cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Browse apps to extend the software in the new JMP Marketplace
Choose Language Hide Translation Bar
Aziza
Level IV

DoE: What to do if some conditions of a design are almost impossible to run?

Hi All,

 

Assume we have 3 factors with 2 levels each. If I understood it correctly, for small number of factors we can consider a full factorial design, in this case would be 2^3 full factorial design. What should we do, if by some reason, 2 out of 8 conditions is not possible to run? Is there any way/concept to handle this situation?

 

Thanks in advance!

Greetings
2 ACCEPTED SOLUTIONS

Accepted Solutions
David_Burnham
Super User (Alumni)

Re: DoE: What to do if some conditions of a design are almost impossible to run?

P_Bartell
Level VIII

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Just to add a bit to @David_Burnham 's advice, if you use the Custom Design platform, you also have flexibility wrt to the # of runs, replication, effect estimation, and other design attributes/characteristics so you'll probably want to give some thought to these as well. For example, the 2**3 full factorial design allows for estimation of all possible main effects, two factor and single three factor interaction effects. Do you really need to estimate all of these effects? Perhaps you don't care about one or more? You can articulate the specific model that you're interested in fitting as you plan the design, then the search algorithm will provide an (probably D?) optimal design tailored to your specific requirements.

View solution in original post

7 REPLIES 7
David_Burnham
Super User (Alumni)

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Aziza
Level IV

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Thank you David, these all are very useful.
Greetings
P_Bartell
Level VIII

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Just to add a bit to @David_Burnham 's advice, if you use the Custom Design platform, you also have flexibility wrt to the # of runs, replication, effect estimation, and other design attributes/characteristics so you'll probably want to give some thought to these as well. For example, the 2**3 full factorial design allows for estimation of all possible main effects, two factor and single three factor interaction effects. Do you really need to estimate all of these effects? Perhaps you don't care about one or more? You can articulate the specific model that you're interested in fitting as you plan the design, then the search algorithm will provide an (probably D?) optimal design tailored to your specific requirements.

Aziza
Level IV

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Thanks for your informative response. This will help me a lot.
Greetings
statman
Super User

Re: DoE: What to do if some conditions of a design are almost impossible to run?

Some questions for a different perspective...Why are those particular treatment combinations impossible?  How do you know they are impossible?Can you provide more context? Are the levels too bold?  Are the levels you set for some factors dependent on the levels of other factors (potentially factors nested within others vs. crossed)?

"All models are wrong, some are useful" G.E.P. Box
Aziza
Level IV

Re: DoE: What to do if some conditions of a design are almost impossible to run?

It was just a question of curiosity, which makes it not easy for me to formulate it in a specific way. But let's assume these factors depend on the levels of other factors. This would be an interesting case.
Greetings
P_Bartell
Level VIII

Re: DoE: What to do if some conditions of a design are almost impossible to run?

@AzizaWhenever I was on teams and we were planning experiments, whether we used a 'catalog of designs' approach (for example, a plain Jane 2**3 full factorial design) or an optimal design approach...whenever we had a proposed experimental design, we'd ask some questions like:

 

1. Will any of these combinations cause a known safety, or otherwise known or suspected hazardous condition that might 'break something' or worse yet, have a potential to cause personal injury or harm?

 

2. Given this design, are there other nuisance factors or experimental constraints we need to consider? If so perhaps we need to add some thinking around blocking (say, we've got to use 2 different suppliers for raw materials...and we want to guard against supplier to supplier variability) or, maybe we've got some 'hard to change' factors and it makes logistical sense to group the levels together (then a split plot approach...a form of blocking should be considered).

 

3. Do we need to consider response testing variability issues for the objects coming out of the experiment?

 

All these answers can influence the final design choice.