Generalized FEAR method to estimate factor effects in two‐level supersaturated designs

Two‐level supersaturated (SS) designs examine more than N SS − 1 factors in N SS experiments. A proper estimation of the factor effects is not evident due to a confounding of the main effects. In this paper, a generalization of the earlier proposed Fixing Effects and Adding Rows (FEAR) method, to estimate effects from SS designs, is presented. Three data sets, examining SS designs ( N SS = 12 or 14), were used to evaluate the FEAR method further. The factor and critical effects estimated from SS designs with the FEAR method are usually similar to those estimated classically from N = 24 or N = 28 screening designs, examining the same factors, when the effect sparsity principle largely is fulfilled, that is very few significant effects occur. It was found that in those cases the FEAR method can correctly indicate the important significant effects. This is not the case anymore when the number of significant effects is too high. However, classically applied methods also fail in these situations. It also was observed that different ( N SS experiments, f SS > N SS − 1 factors) SS designs, which have a different confounding pattern, either can result in good or worse effect estimates.