Introducing confidence bounds and confidence levels into iterated fractional factorial design analysis
Hajas, Wayne C.
This report is further development of a methodology known as Iterated Fractional Factorial Design Analysis (IFFDA). IFFDA uses experimental designs to identify the influential parameters in systems with many (hundreds or thousands) of parameters. At its previous stage of development, IFFDA gives no well-defined measure of the reliability of the results. This report includes enhancements to assign confidence levels and confidence bounds to the estimates produced by IFFDA. These enhancements can be incorporated into the application of IFFDA and the result is a more objective analysis. Two examples are discussed. The first is small and contrived and used to illustrate the capabilities of IFFDA in previous applications. A larger system is required to demonstrate how the confidence bounds and confidence levels can be estimated and a computer model known as SYVAC3-CC3 is used. SYVAC3-CC3 was chosen because it is well known (Goodwin et al 1994 for example) and yet has enough system parameters ($\sim$3300) to be non-trivial. Two strategies are given for incorporating confidence levels and confidence bounds into IFFDA. The first assumes that no expert knowledge of the system is available and the second incorporates expert knowledge into the analysis. In the SYVAC3-CC3 example, the enhanced methodologies gave results that are consistent with the understanding of the system. Results are even more satisfactory when expert knowledge is incorporated into the analysis.