Strangelove: How I Stopped Worrying and Learned to Love A-optimal Designs (2019-US-45MP-281)
Aug 27, 2019 12:45 PM
| Last Modified: Oct 18, 2019 10:36 AM
Heath Rushing, Principal, Adsurgo Andrew Karl, Senior Management Consultant, Adsurgo
Many years ago, a mentor convinced us to approach teaching DOE not as an instructor, but as a motivator to gain disciples, students that would use DOE so effectively and efficiently that others would follow. For years we have used that approach to spread the use of both D- and I-optimal designs in JMP. Then something happened on the way to dinner; A-optimal designs were introduced in JMP 14. While developing examples for these new (in JMP) designs to augment our current teaching material, we found A-optimal designs to be the most adaptable custom design in JMP.
In this talk, we will motivate the use of A-optimal designs. We will start by demonstrating their flexibility: the ability to emphasize some effects (say main effects) over other effects (say interactions) in design selection. Using multiple design metrics, we will then compare an unweighted A-optimal design to other more popular design choices (D- and I-optimal) to show they are not only flexible, but also can perform decidedly better. Augmenting a design with an A-optimal design can produce a superior hybrid design. Lastly, we will demonstrate the use of space-filling designs to compare the utility of weighting schemes for A-optimal designs.