Sensitivity analysis by experimental design and metamodelling: Case study on simulation in national animal disease control

Abstract

Simulation is a frequently applied tool in the discipline of animal health economics. Application of sensitivity analysis, however, is often limited to changing only one factor at a time (OAT designs). In this study, the statistical techniques of Design of Experiments (DOE) and regression metamodelling were applied to a simulation model developed to support decision making in national animal disease control. Since the simulation response of interest was censored, we also applied – besides ordinary least squares (OLS) regression – tobit and logistic regression. Furthermore, a comparison was made with analysis based on an OAT design. The metamodel estimated by OLS regression showed reasonable fit, but was not considered a valid approximation of the simulation model. We concluded that logistic regression can be applied if output data are binary, whereas tobit regression is most appropriate when dealing with censored data. Furthermore, we concluded that the DOE and metamodelling approach – compared with a simple OAT design – is more effective in describing the relationships between model input and output in this case study.