Presenter: @ChristianBille
There were many interesting questions during the session today, answered by @HadleyMyers:
Q: How does the correlation look if the classical design also has the same number of runs (16)?
A: Perhaps the best way to explore alternative designs is to create and compare them using the "Compare Design" platform: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/compare-designs.shtml#
Q: Can you discuss the other Design Evaluation options?
A: Hi, you can find more information about all the features of Evaluate Designs here: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/evaluate-designs.shtml#ww65477
Q: Can you increase power within main screening design instead of switching to another platform (definitive screening design)?
A: Yes, adding runs to a design will also increase the power.
Q: How can we deal with asymmetric design when for example the center point cannot be right in the middle (due to testing/equipment etc. limitations)?
A: You may be able to use "Factor Constraints" in Custom Design for this purpose: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/define-factor-constraints.shtml#
Q: If I do not want to increase run number, can we eliminate some low-power terms in order to increase power for other terms?
A: In general, yes that will work (assuming the terms were not orthogonal in the original design).
Q: Can you elaborate on the ALT optimality criterion, and whether we should choose D-optimal or another design?
A: Very simply, D-optimality is more suited for screening experiments, and I-optimality is more suited for optimization experiments. You can find some information here, although it is very technical: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/optimality-criteria.shtml. Probably the best way to understand these would be to create designs using different optimality criteria, and then comapre them using the "Compare Designs" platform: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/compare-designs.shtml#
Q: Classic stat books does not talk about Definitive Screening. Am I missing something?
A: DSD was invested in 2011, any literature pre-dating that will have no mention of it. You can find more information about them in the JMP Help files: https://www.jmp.com/support/help/en/16.0/index.shtml#page/jmp/definitive-screening-designs.shtml#
Other information: https://community.jmp.com/t5/JMP-Blog/Proper-and-improper-use-of-Definitive-Screening-Designs-DSDs/b...
https://www.wiley.com/en-us/Optimal+Design+of+Experiments%3A+A+Case+Study+Approach-p-9780470744611"
Q: Is there an advantage of using a Definitive Screening Design over of a Custom Design?
A: The advantage is that you can fit interaction and quadratic terms with fewer number of runs than the Custom Designer would allow, assuming the specifics of your situation allow you to use DSDs. You can find information about when and when not to use them here: https://community.jmp.com/t5/JMP-Blog/Proper-and-improper-use-of-Definitive-Screening-Designs-DSDs/b...
Q: Do you provide models only based on Definitive Screening Designs? Or do you screening only for defining critical factors and then another DoE for modelling the interactions?
A: DSDs can detect main effects, interactions, and quadratic terms. If, after analyzing a DSD, it is desired to include higher-order terms (e.g. cubic terms), or if the initial a-priori model includes higher-order terms, DSD is no longer appropriate and the Custom Designer should be used.
Q: What is the confounding in Definitive Screening Designs between quadratic and interaction terms?
A: The defining feature of DSD is that the main effects are completely unbiased with the interaction and quadratic terms. However, there is some confounding between interaction and quadratic terms. This can be decreased by adding runs (possible within the design dialog window).
