A Bayesian approach to the selection of two-level multi-stratum factorial designs (Q2413609)

In a multi-stratum factorial experiment, there are multiple error terms (strata) with different variances. There are some difficulties in extending the minimum aberration criterion (which is popular for choosing fractional factorial designs in the case of unstructured experimental units) to multi-stratum factorial designs. The authors use a framework for Bayesian factorial designs proposed by \textit{T. J. Mitchell} et al. [Stat. Sin. 5, No. 2, 559--573 (1995; Zbl 0828.62069)] and present theorems on Bayesian A- and D-optimal fractional factorial designs. They derive a surrogate criterion (M.S-criterion) which can be related to the generalized word length pattern. The authors also provide a useful tool for screening out inferior designs to facilitate the search for optimal designs. Applications of such a tool are illustrated in several examples.