nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

750570_en-US

liliconcom

Multi component, multi supplier and multi concentration DoE design

I am trying to create a DoE design in which I have 3 supplement types and each supplement have 3 different suppliers. I would like to test more than 3 concentrations for each supplement. So I would need to use discrete numeric design, however this would mean that JMP's suggestion on the ideal concentration wouldn't be outside of these predetermined numbers. Another issue I seem to be facing with is that same supplement types from different suppliers cannot be mixed and JMP doesn't seem to deal with this problem. I have tried to use Covariate factors however it doesn't seem to solve the problems. Thus I am unable to create a Design. Any suggestion and guidance would be greatly appreciated.

Advanced Statistical Modeling

Design of Experiments

Predictive Modeling and Machine Learning

Re: Multi component, multi supplier and multi concentration DoE design

Its not very difficult to recode the concentration levels (see tables attached)

There is a risk that the orthogonality of the design will be effected; however, in this case it looks pretty good.

Recoded concentrations are on the right, default on the left.

Note that in the attached tables, (run the DOE Dialog script) the concentrations are "Hard" to change.

Look at distribution plot using Whole plots and each of the 6 factors. It looks like each concentration is included with each supplement vendor combination (or at least 8 of them)

Re: Multi component, multi supplier and multi concentration DoE design

Thanks Byron, this definitely looks better than how I was trying to set it up before. However, once the numbers are being randomly generated they don't seem to be used for each of suppliers, and I would need to test the same concentrations for each supplement from the different suppliers, unless I can manually change in the created table the concentrations without messing the design and analysis up.

Re: Multi component, multi supplier and multi concentration DoE design

Hi @Moses8,

Welcome in the Community !

I'm not sure to have fully understood your experimental setup :

Are the supplement types independent factors, meaning a supplement can vary in a range of concentrations independently from the other supplements ? If yes, you may use an Optimal/factorial design. If the supplements are components of a mixture, meaning their concentrations/quantities should add up to a fixed value/threshold, then you may use a Mixture design.
Are the 3 suppliers different for each supplement type ? If yes, you can setup your factor with a 3-levels categorical factor (A, B, C for suppliers levels for example), but use supplier as a nested effect in the analysis, as the supplier depends on the supplement type. See about nested effects here : Construct Model Effects (jmp.com)
Do you have restrictions on the possible concentrations values for each supplement ? If not, I would consider using these concentration factors as continuous numeric (instead of discrete numeric), and choose a relevant model so that you can have the minimum number of levels you want. This option would also avoid having a "discrete optimization" for the concentration value choice as you mention (only 3 discrete numeric value available).

If you can provide more information about your project, that could help JMP users provide you help and assistance for your DoE.

Re: Multi component, multi supplier and multi concentration DoE design

Sounds like you want a design like this

For the concentrations of the supplements, I left those to vary between 10 and 50%. Note that they sum to 100%

Don't use discrete numeric for continuous variables. Reserve it for when you have something like the number of teeth on a gear so something that can't be easily changed.

The mixture role lets you have correlated X's and you can have more than 3. The modeling type is Scheffe Cubic which causes multiple concentrations to be used, kind of like when you have a quadratic model for a continuous variable. Note, for polynomial terms you don't add the levels yourself, those are calculated from the range you specify.