermines their ability to act without internal conflict (Harmon-Jones and Harmon-Jones 2007), generating psychological stress often referred to as cognitive dissonance (Festinger 1957). Attempts to reduce this dissonance can involve favouring one cognition over the other, either by ignoring or denying either the corruption accusations or the candidate's military service, or adding more cognitions that tip the balance in favour of one cognition while still resulting in a more nuanced image of the candidate overall (Fotuhi et al. 2013;Harmon-Jones 2002). Alternatively, however, voters might leave the situation altogether and not make the choice between either supporting a war veteran accused of corruption or voting for the opposition. While the dynamics of the tug-of-war between the corruption and war veteran cognition will be further explored below, it is reasonable to expect that identifying a candidate accused of corruption as a war veteran will have the general effect of reducing the electoral impact of the corruption accusations. Therefore we expect that: H2: Veteran status alleviates the negative consequences of corruption charges.
The degree to which veteran status softens the impact of corruption depends, however, on the importance attached to being a war veteran among voters. Indeed, Festinger (1957) argued that attitude change is expected to be in the direction of the most enduring cognition. In other words, people reestablish consistency in their cognitive system by changing whatever is easiest to change. Thus, the condoning effect of being a war veteran on corruption accusations will be most effective on voters for whom the positive image of veterans is firmly engrained and supported by additional cognitions. The most likely of these additional cognitions is copartisanship. Veteran status cues certain ideological views that, even in the absence of a specific party label, attract voters who share them (Rueda 2005). While the causes of the conflicts accompanying and following the protracted dissolution of Yugoslavia were complex, it is safe to say that nationalism was at their core. The direct involvement of veteran candidates in these conflictsparticularly as the fighting was in many instances conducted by volunteer unitsinevitably fuels an association with nationalism, as people are expected to support the cause they put themselves in harm's way for. Because nationalist views are predominantly found on the right side of the ideological spectrum in South-East Europe, veteran status is likely to be a stand-in for rightwing views and rightwing party membership. This association is reinforced by the presence of veteran candidates on the electoral lists of nationalist parties. According to our calculations, for instance, since 2000 around 38% of candidates of Croatia's main rightwing and nationalist party Croatian Democratic Union (HDZ) over six electoral cycles were veterans. For the main non-nationalist party, the Social Democrats (SDP), on the other hand, only 22% of the candidates were veterans. Consequently, rightwing voters are expected to see a partisan ally in a veteran candidate and are consequently more willing to condone corruption accusations than leftwing voters are. We therefore hypothesise that: H3a: Veteran status alleviates negative consequences of corruption charges disproportionally among rightwing partisan voters.
At the same time, we argue that viewing veteran status solely as a stand in for partisanship would be reductionist, and would discount the social status enjoyed by war veterans among certain voters. In postwar societies, veterans are arguably held in higher regard by those for whom the war is politically and socially relevant. The Yugoslav wars of the 1990s were the bloodiest conflicts in Europe since World War II, with over a hundred thousand casualties and a far greater number of people displaced from their homes (Ringdal and Simkus 2012). The wars saw many mass murders and war crimes, which led to the creation of the International Criminal Tribunal for Yugoslavia (ICTY). Despite efforts to reconcile communities and former combatants through international justice, the postwar era has seen the evolution of diverging narratives and memories of the conflicts. Many memorials depict nationally exclusive heroism and victimhood, and commissions set up to investigate war crimes focus almost exclusively on in-group victims (also see Golčevski, von Engelhardt, and Boomgaarden 2013). This is also evident from post-war school education and textbooks: "teaching is divided, interpretations are ethnocentric, the superiority of one group and the inferiority of the others is central, as is the victimisation of 'our' group" (Torsti 2007, 90). We expect these narratives to give rise to a pervasive grievance regarding what other nationalities did during the war(s), and to be present on both sides of the ideological partisan spectrum. In turn, these grievances are projected onto war veteran candidates. Having been on the frontlines and having witnessed the actions of other nationalities first-hand, it is likely these candidates are anticipated to share voters' war grievances and, if elected, to pursue justice for the crimes committed. Consequently, for voters for whom the war is still politically relevant, the positive image of veterans is expected to be more resistant to change, resulting in a higher probability to vote for a veteran candidate despite corruption accusations. We thus hypothesise: H3b: Veteran status alleviates negative consequences of corruption charges disproportionally among voters with high levels of war grievance.
Socially, the impact of war can be felt long after the guns have gone silent through the medical and social care required for direct war participants, war victims, and their communities (Ghobarah, Huth, and Russett 2003). Long-lasting government commitment is needed to repair or alleviate the damage caused by war. This includes assistance for sustained physical injuries and treatment of the mental stress induced by war experiences (Priebe 2010;Ringdal, Ringdal, and Strabac 2012;Solomon and Lavi 2005), as well as funds to rebuild the crippled economic infrastructure. This commitment often does not come cheap and is one of the reasons why scholars have argued that war experiences give rise to a new political cleavage and structure voting behaviour. In Croatia, municipalities affected heavily by the war have been found to be more fiscally liberal and prone to vote for nationalist parties Vuković 2016, 2018). In the same country, individual-level data revealed a preference for nationalist parties among war veterans (Lesschaeve 2019). In Bosnia and Herzegovina, scholars have seen a solidification of ethnic identities in municipalities that have experienced many war casualties (Hadzic, Carlson, and Tavits 2020). In Serbia, refugees who had fled to the country from other former Yugoslav republics showed greater support for the nationalist Serbian Radical Party (Konitzer and Grujić 2009). While partisanship and war grievance could be confounding factors in the relation between war experiences and voting behaviour, as we also suggested above, we want to test whether an additional mechanism is at workone revolving around the material and social needs of direct war participants and war victims. We therefore expect voters who experienced war first-hand or who live in a war-affected community to see in a war veteran candidate someone who will look after their interests. Such a candidate could be viewed as someone who understands what the war did to them, someone who knows their plight. When it comes to policymaking, war participants and victims are expected to believe that a war veteran will make their voices heard. To these voters, the veteran cognition is expected to be the more resistant one, resulting in an increased likelihood to condone corruption among veteran candidates. We therefore hypothesise: H3c: Veteran status alleviates negative consequences of corruption charges disproportionally among voters who personally experienced the war or who come from communities with higher levels of exposure to war violence.

Data and method
This study relies on a 2018 online survey of citizens of Bosnia and Herzegovina, Croatia, Kosovo, Montenegro, North Macedonia and Serbia who were recruited via the social networking platform Facebook. To gather the data, we quota-sampled voters in each country, using Facebook's Marketing API. This approach uses Facebook advertisements to recruit respondents. In addition to being cost-effective, this method has proven that it can lead to similar results as national probability samples (Zhang et al. 2020). The main advantage of marketing API is that it gives researchers control over how they recruit subjects and enables the targeting of specific subpopulations. This allows for the fine-tuning of ads that target specific demographic groups and greatly assists in the quota sampling of a population. For the latter, a large number of strata were identified in each country according to several demographic characteristics (gender, age group, level of education, and region/county). 1 Data supplied by www.internetworldstats.com reveals that the internet penetration is larger than 70% in all countries, while Facebook penetration hovers around 40-50%. While these numbers are unquestionably lower than in most West European countries, the gap is not prohibitively large. In sum, with a panel that essentially covers half the population, using Facebook to gather survey data is a more than a viable method. The survey ran from October to December 2018. In the end, 15,501 respondents were recruited: 3871 in Bosnia and Herzegovina, 3074 in Croatia, 1820 in Kosovo, 2277 in North Macedonia, 1164 in Montenegro, and 3295 in Serbia. Of those 15,501 respondents, 7647 chose one of the two candidates and were thus included in the analysis.
While survey weights would be required to make the sample truly comparable to a representative survey (Ansolabehere and Rivers 2013), there is a great deal of uncertainty regarding their necessity or even appropriateness in the context of survey experiments. The choice essentially revolves around whether the goal is to measure a sample average treatment effect (SATE) or a population average treatment effect (PATE). The choice for PATE (and thus the use of survey weights) is not straightforward. Miratrix et al. (2018) show that weights substantially decrease the accuracy of the treatment effect estimation. They therefore recommend comparing the two measures. We found that the use or exclusion of weights does not substantially change the results. In that case, the SATE-estimate is a sufficient estimate of the PATE-effect, and preferable overall. Therefore, in the Figures below, we report the unweighted results and the weighted models can be found in the Online Appendix in Table A3. 2 Following the breakup of Yugoslavia, ethnic violence engulfed the region and all six countries were heavily affected by war. While some countries were hit more than others, all had to endure violence. As noted above, in postwar elections, veteran candidates have become a common sight on many party lists. Furthermore, countries of this region currently experience high levels of political corruption when compared to Western Europe, placing between rank 60 and 93 on Transparency International's CPI list and being labelled as mostly corrupt (Transparency International 2019). They nevertheless have democratic elections (Michael 2018). Although not perfect, elections in the region are reasonably free and fair, with no parties being banned and with different parties offering different policy platforms. This leads us to conclude that the countries under study are ideal cases to examine the interplay between war past, political corruption, and vote choice.
In the literatures on the impact of veteran status and the impact of corruption accusations, studies have gradually moved away from observational data (e.g. Chang and Golden 2004;Krause and Méndez 2009;Teigen 2008) to experimental designs (Botero et al. 2015;Breitenstein 2019;McDermott and Panagopoulos 2015). Most often, these experiments present respondents with a choice between usually two candidates, whose descriptions vary depending on the condition respondents were assigned to. Our design follows this standard. Respondents were assigned at random to one of ten conditions. Each condition presented them with a short election scenario in which two candidates, given generic names common in each country (as well as common for the ethnicity of the respondent in the cases of Bosnia and Herzegovina and North Macedonia), are running for the national parliament. The two candidates were shown side by side, and the order (left/right) of the candidates was randomised for each respondent. In the first condition, which served as the control, neither of the two candidates was identified as a veteran or as a candidate accused of corruption, nor was information given about their political views. The two candidates were comparable in terms of age (middle-aged), gender (male), ethnicity, having a family, and occupation, resulting in a 48-52% split in the vote preferences for the first and second candidate, respectively. In the second condition, respondents were informed that the first candidate had been accused of corruption by several newspapers. An alternative would have been to have the candidate be officially charged or convicted of corruption. However, in the region with deeply flawed legal systems which are themselves highly corrupt, and where convictions for corruption are rare, we felt that media accusations of corruption were a more appropriate experimental condition. In the third condition, the description of the first candidate mentioned that he was a war veteran. In the fourth condition, the first candidate was identified as a war veteran who was accused of corruption. In the fifth and sixth conditions, the first candidate was identified as having left-and rightwing views, respectively, while the other candidate was identified as having opposite views. Conditions seven and eight were the same as five and six, with the difference that the first candidate had been accused of corruption. Finally, conditions nine and ten were the same as seven and eight, but with corruption accusations swapped out for veteran status. Table 1 gives an overview of all conditions. The description of the first candidate did not differ between conditions. Short biographies of two candidates were also provided in every condition, which in the Croatian version read as 3 : (Ivan Perić) is 48 years old and was born and brought up in your area. After getting a degree in chemistry at university, (Ivan) trained as an accountant, and set up a company ten years ago. In (2016), he ran for office and was elected to the (Croatian Parliament). He is married and has three children. [He believes our country needs to run a more independent and sovereign foreign policy. If reelected, he will work on protecting our nation's traditional family values.] Survey experiments have been found to be a valid approach when studying the impact of candidate characteristics on vote choice, including corruption accusations and war veteran status (Hainmueller, Hangartner, and Yamamoto 2015). However, the main disadvantage of vignette experiments is their external validity. Boas, Daniel Hidalgo, and Melo (2019), for example, found that Brazilian voters were very willing to punish corrupt incumbents, but that this did not translate in electoral action in mayoral elections. The discrepancy between these two results is attributed to the greater saliency of corruption in experiments than in real elections. Nevertheless, vignette experiments remain valuable as instruments to gain insights into voters' social norms and the go-to option for estimating the effect of moderating factors such as veteran status on the electoral consequences of corruption. In sum, the experimental design used in this study is the most appropriate way to test our hypotheses (see, for instance, Batista Pereira 2020). While the estimated effects are most likely an overestimation of the real-life effects, this does not invalidate them.
After reading the candidates' biographies, voters were asked whom they were most likely to vote for. In addition to the two candidates, respondents could also abstain from voting. This answer arguably captures the extent to which the information given about the candidates makes people withdraw from the electoral process. Studies have found evidence that corruption accusations decrease turnout (Stockemer, LaMontagne, and Scruggs 2013). As it deals with an entirely different research question, choosing to vote for neither candidate is beyond the scope of this article. In order to maintain conceptual clarity between the various categories of the dependent variable, we do not pool together respondents who did not want to vote for either candidate and those who expressed a preference for the second candidate. Instead, the former are excluded from analysis. 4 The principal dependent variable is thus a binary variable indicating whether the respondent expressed a preference for the first candidate (value 1), or for the second candidate (value 0). Therefore, when analysing the data, we rely on a binomial model.
The first main explanatory variable is a categorical variable denoting the condition to which the respondent was assigned and thus which version of the first candidate they were shown: control, Corrupt, Veteran, Corrupt & Veteran, Leftwing, Rightwing, Corrupt While this scale is commonly presumed to summarise people's policy views, the support behind this assumption has been lacking. Inglehart and Klingemann (1976) found that of the three components of the left-right scalepartisan loyalties, group memberships, and policy positionsthe first proved to be the most prominent. This has been confirmed by more recent studies showing only a weak relation between people's policy positions and their left-right placement (Lesschaeve 2017;Todosijević 2004). As such, and in spite of how it is predominantly used in the literature, we view the left-right self-placement scale as indicative of people's partisan loyalties. For war grievance, we use respondents' agreement with a single Likert-scale type statement: "I can never forgive the members of the other nationalities in this country for what they have done during the war." A correlation of 0.19 shows that both variables are only weakly related.
When it comes to possible war victimhood stemming from war experiences, however, things get a bit more complex. The literature lacks a standard way of operationalising this concept (Glaurdić, Lesschaeve, and Mochtak 2019). Therefore, for reasons of parsimony, we measure respondents' personal exposure to war violence as a sum scale of six binary indicators. To this end, we asked the respondents whether: (1) they were war veterans, (2) they were ever in physical danger or (3) whether they were forced to migrate during the war, (4) they lost someone in their close family or among friends during the war, (5) they are suffering from a war-induced physical disability, and finally (6) they show any symptoms of war trauma. To qualify as war veterans, respondents had to have actively fought in the wars relevant for their countries. For war trauma, we relied on respondents' answers to six yes-no questions measuring symptoms of war-related trauma (K. Ringdal and Simkus 2012). 5 For contextual measures of war experiences, we unfortunately had to limit our analysis to Croatia and Bosnia and Herzegovina due to municipality-level data availability. Despite this limitation, with data for over 400 municipalities in these two countries which were anyway most affected by the wars in this region, we believe the sample size is sufficient to examine the impact of contextual war experiences. On the municipal level, we measure the incidence of disability caused by war violence. This variable was derived from the 2011 Croatian census and the 2013 census in Bosnia and Herzegovina. We believe it is the best contextual aggregate-level measure of the local population's overall exposure to war violence.
In the analyses, we control for respondents' age, gender, income (deciles), minority status, 6 education level, respondents' interest in politics, and whether respondents thought that corruption was one the most important problems facing their country today. In the contextual analyses, we control for municipalities' ethnic diversity through the ethnic fractionalisation index, unemployment rate, and the average monthly per capita income (expressed in 2017 purchasing power parity US$). The rationale for including control variables is to account for (small) sample differences between the conditions. While the conditions are overall quite similar in terms of our independent variables, a multinomial regression of respondents' condition on the explanatory variables did find some significant differences. Any spurious effects because of these differences were removed with the addition of the control variables.
In our analyses, we pool the data from all six countries. We account for the differences on this macro-level by including country dummies in all models. In addition, to test the robustness of our models, we applied a jackknife procedure, repeating the models reported below, each time excluding one country. In every iteration, the results were substantively the same.

Results
The results of the multivariate analyses are reported in Model 1, Table A1 in the Online Appendix, but the effects are graphically represented in Figure 1. Specifically, Figure 1 shows the predicted probabilities of voting for the first candidate in each of the nine experimental conditions, after having subtracted from it the predicted probability of voting for that candidate in the baseline control condition (46%). Consequently, the dots in the figure indicate the change in vote share in each of the nine experimental condition compared to the control condition, which is represented by the zero-line. The further a dot is removed from the zero-line, the greater the increase (to the right of the line) or decrease (to the left of the line) in vote share compared to the control condition. Figure 1 shows a highly significant negative difference between the control condition and the corrupt condition. As mentioned earlier, in the control condition, the first candidate received roughly half of the votes. When he is accused of corruption, his vote share drops by 31 points, which is a massive decrease. This shows that there are large electoral consequences to being accused of corruption when respondents are offered an otherwise comparable and uncorrupt alternative. This finding is in line with the literature showing the negative effects of corruption allegations on electoral prospects. This clearly supports our first hypothesis. Figure 1 also shows that veterans enjoy a distinct electoral advantage. Squared off against a non-veteran opponent, being a war veteran can increase a candidate's vote share by 14 percentage points. In the successor states of the former Yugoslavia, there does exist a substantial veteran electoral bonus. Hypothesis 2 predicts that being a war veteran softens the negative electoral consequences of corruption accusations. When we look at the corrupt veteran condition and compare it to the corrupt and the veteran conditions, we see that there is still a sizable drop in support for the first candidate, though it is smaller than the drop for the non-veteran corrupt candidate. This provides evidence in support of our hypothesis 2. Veteran status does mitigate some of the negative effects of corruption allegations, though by no means all. When a candidate accused of corruption is also a war veteran, the 31% drop in vote share is reduced to a 25% drop. 7 In Hypothesis 3a, we predict that the interplay between past war service and corruption accusations is dependent on partisanship. To test this, we interact experimental condition with partisanship. The results are reported in Model 2, Table A1 in the Online Appendix, but the effects are visualised in Figure 2. This figure shows the same probabilities as Figure 1, but splits the results up by right-and leftwing respondents (left side of Figure 2), and by respondents with high and low levels of war grievance (right side of Figure 2). Among leftwing respondents, the corrupt candidate and the corrupt candidate with a war service past are virtually indistinguishable. 8 In contrast, veteran status softens the negative impact of corruption accusations on the vote share from 29 points to 19 points among rightwing respondents. The most likely explanation behind this interaction effect is that veteran status is a proxy for party identification. Veteran status is presumed to signal rightwing views to voters. This is confirmed when we look at the conditions where information about candidates' policy views is supplied. The extent to which leftand rightwing voters reward veteran candidates, is similar to the extent that they reward left-and rightwing candidates. Though for rightwing voters the gap between left-and rightwing candidates is substantially reduced when both are war veterans, the latter is nevertheless preferred over the former. This supports hypothesis 3a. The interaction effects between war grievance and experimental condition (coefficients are reported in Model 3, Table A1 in the Online Appendix), appear to follow a similar pattern as partisanship. Voters with strong war grievances are more likely to vote for a veteran candidate, and overlook corruption accusation in veteran candidates than voters with low levels of war grievances are.
In addition, we also find that, for those who still harbour resentment toward other groups for what they did during the war, veteran status is more than just a signal of policy views. In the ninth and tenth conditions, the first candidate in the vignette experiment is a war veteran and identified as holding left-and rightwing views, respectively. The likelihood of someone with strong grievances of voting for the first candidate, however, does not differ between the two conditions. Whether a candidate with a war service record has left-or rightwing views is inconsequential. For people with strong war grievances, voting for a war veteran is more important than voting for a right-or leftwing candidate. Interestingly, in condition seven and eight, in which the first candidate is accused of corruption and identified as holding left-and rightwing views, respectively, rightwing voters and voters with strong grievances punish corruption equally, regardless of ideological views. Leftwing voters and those with little war grievances, however, show a clear willingness to ignore corruption among leftwing candidates, who continue to hold a 16 and 17% lead over corrupt rightwing candidates.
We find no moderating effects, however, of possible war victimhood caused by different war experiences, neither on the individual nor on the contextual level (see Table A2 in the Online Appendix for the models). Our analysis thus provides strong support for hypotheses 3a and 3b, while hypothesis 2c is firmly rejected. The veteran bonus is dependent on the voters' partisanship and war grievance, but not on their possible victimhood grounded in personal or contextual war experiences.
As a robustness check, we reran the models for each country separately. The full models can be found in the online appendix in Tables A4-A9, but the main findings are summarised in Table 2. This table shows the direction and statistical significance of the model coefficients most pivotal for the hypotheses. The first column reports the effects of all countries combined, reflecting the results shown in Figures 1 and 2. The subsequent columns show how the effects turned out in each country. A single sign symbol (+/−) indicates the direction of a substantially large (|β| > 0.10) yet statistically non-significant coefficient (p > 0.05). A double sign symbol indicates a statistically significant effect. We use this approach to overcome the difficulties of disaggregating a pooled analysis, as it results in a reduced overall sample size, and a different sample size in each country.
Overall, we find a large degree of overlap and agreement between the six settings. In all countries corruption is severely punished, veterans receive a bonus, and being a veteran helps alleviate some of the negative consequences of corruption accusations. In addition, in most countries, the left-right continuum, war grievance, or both moderate the electoral bonus of military experience, and the dampening effect of being a war veteran on the negative ramifications of corruption.

Conclusions
The detrimental effects of corruption on postwar societies cannot be overstated. Even in peaceful and more or less functioning democracies, corruption results in the misallocation of investments, tax evasion, lower productivity, higher inequality, policy inefficiency, and ultimately lower state and government legitimacy. In the post-conflict environment of destabilised social and political order, all these negative effects are only amplified. Corruption in postwar societies is easily grafted onto the power structures created by conflict, andif left uncheckedit perpetuates these structures and the divisions they create. Moreover, postwar societies marred by corruption run the risk of being stuck in a vicious cycle with criminal coalitions of corrupt officials and their associates resorting to ever more corrupt practices ostensibly to maintain order, but in reality to protect their own private interests. For example, Afghanistan's President Hamid Karzai frequently claimed he preferred peace to justice, while his administration catered not to the needs of the population, but to its own needs and the needs of its corrupt collaborators, with 35-50% of international aid money wasted (Rose-Ackerman 2008, 334). If political corruption in postwar societies is to be controlled, a system of electoral accountability needs to be created and it needs to function properly. For that to happen, however, voters' decision making needs to be free of bias, particularly bias which may be rooted in the potentially highly polarising recent war past. With this article, we have shown thatat least in Southeast Europethis unfortunately is not the case. While voters from this region generally punish corrupt politicians, they are disproportionally willing to turn a blind eye to corruption if it originates with politicians who are war veterans.
What is more important, this tendency is conditional on voters' ideological commitments and attitudes toward the war past, rather than their war experiences. Voters who see a partisan ally in candidates who are war veterans, and voters with high levels of war grievance directed against other national groups, are the principal drivers of the veteran electoral bonus that allows corrupt politicians with active war military service to get away with criminal behaviour. This means that the system of electoral accountability for corruption does not fail among voters for whom the ended conflict is no longer politically relevanteven if they experienced war first-hand and were traumatised by it. Instead, it fails among voters who are still emotionally stuck in the past, regardless of what they actually lived through during the war. Little wonder that in Southeast Europe it is often exactly the most corrupt politicians who publicly invoke recent history in order to legitimise their hold on power.
The results of this study shed light on one possible explanation as to why many countries in Southeast Europe lag behind other European nations with a similar communist past but who did not experience war in their recent pasts. While we are not the first to discover an enduring legacy of war in contemporary political competition (Rozenas, Schutte, and Zhukov 2017), to our knowledge, this study does represent an important piece of evidence that points to the dangers posed by the political potency of war memory. This should be explored further by future research. For instance, what other considerations in electoral choices (e.g. economic performance) are trumped by the legacies of war? And how does the precise nature of the war experiences (e.g. victory, defeat, or stalemate) affect their continued relevance in present-day political competition? Only by fully understanding the longterm effect of conflicts can we hope to aid postwar societies in leaving the war past behind and turning towards a brighter, peaceful, and prosperous future. Notes 1. We identified 238 strata in Bosnia and Herzegovina,294 in Croatia,28 in Kosovo,88 in North Macedonia,33 in Montenegro, and 400 in Serbia.
2. In addition, the distribution of the sample in terms of gender, age, and education can be inspected in Figures A1 and A2 in the Online Appendix. 3. The parts between round brackets were different for each of the six countries. The parts between square brackets were changed between conditions. 4. Robustness checks show that including non-voters in the analyses does not substantively change the results (results can be requested from the authors). 5. We checked whether war experiences correlate with partisanship and war grievance, but this proved not to be the case (r = −0.03 and 0.03, respectively). In addition, while partisanship and war grievance unsurprisingly correlate in the region (r = 0.18), this is not enough to raise concerns for the analyses. 6. In Bosnia and Herzegovina, we made a distinction between the Federation of BiH and Republika Srpska, designating Serbs a minority in the former and Croats and Bosniaks a minority in the latter. In North Macedonia, Albanians are considered a minority. In other countries, minority membership was based on respondents' self-reporting. 7. We acknowledge that the difference between the Veteran and the Corrupt & veteran condition (42%) is slightly larger than the difference between the Control and Corrupt condition (34%). However, hypothesis one does not predict that corruption accusations work differently for veterans and non-veterans, but rather that being a war veteran can neutralise part of the negative electoral consequences of corruption. As such, testing this assertion is done by comparing the Corrupt condition with the Corrupt & Veteran condition. 8. We take as high and low levels of partisanship, war grievance, and war experiences the mean value plus and minus one standard deviation.