Analysis Methods for Supersaturated Design: Some Comparisons

Supersaturated designs are very cost-effective with respect to the number of runs and as such highly desirable in many preliminary studies industrial experimentation. Variable selection plays an important role analyzing data from supersaturated designs. Traditional approaches, best subset variable stepwise regression, may not be appropriate this sit- uation. In paper, we introduce a procedure screen active effects SSDs via nonconvex penalized least squares approach. Empirical comparison Bayesian se- lection approaches is conducted. Our simulation shows that non- convex method compares favorably approach proposed Beattie, Fong Lin (2001).