The blank ballot crisis: a multi-method study of fraud in the 2006 Italian election

ABSTRACT

Among European democracies, Italian elections have usually seen high levels of protest voting. Then in the 2006 election the number of blank ballots suddenly dropped by one million following the adoption of a new electoral system. This unprecedented change in electoral behaviour, and a margin of victory of 25,000 votes, raised suspicions of electoral fraud. Given that electoral fraud is more commonly associated with voter turnout than with protest voting, it is surprising that the academic literature has neglected this election. A lack of statistical assessment makes this unforeseen event exceptionally suited to the elaboration of a more universal approach to the investigation of voter fraud. For this purpose, this study applies a combination of Benford’s law, distributional tests and regression modelling to the Italian data. Three levels of disaggregation along regional and partisan lines are used to increase the homogeneity in the data. The results are compelling, showing widespread statistical anomalies in the 2006 election, and effect sizes large enough to alter the political outcome. This article concludes that the possibility of regional-level fraud cannot be ruled out, and that the evidence supplements the scholarly work on electoral manipulation in democracies.

KEYWORDS:

• Electoral fraud
• quantitative methods
• democracy
• Italian politics
• political parties

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. See for example in presidential elections, the regional anomalies reported by Lukinova et al. (2011a) in Putin’s favour in 2004, and by Lukinova, Myagkov, and Ordeshook (2011b) in Yanukhovich’s, also in 2004.

2. Note that the opposite argument – more choice reduces the likelihood of protest voting – has also been empirically supported (Moral 2016).

3. For a voter turnout meta-analysis see Cancela and Geys (2006).

4. Consider that here the normal value of the constant α is not 0 or 1, but is slightly higher than each party’s vote share, since in municipalities where protest voting is zero, party vote should be maximized. This makes its predicted deviation in cases of manipulation much harder to pinpoint than in the model for voter turnout.

5. A quantile-quantile test arbitrated whether these two distributions were statistically different. The general trend line is flatter than the prediction line in the left tail of the distribution, indicating that the variable on the x axis (change in blank ballots) is much more skewed to the left than the one on the y axis (change in spoilt ballots).

6. The average variation (2006–2001) in l’Ulivo party shares in Toscana was 1.18%, while the average variation in blank ballots was −2.44%. The ratio between the two is −0.48, which was used as a reasonable benchmark.

Additional information

Notes on contributors

Alberto Lioy

Alberto Lioy is a postdoctoral researcher in political science at the University of Hradec Kralové (Czech Republic). He has published in peer-reviewed journals on topics related to electoral participation, party systems and the politics of the internet.