Too Saturated: When Too Many Factors Are Too Much in a Supersaturated Design (303064)

Keywords: Sparsity, Power, Supersaturated Designs, Design Evaluation

Evaluation of supersaturated designs is not well quantified. Proposed functions of the correlations between design variables are not sufficient indicators of the value of the design and have little to no practical meaning to the scientist customer. Power, the probability of detecting effects without having to specify a particular model, is the ultimate indicator of design value and has meaning to the scientist. Using simulation and appropriate analysis techniques, it can be shown that the power of detecting main effects in a supersaturated design setting is a function of the number of independent variables and the number of experimental runs. In addition, power is also dependent upon the degree of sparsity. A better quantifiable relationship between power, design size and scope, and sparsity is estimated in our empirical model developed from simulated data. This model enhances supersaturated design evaluation and improves the planning of the scope and cost of such designs.