I've found my optimum categorical level; how do I drop the rest during design augmentation?

Brian_Pimentel

Let's say I've designed my initial main effect screening experiment, where one of the factors is a 3-level categorical (A, B, C). 

 

After the first set of experiments, I'm fairly certain that level B is my process optimum decision and want to augment the design to get two-factor interactions, etc. How do I restrict the design augmentation to only explore the B level? Using "Hide and Exclude" leads to a big loss of information for the other factors, which I'd like to avoid.

 

Thanks!

Victor_G

Hi @Brian_Pimentel,

 

Welcome in the Community !

 

May I ask you more details about the goal of your study (optimization/prediction, explanation, both ? ...), the number of factors, the number of runs, design and results you already have ?

 

It looks like you're doing a DoE to "pick a winner", for example to choose the best level in the 3-levels categorical factor. But is this factor significant in the analysis of main effects ?

• If yes, I would highly recommend to augment your design, keep using this categorical factor and look at 2-factors interactions. You may have some significant interactions between this categorical factor and other factors, so level B may not be always the optimal choice in presence of interactions. 
• If no, you may simply augment your design (in menu "DoE", "Augment Design") without using this factor in the augmentation panel. As an example, I used the JMP dataset "Algorithm Data" and augment the design on all factors except the 3-levels categorical factor "Algorithm" :

Algorithm won't be a factor anymore in my design, so I can simply set the level of this factor and enter the level chosen in the corresponding column. 

 

No matter how you're doing the augmentation (with or without the 3-levels categorical factor), I would also recommend to check the option "Group new runs into separate block". More info about the "Augment Design" platform is available here : Augment Designs (jmp.com)

 

I hope this answer will help you,