Bradley Jones, JMP Distinguished Research Fellow, JMP
Screening experiments often require both continuous and categorical factors. Engineers generally prefer continuous factors with three levels over those with two levels because of concerns about curvature in the relationship between a continuous factor and the response.
This presentation provides a method for constructing designs for roughly equal numbers of continuous factors with three levels and categorical factors with two levels. If there are m three level continuous factors and m-1 two-level categorical factors, our designs have 2m runs. Note that m may be odd or even. As for any saturated main effects plan, our designs depend on an assumption of effect sparsity and negligible two-factor interactions (2FIs) to provide accurate estimates of the main effects. With substantial effect sparsity or by not making use of all the columns of the design, it can be possible to identify a large quadratic effect or 2FI. The columns of the two-level factors are orthogonal to the columns of three-level factors. This fact can be used to provide a design-centric analytical approach. When the number of runs is a multiple of eight, all the design columns are mutually orthogonal. This talk provides examples of the designs and shows how to create them.
