Projective three-level main effects designs robust to model uncertainty

This paper is concerned with designing experiments under the assumption that an analysis strategy is used that considers interactions in addition to main effects. A criterion which averages an approximation to A_s-efficiency over lower-dimensional projections of the design is introduced to compare designs. A columnwise design procedure is used to construct three-level designs for six factors in 18 runs. The projection efficiencies of these designs are explored and compared with designs obtained from the L₁₈ orthogonal array. Results are also given for designs with 14 and 17 runs.