Delta Approximation: Parameter Identification for Simulation Models of Human Behavior

There is much concern over the implications of increasing population growth and resource consumption. It may be that traditional academic disciplines have not yet reoriented their goals to deal with this question in a completely satisfactory manner. With the proliferation of computers and computer technology, it is inevitable that some researchers would attempt simulations of long-term human behavior with respect to environment and resources. The great weakness of these ``simulation models'' is that they can never be validated and, therefore, cannot inspire unqualified confidence. Their great strength is that they are summaries of expert opinion and could serve as a focus for multidisciplinary research. The identification of parameters for such models is the main theme of this paper. It is shown that parameter identification, for the proper choice of a ``goodness of fit'' criterion, can be a rapid, always convergent process. This same criterion usually allows identification to occur independently for each state equation. Independence could be especially conducive to multidisciplinary human behavior model building. The technique is illustrated with an identification for a nonlinear, discrete-time simulation model (17th order) of the use of land and energy in California. This model has 80 parameters, which were identified with 150 data points and a few minutes of computer time. Only the hypothetical dynamic relations for agricultural variables need be rejected in this preliminary effort.