The I-optimality criterion is often used when creating response surface designs in JMP's custom design mode (the default when clicking RSM in custom design). The I-optimality criterion is a criterion used to search for a design that minimizes the average value of prediction variance within the design space. While it is useful in the phase of searching for optimal conditions, it can sometimes unintentionally result in designs with poor G-optimality. This is problematic because the conditions selected as optimal may not be very accurate. To address this issue, we considered using the A-optimality criterion. This is because using the A-optimality criterion when creating response surface designs not only produces designs that are quite close to those created using the I-optimality criterion, but also because the "parameter weights at A-optimality" allow for considerable flexibility in controlling the properties of the design. In fact, it has been shown that by utilizing parameter weights, it is possible to create response surface designs with improved G-optimality, and that it is possible to solve some of the problems that can arise when using the I-optimality criterion.
