Suchergebnisse

ABSTRACTPast research in modeling human judgments has been accompanied by continuing debate as to the necessity and effectiveness of using "configural" models, vis‐a‐vis"first‐order" models, to represent complex decision processes. The resounding power of first‐order models adequately to represent apparently configural processes repeatedly has been demonstrated to the frustration of those researchers who intuitively feel that man decides in a complex and configural manner. This paper presents the theory that the apparent weakness of configural models may be attributed to the assumption that interaction effects are "continuous" phenomena when in fact they are "discrete" or local interactions.The definition of subspaces of the predictor set, over which "local" first‐order hyperplanes may be used, is investigated as a viable means of representing "discrete" interactions while preserving some of the parsimony of the "continuous" formulations. The application of the Automatic Interaction Detection technique to define subspaces and local models is attempted with positive results for an installment loan officer. Then a comparison with the results from a recent study which used various "continuous" formulations shows a definite superiority of the "local" modeling approach.