How to Account for Uncontrolled Factors Using a Definitive Screening Design

[#1]

Hi,

I am designing a DOE where I will have 6 factors that are controllable, and two factors that are uncontrolled.

When using custom design, I am able to set factors as uncontrolled, however with definitive screening, I cannot.

1.) Is there a way to set uncontrolled factors in definitive screening designs?

2.) If not, how does one account for uncontrolled factors?

Thanks!

[#2]

Hi @MetaLizard62080 ,

I think it is quite simple. You would just add columns for your uncontrolled factors and add the values for them as you execute the experiment. Then include the uncontrolled factors as effects in your model analysis. 

Adding uncontrolled factors in Custom Design simply adds empty columns for you to record the values and includes the factors as effects in the model script. You can easily do this for a DSD as described above.

Note that you will not be able to include your uncontrolled factors in the specific Fit Definitive Screening analysis as this relies on all factors having the foldover structure of a DSD. However you could do a Fit DSD analysis to determine the active controlled factors/effects and then add the uncontrolled factors to the resulting reduced model. Or you could use a stepwise analysis starting with all controlled and uncontrolled factor effects. I would probably do both and see if they arrive at similar conclusions.

I hope this helps,

Phil

[#3]

Thanks for the help Phil!

So a follow up question.

I wanted to test how JMP handles uncontrolled variables in designing the DOE, so I went into custom design, loaded my factors, modeled the main effects, then added 2 uncontrolled factors. JMP then automatically added more runs I assume to account for the increased expected variability associated with having the uncontrolled factors. If I cannot perform set these as uncontrolled variables in the DSD, and therefore, JMP will not add runs, how can I ensure I will have enough runs to accurately model these factors?

Is it better to still proceed without these factors and retrospectively add the columns, or I could set them as controlled factors knowing I cannot meet the targets and then retrospectively add the data, or I can just add additional runs using the "number of extra runs" feature. What do you suggest?

[#4 · Accepted solution]

Good question, @MetaLizard62080 .

JMP will add more runs to the default number whenever you add more factors, or rather when you add more effects to the model that you wish to estimate. It doesn't matter whether they are controlled or uncontrolled factors. If you have more effects to estimate, you need more runs.

In your situation I would most likely choose extra runs in the DSD dialog and then add the uncontrolled factor columns after the design has been created. This assumes that you have the resource to carry out the extra runs.

A DSD is a small design relative to the number of factors and effects that you might wish to estimate, so extra runs will be very useful even if you don't have uncontrolled factors to consider.

I hope this helps,

Phil

[#5]

Thanks Phil.

Makes sense to me! I'm not really material limited for this so seems like the way to go.

[#6]

Hi @MetaLizard62080 : Can you add some detail to what you mean by "uncontrolled"? I ask because these uncontrolled factors could be random effects, or fixed effects...and which they are can have a big influence on the model/results/interpretation/etc.  

[#7]

Hi Phil @Phil_Kay 

This is an approach i would also like to try. do you have any references for the addition of an uncontrolled factor column to a DSD. it would be good to see an example if possible. in that regard, @MetaLizard62080 did it work for you...?

thanks for any comment on this

cheers

Barry

[#8]

Hi @barry78 ,

I don't know of any real examples where an experimenter has added an uncontrolled factor column to a DSD. Sorry. 

However, my colleague @Peter_Polito , wrote a blog about an experiment (not a DSD) with uncontrolled factors.

If you have a specific question about the approach, I suggest that you post as a new thread.

Hope this helps,

Phil 

[#9]

Thanks @Phil_Kay, will take a look.

Barry

[#10]

@MetaLizard62080  @Phil_Kay , good post here.

https://community.jmp.com/t5/JMP-Blog/Proper-and-improper-use-of-Definitive-Screening-Designs-DSDs/b...