An Experiment in Sensitivity Analysis On an Uncertain Model

Abstract

Experiments on a compartment model into which very large errors have been deliberately introduced show that sensitivity analysis continues to provide valid information about the relative importance of parameters long after other aspects of realism have been destroyed. Thus, sensitivity analysis may be the first area in which a mathematical model yields meaningful information as it progresses through its formulation and development stages, and data-base refinement may be started relatively early in the modelling process.