Graphical Approaches for Planning Experiments - High Resolution Linear Graphs for Three - Level Designs

The lack of consideration for statistical properties in Taguchi's three-level linear graphs is rectified. We propose a new set of linear graphs for the three-level orthogonal arrays according to the maximum resolution criterion. In the presence of two-factor interactions however, the serious bias of all the estimated effects as well as the estimated variance shows that these designs should not be employed. The various alternative designs are discussed.