IDENTIFYING ACTIVE CONTRASTS BY STEPWISE TESTING

Unreplicated fractional factorial designs are frequently used in indus trial experimentation and are characterized by the absence of degrees of freedom for estimation of the error variance. This article presents step-down and step-up testing procedures that efficiently identify th e active contrasts (significant effects) in such experiments, and simu ltaneously provide an accurate estimate of the error variance. The met hods allow the user to restrict the experimentwise error rate to a pre specified level. Use of the new methods is illustrated in terms of sta ndard examples from the literature. Simulation shows that a version of the step-up procedure outperforms its competitors.