Two-Level Parallel Flats Split-Plot Designs

Split-plot designs are widely used when some factors are difficult to change, enabling the implementation of larger experiments. When nonregular fractional factorial treatment structures are used, some effect contrasts can be partially confounded with whole plots, rather than being either orthogonal or fully confounded. This is particularly so for some choices of the number of subplot units per whole plot. Our article proposes clear split-plot designs constructed via a parallel flats structure. We show that these designs divide the factorial effects into two orthogonal subspaces, simplifying model selection. Our proposed class of split-plot parallel flats designs are flexible in run sizes, are straightforward to construct, and enjoy additional benefits versus other nonregular and algorithmically-generated split-plot designs that lack this structure.