How good is an optimal design (DOE)?

[jvbraun]

From Brian Miner:

When we use an optimal design (too many factors, some infeasible combinations removed, a model is postulated....) how should the "diagonality" and confounding matrix be interpreted?

*Is a diagonality the best measure to look at and understand how well you can estimate the effects in the model? Are there any rules of thumbs?

*Are the cells of the confounding matrix to be looked at and "large" (how to define this?) coefficients suggest it is difficult to clearly estimate these effects?

• Are there better measures to look at to know if the design is actually "good"

Here was the full question:

http://stats.stackexchange.com/questions/252174/how-good-is-an-optimal-design-doe?noredirect=1

And in case you are curious, this is the paper that the posting was inspired by: http://www2.sas.com/proceedings/sugi31/196-31.pdf

[jvbraun]

That is an awesome question, and I have been thinking a little about how to visualize the confounding matrix.

[rasmusagren]

Any updates on this? In general, does a value of +/- 1 in the confounding matrix mean that the two effects are confounded? Thanks!