Introduction
In representative democracies citizens hold the government accountable for its actions. By punishing poor and rewarding good performance, voters express their preferences and force political elites to perform in line with their preferences. Most empirical research on electoral accountability and the impact of retrospective performance evaluations focuses on the economy (Lewis-Beck and Stegmaier, Reference Lewis-Beck and Stegmaier2000; Anderson, Reference Anderson2007; Duch and Stevenson, Reference Duch and Stevenson2008; Singer and Carlin, Reference Singer and Carlin2013). However, as voters can assign responsibility to various government outcomes, there is no reason to believe that electoral accountability should be restricted to economic issues (Singer, Reference Singer2011a, Reference Singerb). In this article we focus on corruption – another dimension of the growing performance voting literature.
Corruption comprises activities such as bribery, embezzlement of public funds, misuse of public party funding, and clientelism. Its severe detrimental effects on the economy and society (Mauro, Reference Mauro1995; Rose-Ackerman, Reference Rose-Ackerman1999) should motivate citizens to make vote choices based on their perceived level of corruption. We employ a reward-and-punishment framework to study its electoral consequences. Corruption is a valence issue (Stokes, Reference Stokes1963, Reference Stokes1992) where voters evaluate parties based on their ability to achieve a generally desired policy goal. As a result, citizens should reward government parties for low levels of corruption and punish them for high levels. However, corruption is also a highly persistent phenomenon (e.g. Alesina and Weder, Reference Alesina and Weder2002), which suggests that this simple reward-and-punishment framework to ‘throw the rascals out’ is not the modus operandi in all contexts. Drawing on the rich literature on economic voting, we argue that corruption performance voting varies with individual-level factors and the institutional context. By providing insights into the specific conditions under which the chain of accountability breaks down (Vivyan et al., Reference Vivyan, Wagner and Tarlov2012), we may better understand why fighting corruption is a ubiquitous challenge, even in competitive electoral democracies.
The empirical analysis is based on individual-level data on corruption perceptions and voting behaviour in 20 elections across Western and Central Eastern Europe from Module 2 of the Comparative Study of Electoral Systems (CSES) survey. Past research on the effects of corruption on incumbents’ electoral performance is often based on macro-level data and addresses variation in corruption across countries (Ferraz and Finan, Reference Ferraz and Finan2008; Krause and Mendez, Reference Krause and Mendez2009; Choi and Woo, Reference Choi and Woo2010). Micro-level analyses, in turn, are often linked to specific scandals and certain national contexts (Peters and Welch, Reference Peters and Welch1980; Welch and Hibbing, Reference Welch and Hibbing1997; Vivyan et al., Reference Vivyan, Wagner and Tarlov2012; Basinger, Reference Basinger2013; Praino et al., Reference Praino, Stockemer and Moscardelli2013; Winters and Weitz-Shapiro, Reference Winters and Weitz-Shapiro2013; Wagner et al., Reference Wagner, Tarlov and Vivyan2014). In this article, we employ the CSES data to analyse corruption performance voting and its constraints under various micro- and macro-level factors. The results show some similarities between economic and corruption performance voting, but also point to important differences. As with economic issues, voters generally punish governments if they perceive high levels of corruption. Moreover, corruption perceptions influence vote choice if voters believe that bringing in a new government will make a difference. Non-partisans are more likely to make use of their corruption perception than party-loyal voters. We also find that corruption voting is stronger in national contexts where corruption is a severe, very salient policy issue. However, we find no evidence that corruption voting hinges on a political system’s ‘clarity of responsibility’, that is, the degree to which voters can assign responsibility to particular political actors (Powell and Whitten, Reference Powell and Whitten1993).
These findings suggest that while the state of the economy is a major benchmark that allows voters to evaluate incumbent parties, it is not the only yardstick they use to reward or punish incumbents. Democracies in Central Eastern but also Southern Europe often face high levels of corruption and patronage that may overshadow problems in the economic realm (see, e.g. Anderson and Tverdova, Reference Anderson and Tverdova2003; Kopecký et al., Reference Kopecký, Mair and Spirova2012). In these contexts, including corruption perception enriches the performance voting framework and our knowledge of voting behaviour. At the same time, clarity of responsibility as an alternative context factor seems to be less relevant for corruption performance voting than for the economy. As a result, the context-dependent nature of corruption performance voting allows us to better understand why corruption is highly persistent in some countries.
The economic prerogative in performance voting
We explore whether vote choice is influenced by perceived levels of corruption. Conceptually, this approach relates to the broad theoretical framework of performance voting models (Key, Reference Key1966; Kramer, Reference Kramer1971; Ferejohn, Reference Ferejohn1986). Performance voting is based on the simple idea that voters evaluate parties based on their achievements in office. It assumes that government performance enters a voter’s decision-making process retrospectively. In Kramer’s (Reference Kramer1971: 134) words: ‘if the performance of the incumbent party is “satisfactory” […], the voter votes to retain the incumbent governing party in office […]; if the incumbent’s performance is not “satisfactory”, the voter votes against the incumbent to give the opposition party a chance to govern’. Thus, performance voting is based on voter evaluations of the government parties’ past performance irrespective of how they evaluate the opposition parties’ potential performance. The implicit assumption is that voters first and foremost care about ‘throwing the rascals out’ if their performance was unsatisfactory.
So far, existing studies on performance voting have mainly focused on the economy (Key, Reference Key1966; Kramer, Reference Kramer1971; Fiorina, Reference Fiorina1981; Powell and Whitten, Reference Powell and Whitten1993; Whitten and Palmer, Reference Whitten and Palmer1999; Lewis-Beck and Stegmaier, Reference Lewis-Beck and Stegmaier2000; van der Brug et al., Reference van der Brug, van der Eijk and Franklin2007; Duch and Stevenson, Reference Duch and Stevenson2008; Becher and Donnelly, Reference Becher and Donnelly2013). There are various reasons for its preponderance in the performance voting literature. First, voters generally agree on desired economic policy goals (economic growth and low levels of unemployment and inflation). As the economy is therefore a valence issue, voters can send coherent signals to incumbents if they fail to act in line with the voters’ preferences. This agreement on policy ends effectively enables voters to efficiently structure the behaviour of the government (Ferejohn, Reference Ferejohn1986). Second, the focus on economic outcomes is justified by its importance to voters. While issue priorities vary substantially between voters and electoral contexts (Singer, Reference Singer2011a, Reference Singerb), economic problems, such as unemployment, are typically among voters’ most pressing concerns. Finally, economic outcomes are easy to observe for the average voter. Economic voting allows voters to make rational decisions even if they are politically inattentive (Fiorina, Reference Fiorina1981).
A thriving or deteriorating economy is, however, not the only criterion voters can use to evaluate the performance of government parties (Fiorina, Reference Fiorina1981; Stokes, Reference Stokes1992). As Stokes (Reference Stokes1992: 147) notes:
‘The classic illustration [for a valence issue] is good economic times and bad. But the parties and leaders are at times linked to peace and war, internal order and crime, and many other conditions that are positively or negatively valued. Although their bonding to economic conditions goes back to the early nineteenth century, the conditions that provide the symbolic content of valence politics have progressively expanded with the scope of government in the twentieth century.’
Thus, if voters care about monitoring officeholders, they might also condition their vote choice on other dimensions of good governance such as public safety or peace. Here, we focus on one such non-economic issue: the level of corruption. Corruption is commonly defined as ‘the misuse of public office for private gains’ (Treisman, Reference Treisman2000: 399) or as ‘an act by a public official (or with the acquiescence of a public official) that violates legal or social norms for private or particularistic gains’ (Gerring and Thacker, Reference Gerring and Thacker2004: 300). In both definitions public officials are the major actors, while the extent and type of corruption are left open. Corruption thus comprises activities such as bribery, embezzlement of public funds, misuse of public party funding, and clientelism. Gerring and Thacker’s definition also includes the toleration of corrupt practices. Hence, political elites may be held accountable not just for themselves abusing power and money, but also for failing to limit corrupt behaviour in general (Tavits, Reference Tavits2007).
Recent research highlights the role of corruption perceptions on political behaviour, in particular its effect on voter turnout (Peters and Welch, Reference Peters and Welch1980; Stockemer and Calca, Reference Stockemer and Calca2013; Stockemer et al., Reference Stockemer, LaMontagne and Scruggs2013). Some studies show that high levels of corruption decrease turnout (e.g. McCann and Dominguez, Reference McCann and Dominguez1998; Stockemer et al., Reference Stockemer, LaMontagne and Scruggs2013) as high levels of perceived corruption lead to more negative evaluations of political authorities and the political system in general (Anderson and Tverdova, Reference Anderson and Tverdova2003). Yet, others find a positive effect of corruption on turnout (Stockemer and Calca, Reference Stockemer and Calca2013). Kostadinova (Reference Kostadinova2009), in turn, argues that both effects may occur: there is a direct, mobilizing effect of corruption on turnout and an indirect, negative one that affects turnout through political efficacy. Here, we will extend this research on corruption and study its effect on incumbent voting.
We posit that voters evaluate government parties based on the perceived level of corruption for two reasons. First, corruption is a valence issue (Stokes, Reference Stokes1963, Reference Stokes1992) and thus fits the reward-and-punishment logic of the performance voting framework: voters should prefer less to more corruption and should evaluate the extent to which the incumbent government has performed effectively.Footnote 1 Second, corruption is particularly relevant in some Eastern and Southern European (e.g. Anderson and Tverdova, Reference Anderson and Tverdova2003; Kopecký et al., Reference Kopecký, Mair and Spirova2012) and Western European countries (Singer, Reference Singer2011b). For example, countries such as Romania or Bulgaria do poorly in Transparency International’s Corruption Perception Index (CPI), and voters in these countries also name corruption as one of the most important issues (Singer, Reference Singer2011b). Thus, in these contexts corruption should be a major element in voters’ reasoning when casting their ballots.
In fact, the reward-and-punishment mechanism may even be more persuasive for corruption levels than for the economy. Economic policies also entail aspects where policy means rather than goals are important for making decisions. For example, voters may not only value parties that are competent in ensuring economic growth or low unemployment rates, but also those that advocate policies (e.g. income tax, minimum wages) that are in line with their personal policy preferences. These positional elements of party competition are less relevant for voters’ perceptions of corruption. In contrast to economic policies, parties differ less in how they propose to fight corruption. Instead, corruption can almost exclusively be discussed in valence terms, making the reward-and-punishment logic even more appealing than for economic policies. Moreover, corruption is an issue where it is crucial that voters monitor the incumbents. This is due to the fact that political elites enjoy private gains when exploiting their power and positions for their own good. They thus have clear incentives to pursue their own goals, and these differ from those of voters. Finally, corruption is a particularly suitable issue for performance voting because corruption, as defined above, is directly attributable to the political elites. Thus, it should be relatively easy for the average voter to evaluate incumbent performance on corruption and to punish those who fail to deliver the desired outcome. This leads us to expect that corruption perceptions have a strong effect on vote choice.
The contingent effect of corruption performance voting
However, previous research has shown that performance voting is context-dependent (Powell and Whitten, Reference Powell and Whitten1993; Whitten and Palmer, Reference Whitten and Palmer1999; Hellwig Reference Hellwig2001, Reference Hellwig2011; Anderson, Reference Anderson2007; van der Brug et al., Reference van der Brug, van der Eijk and Franklin2007; Duch and Stevenson, Reference Duch and Stevenson2008, Reference Duch and Stevenson2010). Hence, we expect that the connection between voter perceptions of corruption and vote choice is also likely to be contingent on individual-level attributes as well as on macro-level context factors. We argue that some of these contingency-effects will differ from those for the economy as a performance voting issue. While both issues share many commonalities leading to similar effects and mechanisms, we also observe differences that should affect the circumstances under which economic and corruption performance voting are most likely.Footnote 2
A first moderating factor is partisanship: performance voting is more likely for non-partisans than for voters who lean towards a particular party (van der Brug et al., Reference van der Brug, van der Eijk and Franklin2007; Kayser and Wlezien, Reference Kayser and Wlezien2011). This argument goes beyond the widely known findings that partisanship affects voter perceptions of politics (Campbell et al., Reference Campbell, Converse, Miller and Stokes1960; Bartels, Reference Bartels2002; Anduiza et al., Reference Anduiza, Gallego and Muñoz2013) and vote choice (see e.g. Merrill and Grofman, Reference Merrill and Grofman1999). Rather, it states that partisanship moderates the impact of (potentially biased) perceptions of politics on vote choice (Vivyan et al., Reference Vivyan, Wagner and Tarlov2012). Voters with strong partisan attachment should be less responsive to short-term factors including performance evaluations because of their clear-cut party preferences. Although these voters might be dissatisfied with the performance of their most preferred party, their strong preferences for that party make party switching less likely (Kayser and Wlezien, Reference Kayser and Wlezien2011; Eggers et al., Reference Eggers, Vivyan and Wagner2014). In contrast, non-partisans lack these strong predispositions, and thus even small differences in their perceived party performance can have significant effects on their vote choice (van der Brug et al., Reference van der Brug, van der Eijk and Franklin2007). Following the literature on economic voting, we test whether non-partisan voters put greater weight on the government’s record in tackling corruption:
Hypothesis 1 The influence of corruption performance voting is higher for non-partisans than for voters who feel close to a party.
Retrospective voting should also depend on voters’ belief that throwing the rascals out will not bring new rascals in. Punishing government parties by voting for opposition parties presumes that voters believe in the effect of government turnover. This belief is captured in several theoretical concepts. It relates to voter perceptions of government responsiveness (Kölln et al., Reference Kölln, Esaiasson and Turper2014) and is reflected in what Downs (Reference Downs1957) has labelled the ‘expected party differential’, which is defined as ‘the difference between the utility income [a citizen] actually received […] and the one he would have received if the opposition had been in power’ (Downs, Reference Downs1957: 40). Understood in a broader sense, this notion of party differentials captures in fact any form of perceived party differences including not only policy, but also perceived differences in performance-based assessments and evaluations of party leaders (see Blais et al., Reference Blais, Young and Lapp2000; Wessels and Schmitt, Reference Wessels and Schmitt2008). If they do not believe that government turnover will make a difference, citizens are unlikely to punish incumbents by voting for an opposition party. The more likely option in this situation is to abstain (Downs, Reference Downs1957; see also McCann and Dominguez, Reference McCann and Dominguez1998; Kostadinova, Reference Kostadinova2009). Thus, we expect corruption performance voting to be stronger if voters believe that electing opposition parties will make a difference:
Hypothesis 2 The influence of corruption performance voting is higher if voters expect significant change when the government composition changes.
Turning to factors in the electoral context, we test whether the effect of corruption performance voting depends on the severity of corruption. One of the core elements of issue voting models is that voters evaluate parties based on their most important problems and concerns (Krosnick, Reference Krosnick1988, Reference Krosnick1990). Thus, economic performance voting is most likely in hard economic times and among those who are hit hardest by unemployment and recessions (Stevenson, Reference Stevenson2002; Singer, Reference Singer2011a, Reference Singerb).
Following this argument, we test whether voters pay more attention to their corruption perception when evaluating government performance in electoral contexts where corruption is a prominent problem. In these contexts, the misuse of public goods is more relevant to voters when evaluating government performance. Thus, it is here where voters can use their retrospective evaluation of corruption as an information shortcut (Fiorina, Reference Fiorina1981). In contrast, the effect of corruption perception might be weaker in national contexts where corruption is less severe. Here, voters are more likely to focus on other issues (e.g. the economy) to assess the performance of government parties. Therefore, corruption performance voting should be more likely in contexts where corruption is a prominent problem. This leads to:
Hypothesis 3 The influence of corruption performance voting is higher in electoral contexts where corruption is a prominent issue.
Another factor that might hinder performance voting is the clarity of political responsibility. In order to punish incumbent parties, voters need to be sure that bad performance was truly a result of the officeholders’ mismanagement – and not a result of exogenous factors or actions taken by other political actors. One of the major findings in the economic voting literature is that evaluating parties and candidates is much easier in systems where performance can clearly be attributed to individual actors (Powell and Whitten, Reference Powell and Whitten1993; Whitten and Palmer, Reference Whitten and Palmer1999; Anderson, Reference Anderson2007; Duch and Stevenson, Reference Duch and Stevenson2008). It is easier for voters to evaluate the responsible parties, for example, if the government composition is relatively stable and if the incumbent parties have a majority in parliament. Moreover, the consensus-seeking common in coalition governments makes it difficult for voters to identify the party or parties that are responsible for policy outputs, and this challenge increases with the increasing number of coalition partners.
We test whether the same moderating effect exists for corruption performance voting. Following the arguments from the economic voting literature, corruption performance voting should be more likely in electoral contexts where government parties have unified control over policy-making. Higher clarity of responsibility should make it easier for citizens to identify corruption and scandals, and a high level of information about corrupt practices is linked to corruption performance voting (Winters and Weitz-Shapiro, Reference Winters and Weitz-Shapiro2013). There are of course differences between the economy and corruption as performance voting issues. First, responsibility over economic policies mostly deals with legislative competences, while fighting corruption is largely a matter of executive power. When unable to pass their preferred economic legislation, minority governments or those facing a strong second chamber may diffuse responsibility and blame their political opponents for blocking their most desired policy proposals. In contrast, parties rarely disagree about how to fight corruption. Rather, it is primarily a matter of effective control over the public administration, enforcing existing laws and regulations, and thus lies at the very heart of the executive branch’s responsibility. Second, attributing blame for corruption, especially grand corruption, is much easier than for economic outcomes. Expenses and corruption scandals are often linked to particular parties and politicians, and voters can identify relevant players irrespective of the institutional and political clarity of responsibility. Thus, the signal-extraction problem that plagues voters on economic policies may be less severe for corruption performance voting. Although we expect a moderating effect of political responsibility on corruption voting, its effect might be weaker than previous research found for economic voting.
Hypothesis 4 The influence of corruption performance voting is higher in electoral contexts where the clarity of political responsibility is high.
Data and measurement
There are only a few cross-national studies on corruption performance voting using individual-level data. Most studies focus on particular events (e.g. an expenses scandal) to study the effects of bribery on voters’ opinions and vote choice (Peters and Welch, Reference Peters and Welch1980; Vivyan et al., Reference Vivyan, Wagner and Tarlov2012; Wagner et al., Reference Wagner, Tarlov and Vivyan2014). Moreover, many analyses use survey experiments (Anduiza et al., Reference Anduiza, Gallego and Muñoz2013; Winters and Weitz-Shapiro, Reference Winters and Weitz-Shapiro2013; Eggers et al., Reference Eggers, Vivyan and Wagner2014; Wagner et al., Reference Wagner, Tarlov and Vivyan2014) to provide insights on the causal mechanisms in decision-making processes. Yet, analyses based on a single national context do not allow us to study cross-national differences even though such differences are likely to exist (van der Brug et al., Reference van der Brug, van der Eijk and Franklin2007; Duch and Stevenson, Reference Duch and Stevenson2008). Comparative analyses are rare and largely limited to a macro-level perspective (Krause and Mendez, Reference Krause and Mendez2009; Choi and Woo, Reference Choi and Woo2010). These analyses run the risk of committing ecological fallacies and are not well-suited to testing the causal mechanism linking citizens’ perceptions of corruption and reward-punishment behaviour.
We study individual voting behaviour in 20 parliamentary elections in Western and Central Eastern Europe using the CSES Module 2 survey data (CSES, 2007).Footnote 3 These data enable us to study the effect of corruption perception using identical question wording in various political and economic contexts.
Dependent variable
Our dependent variable is vote choice, distinguishing respondents who voted for an incumbent government party (1) and those casting a vote for an opposition party (0).Footnote 4 The data on the government status of each political party running in the respective election is retrieved from the ERD data archive (Andersson et al., Reference Andersson, Bergman and Ersson2014).
Independent variable
The key independent variable captures each individual’s perceived level of corruption and is based on the following survey item:
How widespread do you think corruption such as bribe taking is amongst politicians in [COUNTRY]: very widespread (4), quite widespread (3), not very widespread (2), it hardly happens at all (1)?Footnote 5
This item takes up the two definitional criteria of corruption perception mentioned before: it highlights the role of public officials as the central agents of corruption, while leaving the extent and type of corruption largely open. Yet, it does not capture which particular political actors are linked to the perceived level of corruption. In that respect, the question mirrors that of the economic vote where respondents are asked to assess the state of the economy without an explicit reference to the government’s impact on the state of the economy (see, e.g., Lewis-Beck, Reference Lewis-Beck1988: 35–37). As for the economy, we assume that voters implicitly hold government parties responsible for policy outcomes because it is these parties that are in charge of making and enforcing laws that prevent undesirable outcomes. We cannot, however, use this question to infer respondents’ views on the effectiveness of government actions (for a general overview of performance and government responsibility, see Glasgow et al., Reference Glasgow, Alvarez and VanBeselaere2002).
Moderating variables
Since our interest lies primarily in the contingent effect of corruption perception, our empirical models feature interaction terms with four moderating variables. First, we test whether partisanship moderates the effect of corruption performance voting: Non-partisans should be more likely to vote based on their performance evaluations than voters who feel close to a political party. We identify voters’ partisanship using the CSES question item ‘which party do you feel closest to?’ and use indicator variables to distinguish between those who feel close to government parties, opposition parties, and non-partisans. As above, the data on government participation of political parties are retrieved from the ERD data archive.Footnote 6 The second individual-level characteristic refers to the voters’ belief that government turnover will bring about change. We capture this belief with an item (measured on a five-point scale) indicating whether respondents think that ‘it makes a difference who is in power’ (5) or not (1).
For macro level effects, we measure the severity of corruption using the CPI provided by Transparency International. The CPI is a composite index and combines surveys and assessments of corruption by several institutions and organizations into a single measure. It is generally considered the most valid and reliable cross-national estimate of corruption and is thus widely used in comparative studies (see, e.g., Treisman, Reference Treisman2000; Persson et al., Reference Persson, Tabellini and Trebbi2003; Krause and Mendez, Reference Krause and Mendez2009; Choi and Woo, Reference Choi and Woo2010).Footnote 7 For each election we capture the severity of corruption via the country’s average CPI score over the entire parliamentary term. In this context, we inverse the original eleven-point scale so that the maximum score (10) corresponds to high levels of corruption, while a minimum score (0) indicates low corruption levels.
Our last moderating variable is clarity of responsibility. There is no established measure of clarity of responsibility in the existing literature (see, e.g. Whitten and Palmer, Reference Whitten and Palmer1999; Tavits, Reference Tavits2007; De Vries et al., Reference De Vries, Edwards and Tillman2011). Hobolt et al. (Reference Hobolt, Tilley and Banducci2013) argue that attributes of the incumbent government (which they label ‘government clarity’) are more relevant than institutional factors (such as bicameralism and the committee structure in parliament). Thus, we follow their approach by compiling a composite index of government clarity ranging from 0 (low clarity) to 1 (high clarity). This index is based upon each government’s (i) single-party status, (ii) the extent of cohabitation within semi-presidential systems, (iii) a measure of its ideological cohesion, and (iv) the main governing party’s share of cabinet posts.
Control variables
The incumbent government’s economic performance is captured via the level of unemployment in the year before the election. Ideally, including measures of the perceived state of the economy at the individual level would allow us to test the effect of corruption performance voting side-by-side with economic performance voting. Unfortunately, such individual-level measures are not included in the CSES Module 2 survey data.Footnote 8 We resort to using the level of unemployment as a macroeconomic indicator that is likely to have the most direct effect on each respondent’s individual evaluation of the government’s economic performance.Footnote 9 These macro-economic data are retrieved from the World Development Indicators database (The World Bank, 2013). At the individual level, we include a series of standard socio-demographic controls, namely age (in years), sex, education (distinguishing four levels of education), and voters’ ideological preferences measured on an eleven-point left-right scale.
Empirical analysis
Given the hierarchical structure of our data with individuals nested within elections and the binary nature of the dependent variable, the subsequent empirical analysis rests upon a series of multilevel logistic regression models of incumbent voting with random intercepts accounting for unobserved heterogeneity between elections.Footnote 10 The results of these hierarchical logistic regression models are shown in Table 1. Model 1 shows the overall effect of corruption performance on vote choice. Model 2 includes moderating effects for both individual-level factors (Hypotheses 1 and 2). We test the electoral context effects of the systemic level of corruption (Hypothesis 3; Model 3) and of clarity of responsibility (Hypothesis 4; Model 4) separately. All estimated marginal effects and associated marginal-effect plots are based on a full model (Model 5) including all covariates.
Random intercept models of incumbent voting
Standard errors in parentheses.*P<0.05, **P<0.01, ***P<0.001.
We start by analysing whether voters resort to corruption as a simple means of retrospective voting. As expected, corruption perception has a significant negative effect on incumbent voting. Thus, individuals perceiving higher levels of corruption are less likely to vote for a party in government. Increasing the level of corruption perception by one unit (on a four-point scale) lowers the probability of incumbent voting by ~5.4 percentage points, with all other independent variables held at their mean or mode, respectively.Footnote 11 To put this effect in perspective, we compare this estimate with that of a one unit change of voters’ economic perceptions. Studying 163 different estimates of economic perception voting, Duch and Stevenson (Reference Duch and Stevenson2008: 64) find that a one-unit change in economic perceptions (on a three-point scale) decreases the probability of incumbent voting on average by 5.0 percentage points. Despite the difficulties of comparing effect sizes across the different models,Footnote 12 we may tentatively conclude that the substantive impact of corruption perceptions on vote choice is similar to that of economic perceptions.
This baseline effect of corruption perception on incumbent voting may serve as a starting point to evaluate the conditional factors in Hypotheses 1–4. Starting with the individual-level factors that may affect performance voting, our results indicate that the effect of corruption perception on incumbent voting is characterized by a strong partisan bias. Figure 1 juxtaposes the particularly influential effect of corruption perception for non-partisan voters with that of voters feeling close to an opposition party, and those who feel close to the incumbent government.Footnote 13 Corroborating our first hypothesis, the perceived level of corruption has only little influence on partisans. In fact, individuals closely tied to the incumbent are likely to support them regardless of their corruption perception, as their likelihood of incumbent voting merely decreases from 94.3 to 91.4% as corruption perception changes from ‘hardly happens at all’ (1) to ‘very widespread’ (4). In a similar vein, individuals close to the opposition are unlikely to support incumbent government parties regardless of their level of perceived corruption (the corresponding change in the predicted probability of incumbent voting is from 4.4 to 3.3%). Overall, it is non-partisans for whom the perceived level of corruption matters most for their vote choice. In contrast, those who feel close to a party have already relatively clear preferences and thus changing corruption perceptions hardly alter their vote choice.
Corruption perception effect depending on party affiliation (H1). Note: This figure shows the estimated marginal effect of a one-unit change in corruption perceptions on the probability of incumbent voting for non-partisans, voters feeling close to an opposition party, and voters feeling close to a government party. All point estimates (dots) and corresponding 95% confidence intervals (vertical lines) are based on Model 5.
A second moderating factor is whether respondents believe that government turnover will yield changes. In line with Hypothesis 2, Models 2 and 5 show a negative interaction between corruption perception and the perceived differences of who is in power, and this effect is statistically significant at conventional levels. Thus, the extent to which voting behaviour is influenced by corruption perceptions increases as respondents believe that opposition parties will govern differently than the incumbent government. Figure 2 shows the marginal effect of corruption perception dependent on the perceived effect of government turnover, holding all other variables at their mean or mode, respectively. Here, we observe that corruption perceptions have almost no impact on the likelihood of incumbent voting (increase by ~0.2 percentage points, insignificant) for individuals who have little faith in government turnover. In fact, a minimum sense that ‘it makes a difference who is in power’ is necessary so that corruption perceptions influence one’s voting behaviour.Footnote 14 For these voters, an increase in the level of perceived corruption negatively affects the probability to vote for incumbents by ~3.7 percentage points, and this effect becomes stronger as the perceived effect of government turnover increases. For individuals who strongly believe in the changing effect of government turnover (five scale points), increasing the level of corruption perception by one unit decreases the likelihood of incumbent voting by 7.8 percentage points.
Corruption perception effect depending on effect of government turnover (H2). Note: This figure shows the estimated marginal effect of a one-unit change in corruption perceptions on incumbent voting, conditional on the perceived effect of government turnover. All point estimates (dots) and corresponding 95% confidence intervals (vertical lines) are based on Model 5.
Turning to the macro-level context factors (Models 3 and 5), we hypothesized that corruption performance voting is more relevant in countries with high levels of corruption (Hypothesis 3). Figure 3 shows the corresponding marginal effect of corruption voting for varying levels of corruption, again holding all other variables in the model at their mean or mode, respectively. Indeed, we observe a strong punishment effect by the electorate in national contexts where corruption is highly salient (e.g. Romania in 2004). If corruption perceptions increase by one unit in such contexts, we would predict that the probability of voting for the incumbent declines by 9.2 percentage points. This effect is approximately three times larger than the corresponding effect in elections where corruption is hardly an issue (e.g. Finland in 2003). Here, increasing the level of corruption perception by one unit decreases the likelihood of incumbent voting by ~3.2 percentage points.
Corruption perception effect depending on corruption severity (H3). Note: This figure shows the estimated marginal effect of a one-unit change in corruption perceptions on incumbent voting, conditional on corruption severity. All estimates (solid line) and the corresponding 95% confidence interval (dashed lines) are based on Model 5. The empirically observed maximum on the Corruption Perception Index (CPI) 0–10 scale is 7.2.
Concerning Hypothesis 4, the results of Models 4 and 5 do not support the claim that corruption performance voting varies with the clarity of government responsibility. While the corresponding coefficient for Corruption perception×government clarity is negative, it fails to reach conventional levels of significance, indicating that varying levels of government clarity do not affect the effect of corruption voting.Footnote 15 Thus, these results indicate that individuals are able to clearly attribute and punish corrupt behaviour, regardless of factors blurring this link (see Figure 4).
Corruption perception effect depending on government clarity (H4). Note: This figure shows the estimated marginal effect of a one-unit change in corruption perceptions on incumbent voting, conditional on the level of government clarity. All estimates (solid line) and the corresponding 95% confidence interval (dashed lines) are based on Model 5.
Finally, Models 1–5 also contain the level of unemployment as a control variable. In line with the expectations of the economic performance voting literature, the coefficient of unemployment is negative: an increase in unemployment by one standard deviation above the mean decreases the probability to vote for incumbent parties by 7.9 percentage points. This suggests that performance voting may encompass economic as well as non-economic performance evaluations and that their combined effect is larger than either that of economic and of corruption performance voting alone.
Conclusion
Are elected officials held accountable for high levels of corruption? The evidence presented here suggests that voters indeed sanction governing parties if they perceive high levels of corruption. Yet, this effect depends on both individual voter characteristics and the electoral context. These findings might help us to understand why corruption is a persistent phenomenon in some contexts but not in others. Regarding individual characteristics, we showed that corruption voting depends on partisanship. It is non-partisans who make vote choices based on perceived levels of corruption. Partisans who lean towards government or opposition parties have a strong predisposition towards these parties. For such voters, performance evaluations are far less important for their vote choice. This is in line with previous research showing varying effects of economic voting for partisans and non-partisans (e.g. Kayser and Wlezien, Reference Kayser and Wlezien2011), suggesting that partisanship moderates the effect of voters’ attitudes on vote choice.
Our results also suggest that electoral control diminishes as citizens lose faith in replacing incumbents with opposition parties. Only citizens who believe that government turnover will lead to substantial changes punish incumbents by voting for an opposition party. In contrast, those who perceive no differences between government and opposition do not vote for opposition parties if the perceived level of corruption is high. This highlights the importance of the voters’ perceived differences between parties, for example in terms of policy positions (see Hellwig, Reference Hellwig2012) and party competence.
The CSES data from 20 elections in different national contexts also allow us to analyse the electoral context of corruption performance voting. Corruption voting increases in a setting with high corruption, indicating that performance voting depends on the saliency of the issue voters use to evaluate government parties. The normative implications of this finding are twofold: First, it implies that voters facing high and persistent levels of corruption indeed try to bring down these high levels by closely monitoring the incumbent parties. Second, it means that corrupt political elites might escape electoral punishment as long as corruption is not among the most salient issues.
In contrast to the economic dimension of performance voting, we find no evidence that clarity of responsibility affects the magnitude of corruption performance voting. This negative finding, which is robust to various approaches to measuring clarity of responsibility, illustrates that performance voting is a multi-faceted phenomenon: while being a strong moderating variable for the influence of voters’ economic perceptions on incumbent support (Powell and Whitten, Reference Powell and Whitten1993; Whitten and Palmer, Reference Whitten and Palmer1999; Anderson, Reference Anderson2007; Duch and Stevenson, Reference Duch and Stevenson2008), the same does not hold for corruption perception voting. Perhaps the signal-extraction problem, fostered by power dispersion, is more severe for economic policies than for corruption.
Economic policies partly entail positional competition, and these conflicts affect legislation on economic policies. Thus, power dispersion leads to a signal extraction problem and makes economic performance voting less likely. In contrast, fighting corruption involves less ideological conflict and is less dependent on legislative politics. Fighting corruption is largely a matter of executive politics, the enforcement of existing rules and laws, and effective control in the public administration. Moreover, corruption scandals are often highly visible and clearly linked to names of parties and individual politicians, irrespective of the political and institutional context. While this remains speculation, extending the theoretical framework of performance voting to explain these differences is a crucial step towards a better understanding of individual voting behaviour.
This study focused on the influence of corruption perceptions on incumbent support, but there are several further implications for other aspects of electoral politics. For example, the results presented here may have implications for the emergence and electoral success of new parties (Lucardie, Reference Lucardie2000; Tavits, Reference Tavits2006; Sikk, Reference Sikk2012). Following Lucardie (Reference Lucardie2000: 182), ‘new parties need a political project that caters to social problems perceived as important by significant numbers of voters in order to win support’. In contexts where voters perceive high levels of corruption, corruption can serve as a fruitful ground for new parties that run on an anti-corruption image (see also Sikk, Reference Sikk2012), with sustainable effects if these parties can defend their distinct profile (Bolleyer and Bytzek, Reference Bolleyer and Bytzek2013). Furthermore, corruption perceptions not only affect party choice but also turnout (Stockemer et al., Reference Stockemer, LaMontagne and Scruggs2013). Yet, there is no consistent evidence whether turnout increases or decreases together with levels of corruption. The results presented here suggest that these differences depend on whether voters believe that replacing the government would make a difference. Those voters who believe in different outcomes under a government led by the current opposition parties punish incumbents for high corruption, while we find no such effect for those who perceive few differences between government alternatives. Thus, the effect of corruption perceptions on turnout may well be conditional on citizens’ beliefs about whether government turnover actually makes a difference (see also Kostadinova, Reference Kostadinova2009). The results presented in this study highlight the importance of such conditional effects of corruption perceptions. It is likely that similar effects are relevant for related research questions.
The present study also encourages research on temporal effects on performance voting. We focus on corruption performance voting in different national contexts, and data constraints do not allow us to include cross-temporal variation in corruption performance voting as well. Nevertheless, we believe that time-series data linking corruption perceptions and vote choice could reveal further insights. For example, voters may be most responsive to sharp increases in perceived corruption and reward incumbents if they perceive decreasing levels of corruption. Including these factors would improve our understanding of voter behaviour in general, and corruption performance voting in particular.
Acknowledgements
We gratefully acknowledge financial support from the Austrian Science Fund (FWF) (grants P25490, S10902-G11, and S10903-G11) and thank Hannah Kieblspeck for excellent research assistance. We are indebted to Michael Lewis-Beck, Markus Wagner, Eva Zeglovits, and the participants of the seminar series at the Department of Methods in the Social Sciences and the Department of Government, University of Vienna, for helpful comments and suggestions on earlier drafts. We also thank the three anonymous reviewers and the editors at EPSR for their support and helpful suggestions.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1755773915000053
Open access
