AN EMPIRICAL EVALUATION OF HEDONIC REGRESSION MODELS

Abstract

Statistical criteria are poor at evaluating spatial exploratory models of hedonic regression because they are heavily dependent on uncertain assumptions concerning spatial relationships. To resolve this problem, empirical evaluation methods are proposed in this paper. A simple linear model and three spatial models for a housing and land price dataset in Tokyo are studied for illustration. The prediction power of models is emphasized. With a cross-validation technique, the housing and land price at each sample point are predicted, and then, numerical and graphical criteria are supposed using these predicted values and real observed prices. The methods not only help us to select suitable models for a dataset, but also provide an alternative for the significance test of concerned spatial relationships.