Turning out but not voting: invalid ballots in post-communist parliamentary elections

Abstract

Large numbers of invalid votes in elections can present a threat to the legitimacy of democratic systems. Focusing on the post-communist countries of Central and Eastern Europe, three theoretical models explaining the incidence of invalid votes are tested using aggregated data from all legislative elections. While the protest behavior powerfully shapes the incidence of invalid votes in less democratic countries, complex electoral systems and concurrent elections are also associated with high levels of invalid voting (indicating voters’ error). Conversely, we find that most socioeconomic variables are weak predictors of invalid voting while controlling for institutional and political variables in the region. The study has policy implications that the simple electoral design should be preferred.

Appendix: Turning out but not voting: invalid ballots in post-communist parliamentary elections

Alternative model specifications and robustness tests

The data file and the STATA15 and R code are provided online: Lysek, Jakub, 2019, “Replication Data for: Turning out but not voting: invalid ballots in post-communist parliamentary elections,” https://doi.org/10.7910/DVN/HIIMRK, Harvard Dataverse.

It contains further analysis and tests with supplementary comments and justifications.

Given the structure of the data we have also run hierarchical linear models. The ICC is 35.66% (variation given by differences between groups). The results of the models are almost identical to the model presented in the main Table 2. Although the Hausman test (Prob > chi2 = 0.99) justifies the random effects specification (or random intercept model), we decided to present fixed effects models to account for unit heterogeneity as proposed by Wilson and Butler (2007). Fixed effects models serve as good robustness checks and are frequently employed in this regard.

We have tested several control variables which we have decided to disregard in the final model. For review of the variables used in models of invalid voting in the literature, see Kouba and Lysek (2019). we have tested the effect of electoral turnout as an independent variable but with no consistent and robust results. We also considered including a measure of ethnic fragmentation to the models (see Martinez i Coma and Werner 2019). The variable is not correlated with invalid votes. However, in some models the regression coefficient is negative. Yet the effect of this variable is highly model dependent. After deep inspection, the variable is correlated with socioeconomic variables such as GDP per capita. The validity of the indicator by Alesina et al. is arguably low. For example, the ethnically homogenous Czech Republic scores 0.322 while Slovakia with a large Hungarian minority scores 0.253. We also did not include the indexes of corruptions (Transparency International, V-DEM index) or The Worldwide Governance Indicators (WGI) by World Bank because they are all highly correlated with GDP per capita and Freedom House score (Fig. 3).

Fig. 3

The correlation matrix of socioeconomic variables. Note: Taiyun Wei and Viliam Simko (2017). R package “corrplot”: visualization of a correlation matrix (version 0.84)

We also had to consider the necessary trade-off between the number of available cases and the number of variables. We therefore controlled for as many relevant variables as possible while keeping the parsimony of the statistical model. We have run models with the key variable only as well as various models controlling for other variables that we decided to disregard in the final model. The model with the key variables provides substantively similar results.

Time trends

Although the share of invalid votes was highest in most countries at the beginning of the transition, we observe a gradual decline afterward and then stabilization (with some exceptions). The tentative explanation is that citizens socialized during the communist regimes (that in practice enforced compulsory voting) were used to turn out, yet in the democratic elections they opted out by casting invalid ballots. With newer generations of voters, larger shares of the electorate stayed at home instead. Alternatively, as voters gained experience with the electoral process, the share of invalid ballots caused by error gradually declined. These two explanations may operate simultaneously. In general, we find that the time trend is not a very strong. (Variable Year is not significant in the models.) The changes are instead attributable to the substantive independent variables (Figs. 4, 5).

Fig. 4

Invalid votes in time by country

Fig. 5

Invalid votes in time

Interaction effects

We also tested various interactions that have been put forward by the existing literature. For example, the margin of victory and parliamentary fractionalization (Uggla 2008), ethnic fragmentation and corruption (Martinez i Coma and Werner 2019), concurrence of elections and literacy rate (Kouba, Lysek 2016), or interwar democratic tradition with time that explains why countries with a historical democratic experience exhibited fewer invalid votes from the beginning of the transition (e.g., the Czech Republic, Slovakia) contrary to countries without prior democratic experience (e.g., Romania, Bulgaria). Only two interactions were significant in some of the various model specifications: the concurrence and literacy rate and interwar democratic tradition and time. The codes for the models are provided in STATA15 do-file (Figs. 6, 7).

Fig. 6

Marginal effect of concurrence on invalid votes as the literacy rate changes

Fig. 7

Marginal effect of the interwar undemocratic tradition on invalid votes as a year of the election changes. Note: The interwar democratic tradition is an index ranging from 1 to 10 as the least democratic country in the interwar period. The index is computed as the year of the breakup of democracy (Berg–Schlosser, Mitchell, 2003: Authoritarianism and Democracy in Europe) within the period of 1918 and 1939 and transformed to a 1–10-point scale