Prediction as a device to evaluate theory
Given that explanation in the absence of prediction is ‘pre-scientific’ (Schrodt 2014), the strong emphasis on evaluating out-of-sample predictive performance allows us to put the theoretical and empirical model components underlying ViEWS to very demanding tests. When doing so, we will also explore why some statistically significant variables fail to improve predictive performance. One possible reason is over-fitting, which is common when data are sparse (Ward et al. 2010). Our methodological work on model selection and averaging will also shed light on whether predictive performance depends on specification problems. We will also explore how our integrated empirical models speak to the theoretical literature.