Abstract
Unreplicated factorial designs are frequently used in industrial experiments. A commonly used method to identify active effects from such experiments is the half-normal plot. Many formal testing methods have been developed to overcome the subjectivity of using this graphical method. Among them, the Lenth (1989) method is simple, yet powerful, as shown by Hamada and Balakrishnan (1998). In this paper, we propose a step-down version of the Lenth method.
It is compared via simulation with the original Lenth method and with stepwise methods proposed by Venter and Steel (1998). It is shown that the step-down Lenth method is better than the original Lenth method and the Venter and Steel step-down method. The Venter and Steel step-up method controlled by the same experimentwise error rate has more power, but it also has a higher individual error rate. Critical values used in the step-down Lenth method are provided.
Additional information
Notes on contributors
Kenny Q. Ye
Dr. Ye is an Assistant Professor in the Department of Applied Mathematics and Statistics. His email address is [email protected].
Michael Hamada
Dr. Hamada is a Technical Staff Member in Statistical Sciences.
C. F. J. Wu
Dr. Wu is H. C. Carver Professor in the Department of Statistics and Professor in the Department of Industrial and Operations Engineering.