Analysis of saturated factorial designs

Fractional factorial designs are commonly used for screening experiments in industrial experimentation, allowing the study of many factors using experiments of only modest size. It is possible for example to estimate the linear main effects of 15 factors with only 16 observations. However, such a design is saturated---it provides no error degrees of freedom with which to independently estimate variability---and this raises some interesting problems concerning the data analysis. The most heuristically appealing methods of analysis of saturated designs use the data adaptively, complicating matters further. The talk will be a survey of problems and results concerning the strong control of error rates in the analysis of saturated designs.

About the speaker: Dan Voss received his B.S. in Mathematics (1979) from the University of Dayton and his Ph.D. in Statistics (1984) from The Ohio State University under the guidance of Angela Dean. He has been in the department of Mathematics and Statistics at Wright State University since 1983, where he is a former Acting Director of the Statistical Consulting Center and is currently Professor and Director of the Statistics Program. Dan coauthored a book on {\em The design and analysis of experiments} with Angela Dean, enjoyed research support from AFOSR, has been an expert witness in court, and has published applied and theoretical work in a variety of journals. His current research interests include applied statistics in general and focus on the analysis of sparse factorial designs from a multiple-comparisons perspective.