483916_en-US

liliconcom

A-Optimal designs generated by default in Custom Designer

Hey folks,

I generated a design in 16.2 for a customer discussion, and was surprised that the custom designer gave me an A-Optimal design. I looked in the documentation and couldn't find any reference to this behavior, although a google search led me to this 2020 JQT paper from Brad and colleagues. 

Is this documented anywhere?

Re: A-Optimal designs generated by default in Custom Designer

I suspect the documentation is not consistent with how the Custom Designer works in 16.2. They indicates D-optimality is the default unless there are quadratic effects. Using Brad's paper that you sited above (if you're an ASQ member, you can access it. Might be available via the SAS library, too), I found a 4 factor main effects + 2fi with 11 runs that produced an A-optimal design by default. Using Evaluate Design, the A-Optimal appears to do a better job in most criteria except D-optimality (particularly average correlation). I'm attaching the designs I created (I forced the D-optimal).

Re: A-Optimal designs generated by default in Custom Designer

This question comes up in JMP Education a LOT. The default in 16.2 is D-optimal with two exceptions. If you press RSM, it's I-optimal. If you have all two-level factors and the model has main effects and interactions, it's A-optimal. Documentation is aware that it's undocumented and they're fixing it.