Created:
Feb 8, 2023 11:27 AM
| Last Modified: Jun 8, 2023 9:37 AM(1343 views)
I designed a Scheffe cubic DOE and completed the experiment. When building the model with the data set, I evaluated the p-values. However, some of my colleagues told me that p-values are meaningless for mixture designs. It does not make too much sense to me since not all mixtures have to have none linear factors, and therefore they don't need to be in the model.
I am hoping to get more input from the community. Any comments and suggestions are welcome!
In addition to @louv's suggestion, remember that there is never a reason or justification to eliminate the first-order terms. The effect tests for them are not valid anyway. The only terms worth considering for removal are the higher-order terms which might not contribute much to the model's accuracy or prediction.
In addition to @louv's suggestion, remember that there is never a reason or justification to eliminate the first-order terms. The effect tests for them are not valid anyway. The only terms worth considering for removal are the higher-order terms which might not contribute much to the model's accuracy or prediction.
Just to pile on...Mixture designs are optimization designs. Typically, you have already screened unimportant factors out, so you know the remaining factors are important (no need for p-values). Also, there can be significant multicollinearity. Your primary analysis is the mixture response surface.
"All models are wrong, some are useful" G.E.P. Box
