USING PARAMETER SENSITIVITY AND INTERDEPENDENCE TO PREDICT MODEL SCOPE AND FALSIFIABILITY

One important criterion for a model's utility is its scope, the abilit y to predict a wide range of results. Scope is often difficult to asce rtain without extensive data fitting, For example, J. E. Cutting, N. B runo, N. P. Brady, and C. Moore (1992) compared 2 models of perceived visual depth by fitting many data sets that were arbitrarily generated from underlying functions. They then defined scope as the number of f unctions a model could account for. We present an alternative techniqu e for scope evaluation that is based on analysis of the behavior of a model's parameters and does not require extensive data fitting. The te chnique examines the ratio between the overall interdependence among m odel parameters and their sensitivity, which we show to be inversely r elated to a model's scope.