Modeling proximity and directional decisional logic: what can we learn from applying statistical learning techniques to VAA-generated data?

Voting advice applications-(VAA) generated data provide an ideal data source for testing competing theories of voting behavior. To that end, this paper focuses on comparing metrics of voter-party ideological concordance based on rival theories of issue voting (proximity and directional theory). Classification performance of the competing models, in terms of correctly predicting party choice, is evaluated in diverse cross-national settings. Drawing on the EUvox dataset (a VAA for the European Parliament elections in 2014) statistical learning techniques are used to model the decisional logic of voters in high- and low-dimensional policy space. The results show that statistical learning methods can improve classification performance significantly and that how dimensionality is modeled affects the performance of competing issue voting models