COLLOQUIUM

Master's Thesis

Single-Replicate Factorials: How Existing Methods Compare to a New Method

Kristin Oines

University of Wisconsin-Milwaukee
Friday, April 30th, 1:00pm in EMS E423

Due to their lower costs, single-replicate designs are useful in industry as a preliminary study to identify what factors merit further study. Therefore, it is preferable to overestimate which factors are active rather than to underestimate, thereby missing a truly active factor. Pursuing `dead ends' is costly and time consuming, however, so an ideal method will correctly identify truly active effects without falsely declaring truly inert effects to be active.

In this paper, we discuss four existing methods used to analyze single-replicate experiments, as well as a new method proposed by Beder. We compare all the methods using actual data, as well as in a simulation study and consider the results of the analysis.