Comparing the Utility of Pairwise and Feature-Derived Similarity Measures for Generating Spatial Representations of Semantic Concepts

This study compares the relative utility of similarity data gathered using either direct pairwise ratings or feature-derived methods in regards to generating spatial representational models of fifteen well-known semantic categories. We assess both the extent to which the different similarity data sets can be accurately represented by a spatial model (representational goodness of fit), and the ability of the resulting spatial models to predict an external empirical semantic variable (predictive validity). The results indicate that feature-derived similarities obtain a better representational goodness of fit than the pairwise similarities, and that the predictive validity of representations based on common features data is far superior to the predictive validity of representations based on any of the other similarity data sets.