A Case Study Illustrating a 10-Step EDA Procedure for Analyzing Two-Level Factorial Designs

• Activity Number: 164
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): James J. Filliben*+
• Companies: National Institute of Standards and Technology
• Keywords: Exploratory Data Analysis ; design of experiment ; graphics ; factorial designs ; two-level designs

Abstract

With increased appreciation--by R&D and industry alike--of the virtues of two-level orthogonal factorial designs, there comes the additional challenge of how to analyze such designs efficiently, systematically, and thoroughly. Least squares factor effect estimates are, of course, computationally trivial, but there is much more insight about the physical phenomenon under study that can be gleaned from these designs beyond the usual estimates, uncertainties, and F-test tail probabilities. A 10-step EDA battery of graphical techniques has been developed to facilitate the analysis of two-level full and fractional designs. This battery augments such classics as the Daniel half-normal plot with a variety of other EDA techniques to provide convincing answers to questions concerning important factors, important interactions, estimated effects, global optima, parsimony, confounding structure, and what settings to use for future experiments. This methodology will be illustrated via a physical sciences application.