JOINT STATISTICS SEMINAR
SPONSORED BY
DEPARTMENT OF EPIDEMIOLOGY AND BIOSTATISTICS
AND
DEPARTMENT OF STATISTICS
CASE WESTERN RESERVE UNIVERSITY
MINIMUM ABERRATION AND MODEL ROBUSTNESS
PROFESSOR CHING-SHUI CHENG
DEPARTMENT OF STATISTICS
UNIVERSITY OF CALIFORNIA AT BERKELEY
Minimum aberration, proposed by A. Fries and W. G. Hunter (Technometrics, 1980) is a criterion for selecting good fractional factorial designs. It is a refinement of the criterion of resolution. A design of resolution r has the property that no k-factor effect is an alias of effects involving less than r-k factors. Among those of the maximum resolution R, a design with minimum aberration has the smallest number of k-factor effects aliased with effects involving R-k factors. In this talk, I will show that minimum aberration is a good surrogate for a notion of model robustness. While minimum aberration is defined in terms of the word length pattern, it can be related to the alias pattern of the interactions. For example, a minimum aberration design of resolution three or higher maximizes the number of two-factor interactions that are not aliased with main effects, and also tends to distribute these interactions among the alias sets more uniformly than do other designs. Some construction results will also be presented.
MONDAY, DECEMBER 16,1996
2:00 PM - REFRESHMENTS
2:30 - 3:30 PM - TALK
327 YOST HALL