Article ID: | iaor2013541 |
Volume: | 7 |
Issue: | 23 |
Start Page Number: | 245 |
End Page Number: | 256 |
Publication Date: | Jan 2012 |
Journal: | International Journal of Services Operations and Informatics |
Authors: | Kawamura Hironobu, Takada Kohei, Furukawa Masayuki, Suzuki Tomomichi |
Keywords: | matrices, statistics: experiment |
A supersaturated design can be used to screen many potentially relevant factors with a smaller number of runs; thus, the resulting matrix has more columns than rows. A three‐level supersaturated design is useful when second‐order effects also need to be determined. However, experimenters using three‐level supersaturated design may actually miss the main effects of factors because the columns of the matrix are not all orthogonal to one another. This paper shows how to assign factors to three‐level supersaturated design that considers low‐order interactions. Specifically, factor assignment that avoids confounding all main effects with all l × l interactions is presented for 27 × 61 and 24 × 28 supersaturated design matrices. The usefulness of the assignment columns created using this method is shown by comparing the presented factor assignment method with the existing one in terms of level of non‐orthogonality.