Sensitivity analysis for models with multiple behavior modes: a method based on behavior pattern measures

Abstract

Sensitivity analysis of system dynamics models is essentially about sensitivity of patterns of output behaviors to inputs, since system dynamics modeling is behavior pattern oriented. In this study, a regression-based procedure for pattern sensitivity analysis is developed, by defining behavior pattern measures such as equilibrium level, trend, inflection point, or oscillation amplitude. A unique feature of the procedure is that it takes into account the possibility of a model generating multiple behavior modes. This pattern-oriented procedure is next applied to the tipping point project management model and a generic supply line model. These test applications yield sensitivity results that are meaningful, and also consistent with previously available sensitivity information about the parameters of these models. Finally, our pattern sensitivity analysis is shown to be a useful and effective method also for oscillatory system dynamics models, an unsolved sensitivity problem previously in the literature.