A robust analysis for unreplicated factorial experiments

摘要

A formal robust analysis for detecting active effects in unreplicated 2~p factorial experiments and their fractions is proposed. Based on an extensive simulation study, the proposed method is shown to be simple and more powerful than the popular method derived by Lent (1989, Technometrics 31, 469-473). Two types of powers were used as criteria in comparing the proposed method with that of Lenth. Precisely, the newly proposed method provides a redescending M-estimator for scale based on the cos-function. Critical values for the proposed test statistic were empirically computed and fitted with tabulated values of the t-distribution for different sample sizes.