Corruption and Electoral Support for New Political Parties in Central and Eastern Europe

More than 20 years after the collapse of the Soviet Union, the electoral volatility in Central and Eastern Europe (CEE) is still remarkably high. A considerable part of the volatility derives from the votes for new political parties, since they are very often on the winning side of elections. This article examines corruption as a potential determinant of their electoral support. It argues that the effect of corruption is twofold: on the one hand, the historically derived corruption level reduces the electoral support for new political parties due to strong clientelist structures that bind the electorate to the established parties. On the other hand, an increase in perceived corruption above the traditional corruption level leads to a loss of trust in the political elite and therefore boosts the electoral support for new competitors. A statistical analysis of all democratic elections in CEE between 1996 and 2013 confirms these two counteracting effects.

therefore analysed the electoral support for new political parties separately from volatility in general (Andrews and Bairett 2014;Birch 2003;Mainwaring et al. 2009;Powell and Tucker 2014;Sikk 2005;Tavits 2008). The impact of the economic performance, institutional variables, the duration of democracy and ethnic as well as political fragmentation have been considered in most of these studies.
This article explores another potential determinant that might explain the electoral support for new political parties, which has only received limited attention in previous research: corruption. 1 The sparse attention paid to the impact of corruption is surprising, since many new political parties, which outperformed the political establishment in the past, have prominently politicised corruption (Bågenholm 2013a;Bågenholm and Charron 2014;Hanley and Sikk 2014). Successful examples are the Bulgarian National Movement Simeon II (NDSV) in 2001 and the Citizens for European Development of Bulgaria (GERB) eight years later, Tradition Responsibility Prosperity 09 (TOP09) and Public Affairs (VV) in the Czech Republic 2010, or more recently the Czech Action of Dissatisfied Citizens (ANO) and Positive Slovenia. In fact, Bågenholm and Charron (2014) have shown in a statistical analysis of Western and Eastern European countries that new political parties that do politicise corruption are more successful than new parties that do not. 2 Hanley and Sikk (2014) show furthermore that these so-called anti-establishment reform parties win more votes in times of increasing corruption, since they benefit from the loss in trust which is caused by such an increase. The negative relationship between corruption and political trust (i.e. citizen's trust in government, administration, parliament and political parties) has already been confirmed by many studies in the past (Anderson and Tverdova 2003;Kostadinova 2012;Mishler and Rose 2001;Seligson 2002). The electoral success of parties politicising corruption and the results from previous research on political trust provide strong evidence that corruption can positively impact the political support for new political parties.
However, when only looking at successful parties politicising corruption, one might forget that corruption, given its strong linkage to the historically derived clientelist structures (see Kitschelt et al. 1999), can also serve as a stabilising factor. Clientelism binds the electorate to the already established political parties, lowering the electoral prospect of new political parties that have neither existing clientelist structures nor access to state resources to build up such structures.
The aim of this article is to analyse the ambiguous role of corruption in explaining the electoral support for new political parties. It makes an important contribution by providing a framework that allows for theoretical and empirical differentiation between the two counteracting effects. By this means, the empirical analysis of this article demonstrates that corruption does play an important role in determining the electoral support for new political parties, but that its real impact is only visible when both effects of corruption are taken into account.
The article is structured as followed: the next section discusses the various definitions of new political parties and presents the coding rules by which parties have been coded. The following section provides an overview of the substantive literature on the electoral support for new political parties. I then offer a theoretical framework for my argument on the two distinct effects of corruption. Section four presents data and measurement used in the statistical analysis of all democratic parliamentary elections in CEE countries between 1996 and 2013; the results are explored in section five. A final discussion provides interpretation and draws conclusions from the findings.

New Political Parties
Previous studies on electoral support for new political parties offer different definitions for new parties. For example, while Powell and Tucker (2014) offer a broad interpretation that categorises all newly appearing parties as new, 3 Sikk (2005) labels a political party as new only when it is the result of neither a party merger nor a party split. In this article, I follow Tavits (2008) and Hug (2001), who suggest excluding merged parties whilst including parties that have emerged after a party's split. Hug (2001: 13-14) argues that party mergers and alliances are only used as a vehicle for maximising the share of votes for already existing parties. A party merger is therefore no new political force in the party systemit just replaces parties of the old establishment. Splits and political parties without any connection to the already existing parties, in contrast, compete against the old political establishment and thus can be considered as a potentially new political force. I chose this definition because it emphasises most clearly the competition between the old political establishment and new political players, and consequently, the dynamics of the party system. Based on this definition, I applied the following coding rules: (1) A party is new if it has achieved at least 2 per cent of the vote for the first time in its history. 4 (2) Electoral alliances and mergers are only considered as new if none of the affiliated parties have ever won seats in national parliament. 5 (3) If a party split occurred, at least one party is coded as new. The party that is identifiable as the successor party (e.g. the same party label and/or the same party leader) is not coded as new. If none of the parties are identifiable as the successor party, all parties are coded as new. Figure 1 shows the accumulated share of votes of new political parties in all democratic elections in CEE between 1990 and 2013. A list of all new parties and their electoral support can be found in the web appendix. The differences between the elections within individual countries are remarkably high. Not surprisingly, most of the second democratic elections had a high share of votes for new political parties. This can be explained by the fact that, in the first democratic elections, political parties often built up ideologically heterogeneous electoral alliances with the only common goal of sustaining or replacing the old regime. In the second democratic election, these electoral blocs have often split into several smaller parties (Rose and Munro 2009: 50). 6 However, also in subsequent elections new political parties have been very successful from time to time. However, systematic differences between countries are harder to identify. With the exception of the non-EU countries Albania, Macedonia and Montenegro, only Croatia and Romania have never had new political parties winning more than 15 per cent of the votes.

Literature Review
Only recently have a few studies attempted to explain the variation in the electoral support for new political parties in Central and Eastern Europe (Andrews and Bairett 2014;Birch 2003;Mainwaring et al. 2009;Powell and Tucker 2014;Tavits 2008). In Western Europe, on the other hand, research on new political parties had already started with the rise of the Greens and right-wing populist parties in the last decades of the twentieth century (see e.g. Kitschelt 1995;Müller-Rommel 1993 Notes: 2 (second), 3, 4, 5, 6, 7, 8 (eighth) election with a polity IV score higher than 6 (year: see Table A1 in the appendix). Albania (AL); Bulgaria (BG); Croatia (HR); Czech Republic (CZ); Estonia (EE); Hungary (HU); Latvia (LV); Lithuania (LT); Macedonia (MK); Moldova (MD); Montenegro (ME); Poland (PL); Romania (RO); Serbia (RS); Slovakia (SK); Slovenia (SI); Ukraine (UK (e.g. post-materialism, immigration) that the new party adopted (see Sikk 2012). Since most of these studies restricted their analysis to one new party or one new party family, the focus on a new salient issue is not surprising. Hug (2001) was one of the first scholars who systematically analysed the emergence and the initial success of all types of new political parties; independent from their political ideology. Nevertheless, in his comprehensive theoretical framework, a new salient issue remains the starting point, indispensable for the appearance and success of a new party. He argues further that a new party only emerges if the established parties do not adopt the new issue in order to prevent the entry of a new party. Finally, the formation of a new party depends on the institutional and political framework, such as the electoral system and costs of party formation. A proportional electoral system facilitates winning seats and thus increases the incentives to participate in elections. Costs of party formation such as signature requirements or a registration deposit, on the other hand, lower the probability of party formation. Cox (1997: Chapter 8) reaches similar conclusions: the decision on party formation depends on a cost-benefit analysis. Again, both the party formation costs as well as the electoral system are crucial factors. In addition, the attractiveness of holding a political office does also increase the likelihood of forming a party. Studies on the support for new political parties in Eastern Europe also examine the institutional design of the country's political system (Andrews and Bairett 2014;Birch 2003;Mainwaring et al. 2009;Powell and Tucker 2014;Tavits 2008). Except for Mainwaring et al. (2009), these studies have found significant effects of at least one attribute of the electoral system, such as the average district magnitude, the electoral design (proportional vs. mixed proportional/single-member district) and the electoral threshold, whereas the electoral design received the most ambiguous results (Andrews and Bairett 2014;Birch 2003;Powell and Tucker 2014). Whether or not the potential benefits of running for office are high is often measured by the political system. For an individual, it is more profitable to form a new party if it can run for the presidency rather than just for a parliamentary seat (Tavits 2008). Andrews and Bairett (2014) as well as Tavits (2008) could confirm a positive relationship between a directly elected president and the entry and/or support for new political parties, whereas Powell and Tucker (2014) as well as Mainwaring et al. (2009) could not.
While the institutional explanation has been applied to the Eastern European setting, the new salient issue argument has often been largely ignored. In one of the few studies, Tavits (2008: 118) mentions the topic of salient issues briefly, but still excludes it from her work. She argues that this theory is only reasonable if one wants to explain the rise of a certain party or party family but not if one wants to explain the emergence and success of all new political parties. It is obvious that when explaining the support for all new political parties one cannot include a direct measure for the new salient issue, since this matters only for a certain type of new parties. Hug (2001: 89) and Harmel and Robertson (1985) provide a solution by measuring the likelihood of the appearance of a new salient issue within a society with proxies such as population size, ethnic fragmentation or economic performance. The validity of these indicators is disputablee.g. ethnic fragmentation and economic performance can serve as an explanation by themselves (as we will see later), and consequently their effect cannot be attributed automatically to the emergence of a new issue. Another reason for excluding new salient issues as explanations might be the different nature of the party systems in Western and Eastern Europe. Whereas successful new parties in Western Europe are counterevidence to the frozen cleavages theory proposed by Lipset and Rokkan (1967) and explaining their success signifies that one has to understand why the party system can still change from time to time, in Eastern Europe the opposite is true. Volatility has been high from the beginning, and successful new parties play an important part in the fact that it is still high today. Since political positions of new or young parties are not as clear-cut as the positions of already existing parties, the overall cleavage structure is also less rigid and definite than in Western Europe. This in turn weakens the argument that new political parties emerge and win seats primarily because of a new salient issue. Empirical evidence for this notion is found by Sikk (2012). His findings show that none of the successful new parties in the Baltic States positioned itself on a new issue and that their winning formula was 'newness' itself. Hug's (2001) proxies for the probability of new salient issuesethnic fragmentation and economic performancehave nevertheless found their way into the analyses of Eastern European new political parties. Different types of indicators for economic performance are included in all studies. The theoretical foundation is the economic voting theory which states that voters punish the governing parties in bad economic times (see e.g. Duch and Stevenson 2008). Tavits (2008) as well as Powell and Tucker (2014) have also analysed the effect of ethnic fragmentation. Theoretically, one can argue in both directions. On the one hand, ethnicity is a rather stable cleavage that can stabilise a party system. On the other hand, it can also be destabilising; this is the case if ethnic minorities do not feel well represented by the political establishment and thus have to rely on new parties that promise to do better (Tavits 2008: 130).
In addition, the fragmentation of the party system has been considered by Powell and Tucker (2014) and Mainwaring et al. (2009) who argue that a high number of political parties symbolises an open political system and therefore has a signalling effect on potential new parties. In order to account for a potential stabilisation over time, the number of years since the first free and fair election has also been included in most studies (see Tavits 2008: 117). Tavits (2008) discussed the effect of ethnic fragmentation and economic performance on the electoral support for new political parties within the broader theoretical framework of disappointed voters. She argues that new parties win votes if voters are dissatisfied with the political elite. Contrary to the institutional approach, this approach is also able to explain within-country variance. However, an important source of dissatisfaction is still missing: corruption. Several studies have found corruption to negatively affect social and political trust (see e.g. Anderson and Tverdova 2003;Kostadinova 2012;Mishler and Rose 2001;Seligson 2002). Although the negative impact of corruption on political trust is undisputed, none of these studies looking at new parties in CEE countries have included corruption. 7 Two recently published articles examine whether corruption affects the electoral support for new (and opposition) parties that politicise corruption (Bågenholm and Charron 2014) ormore generallyof new anti-establishment parties whose opposition to the political elite can include anti-corruption rhetoric (Hanley and Sikk 2014). Bågenholm and Charron (2014) analysing Eastern and Western European countriesshow, first, that new political parties are more successful than their counterparts that do not politicise corruption, and second, that this effect is stronger in countries with higher corruption. Hanley and Sikk (2014) only looking at countries in CEEprove a positive relationship between increasing corruption and electoral support for new anti-establishment parties. However, the level of corruption shows a less clear picture; being beneficial in some, but not all circumstances. These studies provide strong evidence for the claim that corruption must not be neglected when trying to explain electoral support for new political parties in general. Yet, although the results of the latter study indicate it, they fail to discuss the ambiguous role of corruption, which can also hinder the successful entry of new political parties.

Corruption and its Positive Impact on the Electoral Support for New Political Parties
Several studies have found a negative relationship between corruption and political trust (Anderson and Tverdova 2003;Kostadinova 2012;Mishler and Rose 2001;Seligson 2002). Corruption undermines the democratic rules by favouring some citizensfor example the wealthyover others (Anderson and Tverdova 2003). Consequently, this creates a large amount of mistrust between the majority of citizens and their political leaders. For example, Rothstein and Uslaner (2005) have shown that corruption exacerbates social inequality which in turn reduces peoples' social trust. In particular, Kostadinova (2012: Chapter 8) found a loss of trust in several different political institutions: corruption not only lowers trust in government and public administration, but also in parliament and political parties in general.
In regard to new political parties, I argue that a loss in political trust, especially in political parties, decreases the political support for traditional political parties and thus increases the electoral prospects for new political parties. So far, previous studies have only asked what consequences corruption has on political incumbents. Peters and Welch (1980), as well as Ferraz and Finan (2008), have shown that corruption allegations lower the electoral prospects of the accused politicians. Other findings demonstrate that if corruption is perceived to be high, the electoral support for governing parties is decreasing (Bågenholm 2013b;Klašnja et al. 2014;Krause and Méndez 2009). Whether frustrated citizens vote for new political parties or for traditional opposition parties has not yet been systematically analysed. However, the success of new anti-corruption parties in the last two decades (Bågenholm 2013a;Bågenholm and Charron 2014), in particular in countries with increasing corruption (Hanley and Sikk 2014), suggest so. Theoretically one can explain this by the assumption that citizens rather find a link between corruption and traditional opposition parties than with new political parties, since new parties have not been part of the 'corrupt' political system before.
Several studies support the assumption that emerging political parties benefit more from a loss in political trust. Bélanger (2004), using survey data, has shown that traditional parties tend to lose political support to third parties (i.e. not governing or main opposition party) from voters that had a generalised antiparty sentiment, but only if the third party was able to incorporate such antiparty feelings. Otherwise citizens do not give their vote to the main opposition party; they just abstain from voting. Furthermore, an experiment by Pop-Eleches (2010) provides additional empirical evidence of this claim. In his study, he asked people from Bulgaria which party they would vote for if the governing parties were faced with corruption allegations. Most of the participants chose the new political party. In a first step, I therefore argue that if corruption is perceived 8 to be high, electoral support for new political parties increases.
Before analysing the relationship between perceived corruption and the electoral support for new political parties, one has to consider that there are systematic differences between levels of corruption from one country to another. 9 Whereas Slovenia has always had one of the lowest levels of corruption in CEE (only outperformed by Estonia from 2010/11 onwards), Ukraine has remained one of the countries with the highest corruption during the period from 1996 to 2013 (see also Figure A1 in the appendix). 10 To test the relationship stated above, the analysis must account for the path dependence of the main explanatory variable. This problem is more prominent in the theory of economic voting, where it is assumed that the economic situation explains the electoral outcome (see e.g. Duch and Stevenson 2008), the main explanatory variable (e.g. GDP, unemployment) being highly path dependent. In order to avoid testing the very unlikely case that voters punish the political incumbents only because the country is still worse off than other countries, although it is growing faster, scholars include measures such as economic growth that compare the current economic situation with the previous one (see e.g. Tavits 2008). I applied a similar strategy here. It is unlikely that, for example, every year Slovenians will reward the political elite for still being one of the least corrupt countries, although corruption is increasing. Therefore a relative measuresuch as the change in corruption or the deviation from the countryspecific levelis necessary. Considering only the change in corruption from one election to another, one would presume that voters evaluate the current level of corruption from a short-term perspective, independently from the corruption levels in earlier years. The deviation from the country-specific corruption level, on the contrary, allows a long-term perspective and therefore also accounts for the countries' structures that are responsible for the corruption's path dependency (discussed in more detail below). It is very unlikely that an increase in corruption will be evaluated equally by voters in a country that has never experienced such high corruption before as in a country where the increase is just a reversion to the norm. The latter approach incorporates all past values and is therefore theoretically more adequate than a short-term perspective which assumes that every change in corruption has the same meaning in any country at any point in time. This is also illustrated by the opposite situation: Looking at the deviation from the country-specific corruption level, a decrease in perceived corruption after a series of corruption scandals that raised the corruption level above the country-specific level, is measured as normalisation and not as an improvement (as it would be from a short-term perspective).
Only an actual reduction in the corruption level below the country-specific level, for instance, through a successful anti-corruption campaign by the government, would increase the political trust of voters while decreasing the electoral prospects of new competitors. Thus, the first effect of corruption is hypothesised to be the following: H1: An increase (decrease) of the perceived corruption level above (below) the country-specific corruption level increases (decreases) the electoral support for new political parties.
The Country-Specific Corruption Level and its Negative Impact on the Electoral Support for New Political Parties In the section above, I argue that new parties benefit from an increase in corruption. However, I expect the opposite effect for the country-specific corruption leveli.e. that new parties are more successful in low-corruption countries than in high-corruption countries. Contrary to the fluctuations of corruption that influence the (dis)satisfaction with the political establishment, the country-specific level of corruption indicates the nature of party-voter linkages and how important time and state resourcesboth missing in new political partiesare to maintain them.
In their seminal work, Post-Communist Party Systems, Kitschelt et al. (1999) argue that path dependence in corruption derives from the differences in the historical legacies of post-communist countries. The different legacies determine whether the linkage between citizens and parties is either organised in a more programmatic or more clientelistic manner. 11 Kitschelt (2001: 316-15) demonstrates this relationship by comparing the countries' corruption levels, which he used as an indicator for clientelism, with their typology of the historical legacies (see Table 1). He found that corruption levels vary in the same way as predicted by the typology. In particular, the patrimonial countries, predicted as having the strongest clientelistic party-voter linkages, were the most corrupt countries in the sample.
According to Kitschelt et al. (1999;see also Kitschelt 2001: 307ff.), the reasons for the different degrees of clientelism can already be found in the inter-war period. Back then, the countries of patrimonial communism were still agrarian societies, and thus, contrary to type (1) and (2), there was no urban middle class that opposed a communist regime. Whereas the communists could stay in power due to a mobilised and relatively strong working class in the bureaucratic-authoritarian type and concessions to the middle class in the national-accommodative type, these strategies were neither necessary nor feasible in patrimonial communism. Instead, the communists secured their power through clientelistic and repressive structures. The absence of an urban middle class also had a significant influence on the transition to democracy. Whereas the existing middle class in the other two types could rely on the political experiences gained during the inter-war period and/or the capacity to organise themselves under communist rule, the greatly oppressed opposition in patrimonial communism was too weak to oust the communist party from power. The communists remained in power and could continue to exert control over the state apparatus and, consequently, also over economic and political reforms. This gave them enough resources to maintain the clientelistic structures. The communist party was also not forced to reform itself, contrary to those in type (1) and (2), where the communists started to rely more on programmatic claims in order to increase their electoral prospects. According to Kitschelt et al. (1999), the reason why clientelism is still present in today's politics of the patrimonial countries is that the communists could design the political institutions accordingly. Additionally, Hale (2007) argues that clientelism continues to exist because less radical economic reforms have been made in the beginning. Without such reforms, the economy continues to be strongly

TA B L E 1 T H R E E T Y P E S O F C O M M U N I S T R U L E B Y K I T S C H E LT E T A L . (1 9 9 9 )
(1) Bureaucraticauthoritarian communism (1) and (2) (2) Nationalaccommodative communism (2) and (3) ( 3) Kitschelt et al. (1999: 39) and Kitschelt (2001: 315). connected to the state and therefore the population remains dependent on the state. This again, facilitates clientelism. Sixteen years after the publication of Kitschelt et al.'s (1999) typology, the predicted relationship still exists (see Figure A2 in the appendix). The corruption levels of the patrimonial countries are still the highest; none of the nonpatrimonial countries have reached the same level. The picture of the mixed category is less uniform. While the Baltic States are closer to the non-patrimonial countries, Serbia and Montenegro (and to a lesser extent Slovakia) have corruption levels comparable with the patrimonial countries. This is less surprising when considering the long period of time it took Serbia and Montenegro to break with the authoritarian past, having therefore a more similar legacy to the patrimonial countries. In addition, Slovakia under Prime Minister Mečiar (1990Mečiar ( -1998 was an illiberal democracy similar to patrimonial Bulgaria and Romania and thus had a similar pace towards democracy and a market economy (see Vachudova 2005: Chapter 2).

Mix of
Assuming that the path dependence of corruption is mainly characterised by the different extent of clientelism, the country-specific corruption level determines how strongly politics function through exchange mechanisms between a party and its supporters. Instead of programmatic claims, parties obtain votes in exchange for provision of goods, jobs in the public sector and facilitating access to social security services. Beside the mutual relationship between voters and parties, clientelism also includes networks with resourcerich constituencies (e.g. entrepreneurs, investors) that exchange money for privileged treatment in the public procurement system, regulatory decision proceedings and other sectors controlled by the party (Kitschelt 2000: 849). This, in turn, generates more resources for the electoral campaign, as well as for maintaining the resilience of clientelist networks with the party's voters.
The Montenegrin Democratic Party of Socialists (DPS) is an example of a former ruling communist party that has very successfully bolstered its clientelist networks, and as a consequence, never had to leave office since the first multiparty elections in 1991. The local NGO, the Network for the Affirmation of the NGO Sector (MANS), provides evidence of the use of clientelistic strategies even in the most recent elections in 2012. It reports a significant increase in public spending during the months before the elections, with evidence that the entitlement to benefits were dependent upon party affiliation. It also found state-owned companies and public institutions compiled lists of their employees, indicating party affiliation. 12 Not only communist successor parties rely on their long-standing clientelist networks. During the opportunistic periods of transition, parties created by influential people, who were able to adapt rapidly to the new post-communist environment, were also able to create new clientelist networks. This was most feasible in patrimonial countries, where the process of transformation to a market economy was slow. Consequently, the lines between private and public sector were kept blurred while a large part of the population and economy remained dependent upon the state (Hale 2007;Zimmer 2005: 370). Zimmer (2005) demonstrates this with the example of the Ukrainian Party of Regions, formerly led by Viktor Yanukovych. The party secured its high level of support in the Donets'k region due to a clientelist network with its voters and the business sector.
These examples illustrate how important state resources are in maintaining clientelist networks. Therefore, particularly governing parties provided with state resources have the most opportunities to reward its supporters. This observation is also confirmed by several studies that show how governing parties greatly benefit from clientelism (see e.g. Birch 2007; Manzetti and Wilson 2007;Nichter 2008;Stokes 2005). Since new parties have access neither to existing clientelist structures nor to state resources that could help to develop such structures, it is more difficult for them to succeed in a clientelistic environment than in a programmatic one. Also existing opposition parties have only limited access to state resources. However, unlike new parties, they have several advantages. First, the time dimension differs. Existing opposition parties could build up a clientelist network with their constituencies and business sector supporters during opportunistic periods of transition where state structures were more fluid. Additionally, they could strengthen these relationships over time. Second, although they are not in the national government, they might have access to state resources due to mandates on the local level or in parliament. My second hypothesis is therefore the following: H2: The higher the country-specific corruption level is, the lower the electoral support for new political parties.
Whether clientelism is always a corrupt act is not without controversy. 13 In order to avoid measuring something different than the effect of clientelism, I use supplemental data on clientelist structures. The Democratic Accountability and Linkages Project by Kitschelt (2013) measured clientelistic behaviour by political parties in 88 countries, including all countries in my sample except Montenegro, using expert surveys conducted in 2008/9. 14

Data and Measurement
To assess the impact of corruption on the electoral support for new political parties, I assemble a cross-sectional times-series dataset covering all democratic parliamentary elections in Eastern and Central European countries from 1996 to 2013. 15 The units of analysis are therefore elections. The first free and fair election is excluded in order to make sure that an increase in the electoral support for new parties is not just due to a democratic opening. I limit my sample to Eastern and Central Europe for two reasons. First, all countries are young democracies with a still very volatile party system. Explanations for the success of new parties in established democracies might differ from those in developing democracies; as we have seen for example with respect to the importance of new salient issues. Second, the restriction on one region with a common communist past controls for possible historical and cultural factors that may also influence outcomes. A total of 72 elections in 17 countries are included in the sample. Due to missing values in corruption, 66 elections can be included in the analysis. Table A1 in the appendix lists all elections.

Dependent Variable
The dependent variable is the total share of votes all new parties won in a certain election. 16 The main data source is the Comparative Political Data Set III (Armingeon et al. 2014) for the EU-10 countries and the Comparative Political Data Set II (Armingeon and Careja 2007) for all non-EU countries and Croatia. 17 A party is coded as new according to the coding rules specified above. Information for the coding is taken from Bugajski (2002), Ismayr (2010), the Parliament and government composition database (ParlGov) by Döring and Manow (2014) and several media sources.

The Main Explanatory Variable: Corruption
In order to capture corruption, I use the Control of Corruption Indicator, which is part of the Worldwide Governance Indicators (WGI) provided by the World Bank. 18 Data is available biennially from 1996 to 2002 and annually from 2003 on. This index can be used for cross-national as well as longitudinal comparisons (Kaufman et al. 2010: II). It measures the perceived corruption defined as 'the extent to which public power is exercised for private gain' (Kaufman et al. 2010: 4) and thus corresponds with the definition by Rose-Ackerman (1999: 91) used in this article. In the original dataset, the highest corruption value is -2.5 (low corruption control) and the lowest level of corruption is 2.5 (high corruption control). To facilitate the interpretation of the results, I multiply all values by (-1).
The two hypotheses require a different type of corruption measure. In order to test hypothesis 1, I calculate the deviation of corruption from the countryspecific corruption level (DEVmean) at election t. The country-specific corruption level is measured by the arithmetic mean of the corruption levels between 1996 and 2013 of the respective country. The deviation from it, in turn, is the difference between the corruption level of election t and the country-specific corruption level. To test hypothesis 2which states a negative relationship between the country-specific corruption level and the electoral support for new political partiesthe calculated country-specific corruption level is imputed as a time-invariant variable for every (election of a) country. Additionally, models are calculated with an index of clientelism ranging from 0 (no clientelistic practices) to 20 (strong reliance on clientelistic practices) from the Democratic Accountability and Linkages Project by Kitschelt (2013). The index is based on expert evaluations of the exchange mechanisms used by the single parties in a country, added for each country and weighted by the parties' size. Values for Montenegro are missing.
The measurement of DEVmean underlies the problematic assumption that the voters' perception of the country-specific corruption level is influenced by future values. A theoretically better measurement would be to subtract the corruption at election t from the arithmetic mean of the corruption levels between 1996 and the last year before the election t (deviation from the retrospective country-specific corruption level; DEVretro). Since values are only available from 1996 onward, this retrospective measurement becomes more accurate over time: in 2000, the arithmetic mean depends on just two values, in 2002 on three values, in 2003 on four values, and so on. In addition, fewer elections can be included in the analysis since there is no country-specific country level to compare with for the elections in 1996 and 1998. Because the accuracy of DEVmean is not dependent on time and because it allows all elections from 1996 to be included, I use DEVmean as the main independent variable for hypothesis 1. In order to ensure that the results cannot be driven by future values, additional models are calculated with DEVretro.

Control Variables
In addition, I include all variables that previous research has shown to also affect the electoral support for new political parties in CEE. An overview of all included control variables and the direction of their expected effect, operationalisation and data sources is given in Table A2 in the appendix.
I include the electoral design (1 proportional, 0 mixed proportional/singlemember district), 19 the electoral threshold and the average district magnitude 20 . In order to measure the potential benefit of a won office, I include a dummy variable for the political system (0 parliamentary 1 semi-presidential) 21 based on Lijphart's (2012: 109) definition. He defines semi-presidential systems as systems that have both a prime minister (i.e. head of government) and a president elected by popular vote. The economic situation is measured by economic growth (annual growth of real GDP in percentages) and the annual inflation rate. Some countries experienced hyperinflation during the observation period which makes its distribution highly right-skewed. The log of inflation rate is therefore included. Alesina et al. (2003) provide an index of ethnic fragmentation regarding linguistic and racial minorities with values between 0 (low fragmentation) and 1 (high fragmentation). The fragmentation of the party system is measured by Laakso and Taagepera's (1979) index of effective number of parties that accounts not only for the number of parties but also for their electoral strength. To be certain that the effective number of parties is not the result of the election under consideration, the value of the previous election is included. The last variable included is the length of democracy counting the number of years between the year of election and the year of the first free and fair election since independence.

Results
The two hypotheses are tested statistically using OLS-regression analysis. In cross-sectional time-series data, observations (here: elections) are not independent of each other and problems of serial correlation and heteroscedasticity can distort the results (Beck and Katz 1995). In order to avoid them, robust standard errors clustered by country are calculated. Because my data includes a higher number of countries than elections for any country, temporal dependence between observationsthe third problem of cross-sectional time-series datashould be less of a problem (Beck and Katz 1995). Results are presented in Table 2.
Model 1, which includes the country-specific corruption level as well as DEVmean, confirms the two hypotheses: The country-specific corruption level has a significant negative impact on the electoral support for new political parties, whereas the deviation from the country-specific corruption level has the expected positive effect. In Model 2, the average district magnitude and the electoral threshold are added to the analysis. Both have missing values for Serbia and Montenegro as well as the elections in 2013. This explains the significant reduction of observations. Nonetheless, the results remain the same. DEVretro is included in Model 4. Again, it shows a significant positive effect. The positive effect of the deviation from the country-specific country level shown in Model 1 and Model 2 therefore cannot be the result of the deviation from future corruption values. 22,23 Countries with a high corruption level are often countries that do not perform well in democracy scores. Although I have controlled for a certain degree of democracy by excluding countries and elections that have not fulfilled the criterion of being democratic (according to the Polity IV score), there is still variance between the countries regarding the quality of democracy, Ukraine being the least democratic country in the sample. A negative effect of the country-specific corruption level could therefore be driven by the lower democratic quality of some high-corruption countries. In Model 3 and 5, the Polity IV score is included. Even though the coefficient has the expected negative sign, the effect is not significant. Also, the effect of country-specific corruption level is not affected by the inclusion of the Polity IV score. 24 Model 6 and 7 replaces the country-specific corruption level by the index of clientelism, Model 7 controlling for the Polity IV score. The coefficient of clientelism is negative and significant. This strengthens the underlying assumption of H2 that new parties win fewer votes in high-corruption countries due to clientelist structures.
Several control variables significantly influence the electoral support for new political parties: Economic growth, the political system and ethnic fragmentation. The latter, which theoretically could have an impact in both directions, has a positive effect on the electoral support for new parties. This is consistent with the findings of Tavits (2008) establishment, and therefore feel attracted to new actors. However, this relationship must be analysed in more depth, in particular because the coefficient is not significant in all models. The political system is the only institutional factor that has a significant effect on the electoral support for new parties in all models. 25 The electoral design shows a negative sign, though only significant in a few models. Although it is theoretically surprising that new political parties are less successful in purely proportional systems, this finding is not novel (see Hug 2001: 8). Hug (2001: 58-9) assumes that the selection process in less proportional systems prevents weak new parties participating in elections, which may again increase the potential of strong new political parties to pool and mobilise unsatisfied voters. The importance of distinguishing between the two counteracting effects of corruption is highlighted in Model 8. Instead of including the time-invariant country-specific corruption level and the deviation from it, only the corruption level per se is included. The effect is negative and significant, but only onesided at the 0.1 level. The large country differences in corruption outweigh the smaller developments within countries. This illustrates that disregarding the two counteracting effects of corruption misleadingly makes corruption seem either unimportant or only disadvantageous for new parties.

Discussion
This article explores the impact of corruption on the electoral support for new political parties. The statistical analysis of all democratic elections between 1996 and 2013 in CEE demonstrates that corruption is important in explaining differences within countries as well as more systematic country differences. The article makes an important contribution by distinguishing between two counteracting effects of corruption: the historically derived, country-specific corruption level reduces the electoral support for new political parties, whereas an increase in the perceived corruption above the country-specific corruption level leads to a loss of trust in the political elite, therefore raising the electoral support for new competitors. The effect of the historically derived corruption is due to corrupt clientelist structures that characterise politics in countries with high levels of corruption and bind voters to the established political parties. By ignoring the theoretical and empirical differentiation, the real effects of corruption would not be observable, thus leading to the mistaken conclusion that corruption is either negligible or only disadvantageous for the electoral fate of new political parties.
Montenegrohaving had no changes in government since the first electionsillustrates the stabilising effect of corruption at an extreme, but Macedonia, Romania and Albania are also countries with high levels of corruption, where new parties have won only a small share of votes. There are, however, countries that contradict this general claim and do show support for new political parties, despite having high levels of corruption. This can often be explained by other factors that matter for electoral support for new political parties, such as the political system, the economic situation or a strong increase in corruption. My models prove only that clientelism explains part of the variance, all other factors being equal. In Ukrainian politics, for example, new political parties are a popular vehicle for politicians coming to power due to a strong presidency and a high degree of personalisation, thus restricting the stabilising effect of clientelism. The victory of the new Bulgarian party GERB, in 2009, shows that increasing corruption (together with a worsening of the economic situation) can also have an impact in countries with high levels of corruption. Although clientelism is generally stabilising, there is no guarantee that the vicious cycle of a highly corrupt and stable party system cannot collapsein particular, if corruption continues to increase and the costs of it become too high. The statistical models in this article try to capture such situations by differentiating between the two types and effects of corruption. However, situations where a rather stable party system collapses despite clientelist structures (as happened twice in Bulgaria) also expose the limitations of regression models assuming log-linear relationships. Italy serves as an illustrative example. Even though Italians had little trust in government, the Italian party system had been highly stable and voter turnout remained high for many decades. The clientelist and corrupt structures had made this possible. Nevertheless, over time the corrupt system did destabilise itself by increasing inefficiency in the public administration and the economy. This eventually led to the 'Mani Pulite' investigations in 1992 followed by the collapse of the entire Italian party system in 1994 (della Porta and Vannucci 1997). A sudden collapse of a formerly resilient and very stable clientelist system is hard to capture adequately by statistical methods assuming only linear effects. More in-depth analyses of such events, complementing the more general results of this article, are therefore needed.
Another question that needs further elaboration arises from the findings: voters can only respond to corruption if they know about it. It is therefore important to identify the circumstances under which an increase in corruption is also perceived as such by the voters. An interesting study by Klašnja et al. (2014) on Slovakia shows that the emergence of a new political party increased the personal evaluation of corruption of Slovakian citizens. Other authors have argued in a similar manner, stating that the stronger the opposition, the more visible corrupt actions are (Grzymala-Busse 2007), and consequently the more people are dissatisfied with politics (Ceka 2013). Therefore, a new party can benefit not only from citizens already tired of corruption on Election Day, but can also, at an earlier stage, have an influence on whether citizens perceive corruption to be a substantive problem.
The findings of this article and the possible follow-up questions also contribute to the wider discussion on the causes for the still very high electoral volatility in CEE. For example, my results suggest that as long as corruption scandals are recurring events in CEE, frustrated citizens will punish the political establishment by voting for new parties and thus the volatility will not decrease. I hope this article will be perceived as an impetus for further studies on the relationship between corruption and party systems in CEE and other regions.
1. Corruption is defined as 'misuse of public power for private gain' (Rose-Ackerman 1999: 91). 2. Bågenholm and Charron (2014) provide a list of all parties politicising corruption in 32 European countries between 1981 and 2011. 3. The only exceptions are parties that are the result of a major party's merger with one or several insignificant parties. 4. The share of votes in the national parliamentary (i.e. lower house, if bicameral) election is considered. Powell and Tucker (2014: 7) argue that choosing the threshold of 2 per cent ensures equal standards for reporting election data across countries, since parties below that threshold are often subsumed under the category 'others'. Due to the 2 per cent threshold, not only genuinely new parties but also previously minor parties (< 2 per cent) are coded as new. Usually, this has only a minor impact on the dependent variable, since such parties have won very little more. In addition, just like new parties, very minor parties cannot yet be counted as part of the political establishment. Theoretically, the consequences of corruption for such 'outsiders' can therefore be predicted to be very similar. 5. According to this rule, neither the Viktor Yushchenko Bloc 'Our Ukraine' nor the Yulia Tymoshenko election blocboth running in the parliamentary elections in 2002 for the first timeshould be coded as new, because both blocs included some minor parties that had won seats previously. However, the leader of both blocs had been Viktor Yushchenko (not affiliated to any party at that time) and Tymoshenko (leader of the main party of the electoral bloc, founded only in 1999), respectively. Due to the high degree of personalisation of Ukrainian politics and the importance of the party's leadership, I decided to code them as new. However, all analyses have also been calculated with a value for Ukraine 2002 not including these two parties (i.e. Ukraine 2002: 4.1 per cent). The results do not change. 6. The majority of the second democratic elections are excluded from the analysis, since data on corruption is not available for this period of time. Elections that are included in the analysis and labelled as second democratic elections in Figure 1 are the second election with a polity IV score higher than 6. They are not, however, the first democratic elections with a less stringent definition of democracy. Lithuania 1996, Moldova 1998, Ukraine 1998 as well as Macedonia 1998 and Slovenia 1996 are the second elections since independence, but all these countries had more or less free and fair elections in 1990. Their characteristic as second elections might still have an additional impact on the share of votes of new parties (in particular, the dissolution of Sajūdis in Lithuania and the formation of a new communist party in Moldova). All analyses presented in the article are therefore also calculated without these elections. The results do not change. 7. Only Mainwaring et al. (2009) incorporated corruption in their model. They found a significant positive relationship. However, this result has to be interpreted with caution, because Mainwaring et al.'s (2009) worldwide sample includes not only developing but also developed democracies. The latter have both more stable party systems and lower corruption levels than developing democracies. Whether corruption also explains the differences within developing democracies (e.g. CEE countries) has therefore not yet been answered. 8. Because of its clandestine nature, the actual corruption level is not automatically the perceived corruption level. Citizens can only be responsive to the latter. For the empirical analysis, an indicator for perceived corruption is used. 9. The corruption level of the previous year accounts for more than 95 per cent of the variance in corruption level. 10. Data derives from the Control of Corruption Index which is part of the Worldwide Governance Indicators (WGI) provided by the World Bank. 11. Clientelism is defined as 'the trade of votes and other types of partisan support in exchange for public decisions with divisible benefits' (Piattoni 2001: 4). 12. 'Report on the misuse of state resources and public authorities in the 2012 parliamentary election campaign', http://www.mans.co.me/en/about-mans/publications/#sthash.gQSfIQpT.dpuf (accessed 15 May 2015). MANS is a local partner of Transparency International. 13. For a deeper discussion see Holmes (2006: 28). 14. The country-specific corruption level and the measure on clientelism by Kitschelt (2013) correlate with r = 0.58 (significant at the p ≤ 0.05 level). The main discrepancy derives from Hungary being a country with relatively low corruption (although increasing over time) but high clientelism. 15. An election is included if the country's democracy score in the Polity IV Project is 6 or higher.
The only exception is Georgia 2008 where elections had been held during the rising conflict over the secession of the regions of Abkhazia and South Ossetia (see e.g. http://news.bbc.co. uk/2/hi/europe/7411857.stm, accessed 18 June 2014). 16. I use log (total share of votes +1), since the distribution of the y-values is highly right-skewed. 17. Since the CPDS II has not been updated since 2008, I updated the dataset based on the European Election Database by the Norwegian Social Science Data Services (NSD) and the dataset Parties and Elections by Wolfram Nordsieck (http://parties-and-elections.eu/). 18. Another reliable data source is the Corruption Perception Index (CPI) by Transparency International (Tavits 2007). However, Transparency International itself states that the index should not be used for comparisons over time (http://www.transparency.org/cpi2011/in_detail#myAnchor6, accessed 18 June 2014). It should thus not be used in a cross-national time-series analysis. Nevertheless, the correlation coefficient between the CPI and the WGI is r = 0.92. 19. No country in the sample uses the pure majority rule. 20. Due to a highly right-skewed distribution and following others (see e.g. Andrews and Bairett 2014; Powell and Tucker 2014) I include the log of the average district magnitude. 21. No country in the sample has a pure presidential system. 22. The number of observations of a model that includes all control variables (e.g. also log_district and the electoral threshold) and DEVretro is too small to have meaningful results (N = 49). However, excluding all non-significant variables (in order to increase the degrees of freedom) and including log_district and the electoral threshold does not change any substantial results.
23. Additional calculation with a short-term perspectivei.e. measuring the difference in the corruption level from one election to another or the difference between the average score of two subsequent legislative periodshave been conducted. None of these indicators have shown a significant effect. This confirms the theoretical assumption that voters include long-term developments in their evaluation of corruption performance. This is also confirmed when including a simple trend dummy (1 positive trend; 0 negative) in the model. It shows that new political parties have won more votes in countries where the general development of corruption over time is positive. However, this does not denote that the trend over time is the only driver of the positive effect of DEVmean and DEVretro: having excluded all countries with a positive trend, these variables still have a significant positive effect. One can conclude that whether corruption is higher or lower matters; but only under consideration of the general state of corruption that changes the meaning of up-and downturns between the different countries at different points in time. 24. Since the correlation between the country-specific corruption level and the Polity IV-score is r = -0.65 the insignificant coefficient of the latter could be a result of multicollinearity. However, excluding the country-specific corruption level does not change the resultthe coefficient remains insignificant. 25. In Model 8 the effect is significant one-sided at the 0.1 level.