nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

DOE how to force a given combination?

anne_sa

Hello everybody,

I use the Custom Design platform and I would like to include a specific combination in my design (which correspond to a "control", well-known combination of factor levels).

Is there a way to do that during the design creation phase or should I add it afterwards in the table (but in that case the design could be less balanced or less optimized...).

Thanks in advance for your inputs!

Accepted answers

Dan_Obermiller

Or, how about creating a data table with the one control run and then use Augment Design to add runs to the control run? That would be optimized for all of the runs.

All comments

P_Bartell

I'm the simple type...so I'd just add the treatment combination to the table as long as it doesn't exceed your budget for # of runs. If it does, well then the question is 'Which treatment combination do I eliminate?" Any way you do this you can still use the Evaluate Design Platform to compare performance measures.

But this addition of one "control" treatment combination adds another layer...since the optimal design is model driven, what does including this new treatment combination do to your ability to estimate effects? If it causes more trouble than it's worth, you may want to exclude the treatment combination from the analysis. When I worked in industry we routinely included what you called a "control" in the experiment event as a means of reassurance that some nuisance variable(s) didn't rear their ugly heads during experimental conduct or testing/measuring of responses. If the results of the "control" were close to what we expected practically, then we were more confident that the real experimental design of interest was not struck by the gremlins of chance.

Mark_Bailey

I am even simpler than @P_Bartell.

I wonder about designing the experiment without the control runs, running the experiment with the control runs, and then analyzing the data in the following way. Exclude the control runs when fitting / selecting the best model. Predict the response for the control runs. Then use only the control runs in an equivalence test based on the prediction.

P_Bartell

Works for me. Only issue I see is the original poster seems to suggest including only one control run.

Dan_Obermiller

Or, how about creating a data table with the one control run and then use Augment Design to add runs to the control run? That would be optimized for all of the runs.

anne_sa

Thanks a lot for your your answers!

I selected the post of @Dan_Obermiller as "Solution" since it's indeed a nice way to optimize the design while including this run. But I will also keep in mind the approach suggested by @P_Bartell and @Mark_Bailey!