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; } } }

Custom DOW with mixture factors

akrishna39

Hi,

I have to design a 48-run design with 16 Categorical variables (2-level each) and 4 mixture variables. I need the mixture variables to have values 0, 0.5 and 1. But there exists the option to put only 2 values in the Custom design template (snapshot attached). Please let me know how can I generate a design with 3 distinct values of the mixture variables.

JMP_2_levels_only_for_mixture_factors.jpg

Accepted answers

cwillden

What often throws people off about Custom Design is that for continuous factors (and mixture factors) you are not enumerating possible levels, but just the ranges for those factors. The algorithm that generates the design, depending on the model you specified, will optimally determine interior points. For example, if you specify a continuous factor with range 10 - 20 and specify a quadratic term for that factor in the model, the design will end up with design points at 15 as well in order to fit the quadratic term.

If you specify in your model main effects and interactions only for the mixture variables, you will get your desired levels.

What model did you have in mind? With this many factors at 48 runs, the only thing you will be able to accomplish is to identify really, really big signals for the main effects and 2-factor interactions for the mixture components. Power will be good on the categorical factor main effects.

All comments

stephen_pearson

Without a better understanding of the objective of the experiment it is hard to advise. The question you are trying to answer will dictate the number of runs that will be required. By having none of each mixture component for some experiments you risk the variable behaving like a categorical rather than continuous factor.

Are you sure it is a mixture space you are trying to explore? Why must it be 0.5? If you bring your minimum value up to 0.08, then you will be able to explore more of the interior of the mixture space.

cwillden

If you specify in your model main effects and interactions only for the mixture variables, you will get your desired levels.

akrishna39

Thanks cwilden and stephen,

I did not know that in mixture variables I am providing the range and not the levels of the factors as the input. I guess since my number of runs is too small and I need to estimate only the main effects, I am getting only 1 and 0 as values for the mixture variables. I am trying to make a model to screen varaibles. 48 runs is the maximum that I can afford. So, I was looking for a design for 48 runs.

cwillden

You can still get the 2-component blends at 50% each with 48 runs. You would just have to specify interactions only for the mixture variables. Still, you will only detect huge effects.
For the categorical factors you definitely can only afford main effects unless you specify just a small handful to look at.