Statistical simulation as an effective tool to evaluate and illustrate the advantage of experimental designs and response surface methods

Abstract The most illustrative way to evaluate the benefit of experimental designs is to use them in practice. This includes setting up the designs, generating data and performing statistical analysis on this data. However, testing several different experimental designs in the laboratory is in many cases both time consuming and expensive. As a cost-reducing alternative to this traditional testing, statistical simulations can be used. This paper shows how simulations can be used to evaluate three different experimental designs' ability to describe the interaction between three process variables (surfactants in an oil spill dispersant) and a quality describing variable (the dispersant effectiveness). Two simplex-lattice designs and one traditional change `one-variable-a-time' (OVAT) approach are compared and the ability of the first two designs to find an optimum composition at a lower cost (fewer experiments) is demonstrated. Furthermore, simulations are used to evaluate the increased cost of replicate measurements against the increased precision. This paper shows that statistical simulation is a powerful tool to evaluate and illustrate the benefit of using designed experiments.