Two-stage factor screening procedures using multiple grouping assignments

A generalization of the group screening technique for finding the non-negligible factors in a first order model is presented. In conventional screening designs, each factor is assigned to a single factor group, and the sum of effects associated with each group is estimated in the first stage. The new designs assign a factor to multiple groups in the first stage. An individual effect is estimated in the second stage only if each group to which the corresponding factor is assigned is “active” in the first stage. The performance of these designs is compared to conventional group screening designs for cases in which the direction of each factor, if active, is assumed known a priori and measurement error is negligible.