GUEST EDITORS' INTRODUCTION
Special Issue on Model Selection
A measure of advancement in psychology, as in any other discipline, is the discovery of general laws and principles that govern the phenomenon under investigation. Mathematical models are often used to capture such regularities. In the process the scientist will eventually come to ask the question: How should one decide among a set of competing models of the same phenomenon? This model selection problem is at the heart of the scientific enterprise. Occam's razor has been proposed as a guiding principle for the problem. According to the principle, the model that fits observed data sufficiently well (i.e., descriptive adequacy) in the least complex way (i.e., simplicity) should be preferred. Development of selection methods that formalize this principle, however, turns out to be a nontrivial problem that commands its own field of scholarship. This special issue represents a collection of the state-of-the-art techniques currently available in the field. Utility of these techniques goes beyond the boundary of psychology as they are equally applicable to model selection problems that may arise in any other scientific discipline.
In cognitive science in particular, we have witnessed in the past two decades a tremendous growth in the use of computational and mathematical models to infer the underlying process, yet the development of well-justified methods for evaluating the adequacy of the models themselves has lagged behind. We hope the present special issue will fill this gap by providing scientists with a range of selection methods that may help advance theory development in cognitive science.
The present issue consists of 11 papers contributed by 13 authors whose fields of research include psychology, statistics, computer science, and philosophy, attesting to the ubiquity of the model selection problem. The special issue begins with two motivating papers. The opening paper by Cutting illustrates model selection in cognitive psychology and offers insights from an experimental psychologist's point of view. This is followed by Bamber and van Santen's paper, which addresses model testability and identifiability, another important issue in mathematical modeling. The rest of the nine papers review various selection methods and relevant theoretical issues. Zucchini's paper provides an introduction to model selection and explains the basic issues involved. The five papers that follow present an expert review of each selection method: Bozdogan's paper on the Akaike information criterion and informational complexity criterion, Wasserman's paper on Bayesian model selection, Browne's paper on cross-validation methods, Gru nwald's paper on minimum description length principle, and Golden's paper on model selection test methodology. This is followed by Busemeyer and Wang's paper on generalization criterion methodology, which unlike all the methods described above is designed to