TABLE OF CONTENTS It is still a widely held view that the scientific approach to designing experiments is to investigate one system parameter at a time. That this is badly wrong is demonstrated below. This chapter introduces statistical strategies for the design of experiments for the systematic investigation of complex systems. It is not intended as a comprehensive account of experimental design, but it will give the reader a useful approach to studies requiring investigation of several system parameters. The 'factorial designs' covered here will be useful for method development, troubleshooting, and robustness/ruggedness testing during method validation.
The ideas are introduced in the context of the simplest possible extension of the discussion of Chapter 4: two system parameters are to be studied and two settings for each will be investigated. The approach will then be extended to studying the effects of varying three or more system parameters. The 'fractional factorial designs', which are used for ruggedness testing, are also discussed in this chapter. These allow many system parameters to be investigated simultaneously, without requiring excessive experimental runs. An example in which twelve parameters were studied using only sixteen runs is described.
Two key aspects of multi-factor studies are the need to acquire the maximum information from a given amount of work and the importance of discovering whether varying one system parameter influences the effect of varying a second parameter. These are discussed in Section 5.2. The importance of graphical analysis as a method for validating statistical assumptions was a feature of...
