Pareto plot threshold for multiple management factors in design of experiments

Abstract

Economical, technical and strategic reasons make unreplicated experimental designs necessary and popular in industrial and management settings. This paper investigates the legitimate use of well known methods that appoint effects as significant when they are larger than a defined threshold. We found that established estimation methods, such as Lenth’s method and Dong’s method, are too tolerant and a larger threshold should be used before an effect can be determined as influencing. This comparing analysis was done by using the degree of freedom that is available when the resolution of the design is one less than the maximum value. Using the Pareto Principle we suggest a more accurate threshold estimator for significantly influencing factors and/or interactions based on analysis of variance.

Keywords:

• Design of experiments
• Significant effect
• Sum of squares
• Factorial design
• Experimental error