Matching algorithms based on competing theoretical models of issue voting, namely proximity and directional theory, are presented with a view to evaluating their predictive performance in diverse cross-national settings. Both high dimensional models (based on the thirty or so policy items usually included in a VAA) as well as low dimensional models (typically left-right versus conservative-liberal) are tested. A core concern is whether voters' decisional logic is conditioned by contextual factors such as centripetal versus centrifugal party competition. Drawing on recent literature that argues proximity models perform best in less polarised setting whereas directional models work better in a more polarised context, this paper shows how VAA-generated data can be used for evaluating the performance of competing theories of issue voting. Furthermore, it argues that engaging in such analyses can also lead to theoretically grounded improvements in the algorithms deployed by VAA designers.
