Criteria sensitivity analysis: a new approach to parameter sensitivity

The problem of parameter sensitivity is not well defined. This paper aims to make it well defined. This is done by requiring the modeller to provide a performance indicator l(X,t,β) which is a measure of how the behaviour of the system is to be judged. This forces the modeller to be precise about what he/she regards as acceptable behaviour. Mathematically we define behaviour to be acceptable at β if l >0 for that β, and we define the system as sensitive if the set of acceptable β's is small. The analysis which is required is termed criteria sensitivity analysis. This method is applicable to many different kinds of models. The basic ideas of criteria sensitivity analysis are presented here. A static model is used to demonstrate major points. The power of this new approach is illustrated by an application to a dynamic model.