The results of a fractional factorial experiment can sometimes be ambiguous due to confounding among the possible effects. More than one model will often be consistent with the data. We develop a Bayesian method based on model discrimination for designing a follow-up experiment to resolve the ambiguity. The idea is to choose runs that allow maximum discrimination among the plausible models. The method is more general than methods that algebraically decouple aliased interactions and more appropriate than optimal design methods that require specification of a single model. The method is illustrated with several examples of fractional experiments.

This is joint work with Dan Meyer and George Box.