Classical Design Structure of Nonregular Designs from the Custom Design Platform
Mark Johnson, PhD, Professor – University of Central Florida Bradley Jones, PhD, Co-Author – SAS
Custom Design in the DOE platform of JMP produces nonregular designs with desirable projection properties (Leoppky, Sitter and Tang, 2007). Jones and Montgomery (2010) further investigated nonregular designs to identify designs that allow for unambiguous estimation of main effects and a few two-way interactions. Adoption of these partially confounded designs by a statistical community accustomed to fully confounded designs could be enhanced by demonstrating their classical design features. For example, the recommended nonregular design for six factors in 16 runs can be shown to arise from a 24 full factorial in factors A, B, C and D with additional factors introduced via the generators E = ½(AC+BC+AD-BD) and F = ½(-AC+BC+AD+BD) to yield the principal fraction. The remaining 26 two-level, non-regular orthogonal designs for 6-factors in 16 runs will be examined from the classical design perspective. Finally, an example involving the design of an experiment to assess arson fire patterns will be described in which the aforementioned nonregular design is ideally suited.