Two factorial designs with quantitative factors are called geometrically equivalent if the design matrix of one can be transformed into the design matrix of the other by row and column permutations, and reversal of symbol order in one or more columns. In this paper, we compare two known methods for the determination of geometric equivalence and propose a modified method based on the “split weights” of the rows of a design matrix. We also propose and evaluate new screening methods for geometric non-equivalence. Most of the paper concentrates on symmetric designs with factors at three levels, but extensions to designs with factors at four or more levels and to asymmetric designs are indicated.
Draper, N.R. Mitchell, T.J., 1968. Construction of the set of 256-run designs of resolution ≥5 and the set of even 512-run designs of resolution ≥6 with special reference to the unique saturated designs. Ann. Math. Statist. 39, 246–255.MathSciNetCrossRefGoogle Scholar
Draper, N.R., Mitchell, T.J., 1970. Construction of a set of 512-run designs of resolution ≥5 and a set of even 1024-run designs of resolution ≥6. Ann. Math. Statist. 41, 876–887.MathSciNetCrossRefGoogle Scholar
Evangelaras, H., Koukouvinos, C., Dean, A.M., Dingus, C.A., 2005. Projection properties of certain three level orthogonal arrays. Metrika 62, 241–257.MathSciNetCrossRefGoogle Scholar
Fontana, R., Pistone, G., Rogantin, M.P., 2000. Classification of two-level factorial fractions. J. Statist. Plann. Inf. 87, 149–172.MathSciNetCrossRefGoogle Scholar