Do Politics in Europe Benefit from Politicising Corruption?

Abstract

In this article, two unexplored trends in European electoral politics are highlighted. Using newly collected data the article tracks the politicisation of corruption in electoral campaigns from 1981 to 2011, an electoral strategy that has been increasing over time in most European countries. It then empirically tests two aspects of this campaign strategy. First, what are the factors that are systematically associated with a party’s decision to politicise corruption? Second, what are the electoral effects in terms of relative vote share for parties that politicise corruption? Using an original data-set that employs multi-level data (parties nested in countries) the results demonstrate first that politicisation of corruption occurs systematically more often among established parties from the main opposition, new parties and parties on the political right, and occurs as a function of country-level corruption, district magnitude and public party financing. Second, it is found that the main opposition and new parties that use such a campaign strategy make significant electoral gains relative to the previous election compared to parties that do not politicise corruption. Yet gains are offset in low-corruption countries. The findings demonstrate salient implications for research on party systems, corruption studies and democratic legitimacy, among other areas of investigation.

Notes

1. In some academic circles, this growing attention among salient political actors has been called the ‘anti-corruption movement’ (see Charron 2011: 70–72 for more details).

2. Some of the parties, especially the new ones, that politicise corruption have been argued to be considered a new type of niche party on a par with Green or Radical Right parties (see Bågenholm 2013a; for a definition of niche parties see Miller and Meyer 2011 and Wagner 2012). However, in this article we do not distinguish between anti-corruption niche parties per se and other parties that politicise corruption, as the exact boundary between the two categories is difficult to draw. We do distinguish between a number of other party characteristics however, such as incumbency, newness and left–right position.

3. The politicisation of corruption differs somewhat between the established, less corrupt democracies in the West and the more corrupt, post-communist countries. In the less corrupt countries, it is usually parties on the far right who employ the strategy and rather sweepingly so, accusing the so-called ‘political establishment’ of being corrupt and in a collusive arrangement with each other. In Austria, specifically, the Proporz system is attacked by the FPÖ in almost every election. In the more corrupt countries, on the other hand, it is rather the mainstream or new parties that politicise the issue, often as a response to a specific scandal, or to propose certain reforms.

4. Since many small, new parties that do not manage to win seats are unfortunately not reported in election data, we elect to take only those new parties that did in fact win at least one seat. Thus we are aware that we are overestimating the impact of new parties broadly speaking, but we do so knowingly and thus our data allow us to interpret the effects of new parties given that they won seats, and not otherwise.

5. Compared with other measures of corruption such as Transparency International’s Corruption Perceptions Index (CPI), or the World Bank Governance Indicators, the ICRG variable is preferred in our case because it extends back to 1984 in most cases while the other two measures only extend as far back as 1996. In elections prior to 1984, we elect to impute the data back to 1981 using 1984 data. While this is not optimal, it does allow us more observations in the analysis, and since corruption is a rather ‘sticky’ variable over time, we are fairly confident that this limited imputation still gives us an accurate picture of corruption during these years.

In addition, readers might be concerned that, because this measure is an expert assessment (rather than a citizen one), it might not match as closely to voters’ perceptions of corruption in their country as a citizen measure might do. First, citizen measures are not as often taken (Eurobarometer, World Values Survey) as the expert assessments are published (annually) and thus many country-years would have to be imputed or omitted. Second, when comparing measures of corruption perceptions/experiences between experts and citizens across European states, the correlation is very high. For example, in a recent survey of 85,000 EU citizens, Charron (2013) finds that the citizen measure of perceptions of corruption in the public sector correlates with the ICRG and WGI measures at 0.85 and 0.86 respectively.

6. We checked the robustness of this measure by also taking 6 and above along with 8 and above.

7. In the case of the latter, we run mixed-effects models for binary response (when politicising corruption is our dependent variable, ‘xtmelogit’ in STATA) and for continuous responses (when it is the independent variable, ‘xtmixed’ in STATA) which explicitly models multi-level, hierarchical data. The models use a nested structure for the country-level clusters and allow random effects for hierarchical factors. A random intercept (country-level) in the models adjusts for different levels of counts among the country clusters. In addition, since several countries have no cases where any party has politicised corruption, we considered alternative estimations such as using a Heckman Selection model that would account for all parties in a country being zero for the entire sample, which we assume would be linked with the level of corruption in the country. However, several ‘low-corruption’ countries have cases of politicised corruption (Austria, Iceland and Finland, for example), so this estimation is not appropriate in this case. Thus we simply control for country-level corruption in most models.

8. A full list is provided in the Appendix.

9. There are most likely many small parties that never won seats that also politicised corruption, which our analysis does not pick up; thus the model necessarily overestimates the effect of new parties on ‘vote share’. However, due to systematic under-reporting of small parties that fail to win seats (along with reporting on their key platform issues), this is unfortunately unavoidable; the results should therefore be interpreted cautiously.

10. Due to the panel data we employ, we check for problems associated with autocorrelation and heteroscedasticity, For example, we run correlations with the residuals of the model over both country-level variables and time variables. We find no significant correlations. An example of this can be found in the Appendix, showing the residuals (Y-yhat) over time from model 4 in Table 2.

11. Graphs were produced in STATA using the ‘margins’ command.