Fit Models (Responses compared to reference, Mixture)

AnnaSan

Hello all,

I have started to use jmp recently to learn how to plan experiments. So far I did some trials, but I have problems with all the different options and data arrangements when building models. I share some data to show you where I stuck:

Experiment 1:

3 Factors (Mixture Component 1, Mixture Component 2, Continous Factor 1), 9 Experiments

3 Responses (only 2 in the data sheet; aim: maximize)

 

My first questions are about the responses:

The responses are measured as the difference to the values before the experiment (also a mean of several measurements) and I dont know how to add this information to the data to build the model (in the sheet I just substracted the mean to have something start with).

I am also not sure if I need to add the experiment number to the model. Is it assumed that data come from different experiments if I do not?

 

The other part I am wondering about is how to deal with the mixture components in the model. In that particular example I have only 2 components, so is it necessary to fit it as mixture model?. Mixture component 1 does not seem to effect the response, but it seems not correct to remove a mixture factor from a model.

 

Hope you can help me with that. I added the data and some plots I made.

I have only used the interactive tools so far.

Anna

Results Experiment 1.jmp

Interaction plot 2.png

Interaction plot 1.png

statman

I have some thoughts, but I'm not sure exactly what you are asking:

1. It appears you have 9 treatment combinations which is a fractional factorial of the 3^3 design, but I'm not sure how you chose the treatments you did?  Typically when designing an experiment, you will start with an idea of the model you want to investigate.  This is usually derived from the hypotheses you have regarding what factors effects the response variables.  I don't know what model you are investigating, but the design you chose for 9 treatments could have been chosen better.  You have confounded or partially aliased terms in your experiment.

 

2. You have multiple measures for each treatment.  I don't know if these are repeated measures or replicates as there is no run order in your data table.  Even so, one treatment has 6 responses, the others have 10.

3. If you want the delta change from before treatments to after, just add a column for that contains the measures before and then create a column which is a formula that calculates the delta.

4. You use the term "mixture" component?   Are there any constraints?  There are multiple schools of thought regarding mixture designs (I suggest you read Cornell "Experiments with Mixtures".)  For example, are there other components of processing that are not part of the "mixture batch" like agitation rate or temperature? You may want to investigate these factors as well.  Some times you can start with ratios of components. Mixture designs are a special case of experimentation.  This is because changes in the levels of one mixture factor necessitates changes in the levels for the other mixture components (usually the components must some to 1). , hence some of the orthogonal properties of the experiment are compromised.  I usually think of mixture designs as optimization type experiments.  Typically, this means you have already defined the first order model and completely understand noise effects.  You will be analyzing response surfaces to determine where the best area is to work.  But it doesn't do any good if the surface is the base of the mountain you are trying to climb.

Maybe start here:

https://www.jmp.com/en_us/events/mastering/topics/mixture-designs.html

5. No, you should not add experiment number to the model.  It would be advised to keep track of run order as often we will sort in run order when evaluating the residuals of the model.

 

Lastly, I would suggest you start with some basic understanding of experimentation.  JMP offers some on-line  options to learn about experimentation:

 

https://www.jmp.com/en_us/online-statistics-course/design-of-experiments.html