Seminar 13th May 2010 2:15 p.m. 54/10B
The design, analysis and practical use of supersaturated experiments
Chris Marley
University of Southampton
- Categories
- Complex Systems
- Submitter
- Stefanie Biedermann
In screening experiments, it is often necessary to investigate a large number of factors to establish which have a significant effect on the response of interest. If it is not possible to carry out a large number of runs, a supersaturated design may be used. This is a design in which the number of observations is less than the number of parameters to be estimated. As a consequence, not all parameters can be estimated simultaneously.
There is much work in the literature on how to design and analyse such experimental plans, and it is widely believed that their effectiveness in detecting active factors relies on the assumption that, in reality, there are only a small number of dominant factors – a concept known as effect sparsity.
This talk presents some interesting results on the effectiveness of supersaturated designs and associated analysis methods for different numbers and sizes of active effects. A new class of criteria for supersaturated designs based on measures of multi-collinearity among subsets of the factors is also discussed.