An empirical estimation of statistical inferences for system dynamics model parameters

For system dynamics simulation (SD) models, an estimation of statistical distributions for uncertain parameters is crucial. These distributions could be used for testing models sensitivity, quality of policies, and/or estimating confidence intervals for these parameters. Assumptions related to normality, independence and constant variation are often misapplied in dynamic simulation. Bootstrapping holds a considerable theoretical advantage when used with non-Gaussian data for estimating empirical distributions for unknown parameters. Although it is a widely acceptable approach, it has had only limited use in system dynamics applications. This paper introduces an application of Direct Residual Bootstrapping (DRBS) for statistical inference in system dynamic model. DRBS has been applied successfully to `The Irish Elderly Patient Delayed Discharge´ dynamic model to estimate empirical distribution for some unknown parameters with a minimal computation effort. The computational results show that bootstrapping offers an efficient performance in cases of no availability of prior information of model parameters.