A spatial model of voting behaviour – an empirical study based on Budapest data

Abstract

One condition of the stable and balanced functioning of democratic systems is the active participation of citizens. Considering the political subsystem, civilian activity is primarily, although not exclusively, manifested in participation at elections.

The relationship between participation in the 2010 parliamentary election and the number of votes cast in favour of the winning party was in the focus of our investigations in this study. According to Klimek et al. (2012), who greatly inspired the current research, conclusions can be drawn about the fairness of the election based on the concentration of votes cast for the winning party. This is possible because salient differences observable in the joint distributions of concentration and participation may suggest problems concerning the fairness of the election. Since the chance of salient differences occurring is negligible, these might suggest manipulation. In our study, we argue that a method which ignores spatial effects may be inefficient in the case of auto-correlated error components because the independence assumption is not met, so this method may lead to biased estimations.

In order to answer the research question, we relied on data derived from voting districts in Budapest which were used to create a geo-referenced spatial point database. Data were entered by voting district, then areal data was created with the help of Voronoi tesselation. In order to shed light on the relationship between participation and the votes cast for the winning party, we relied on spatial filtering as suggested by Getis and Ord, which enables the explicit expression of spatial effects.

The non-parametric Getis spatial filtering has been investigated with the help of optimal distances and first-order topological neighbourhood as well, although the method is primarily based on determining distance-based auto-correlations. The approach which took topological relations into consideration did not result in significant differences in parameter estimates, although it had a somewhat better fit (higher R2 and lower AIC value). This calls attention to the problems of applying distance approaches in the case of very unevenly sized irregular configurations.

By the decomposition of variables to spatial and non-spatial components and by the explicit expression of spatial effects, the independence assumption of error components can be fulfilled. As a consequence, it is possible to return to using the traditional Ordinary Least Squares (OLS) method.

The most important finding of our research was that the relationship between voting turnout and votes cast for the winning party in Budapest in 2010 was unjustifiable. Relying on the OLS method and disregarding spatial effects led to serious consequences; it resulted in biased estimations. For this reason, we emphasize the importance of handling auto-correlations properly in the case of spatial data.