An efficient screening methodology for a priori assessed non-influential factors

• Authors: Thomas M. Cioppa
• Venue: WSC '04 Proceedings of the 36th conference on Winter simulation
• Year: 2004

The choice of which factors to choose to vary in a simulation to effect a change in the measure of interest is difficult. Many factors are a priori judged not to effect the measure of interest. There is often no subsequent analysis of whether this judgment is valid or not. The proposed methodology outlines a sequence of limited runs to assess the accuracy of the a priori belief of non-influential factors. These runs can either identify if an influential factor(s) has been omitted or confirm a priori beliefs. Executing these sequences of runs before focusing on the suspected influential factors can contribute to subsequent analysis by confirming that suspected non-influential factors do not significantly impact the measure of interest. An example of the methodology using an agent based model is presented.