The analysis of voting behaviour is core subfield of political science. Rival theories posit different social and cognitive mechanisms in determining voters' specific vote choice. For instance, one common method from an issue voting perspective is to have respondents' place themselves and political parties (or candidates) on the same underlying issue scales and to then use various metrics -that assume different cognitive mechanisms- to compare levels of voter-party congruence. In this paper, we compare the performance of traditional approaches with methods that draw their inspiration from social computing. Specifically, we compare the performance of collaborative filtering - which uses item ratings from like-minded users to make predictions about new users - with issue voting approaches. Furthermore, in doing so we introduce statistical learning techniques that are more commonly used in computer science to the broader field of political science.