Mixture design question

[#1]

Hello everybody,

I would like to know if it is possible to use a Mixture design if we want to study the impact of the proportion of the mixture factors but also the total quantity of the mixture.

Basically I would have two mixture factors (corresponding to the proportion of my 2 components ) and a continuous factor to represent the total quantity of the mixture (e.g. from 100ml to 500ml). Is it possible to do that?

Thanks

[#2 · Accepted solution]

Yes, you can do this. Your mixture components are expressed in terms of proportions. Your mixture amount would be a continuous factor. This is called a combined mixture-process design. This name will be helpful as you start looking at how you analyze these types of designs. The model is what I would call non-intuitive, so you will likely want to review the modeling approach.

[#3]

Many thanks @Dan_Obermiller for this quick answer!

I will follow your advice and investigate about combined mixture-process design.

[#4]

To add just a bit to @Dan_Obermiller 's thoughts you'll definitely want to build the design in JMP's Custom Design platform.

[#5]

Thank you for this complement @P_Bartell .

I started to read some documentation. If I understand correctly my situation correspond to this example , right?

And a small additional question, if I want to study this kind of experiment but if the design was not created with JMP, I can use the macro "Mixture Response Surface" from the Fit Model platform and remove the main effect term for non-mixture factors that interact with all the mixture factors?

[#6]

Yes, the link that you have is an example of a combined mixture-process design. Following that approach should work for you.

If the study is NOT designed in JMP, there will be some difficulties. I will assume that you only have one process (meaning continuous) factor. First, the mixture response surface macro will not include the process variable. Second, the process variable should not appear in the model, but should appear as interactions with all of the mixture terms. But then other difficulties come into play. There are column properties for mixture terms that a design from JMP automatically adds. You will need to add those by hand for each of the mixture factors. You will also need to add a Design Role of Mixture to each of the mixture factor columns. These things tell JMP that it is a mixture and allow the Mixture Profiler and Prediction Profiler to work properly as well as fitting the Scheffe model properly instead of a "no-intercept" model.

You might want to create the design in JMP (ALWAYS the best option!). But if you MUST use a non-JMP designed table, create a similar design in JMP and see what the "metadata" is in the data table, and mimic it in the non-JMP designed table.

[#7]

Many thanks for this complement @Dan_Obermiller 

Actually I've just got the dataset from a colleague, that's the reason why I wanted to know how I could analyze it, and above all how to this with JMP next time!!