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; } } }

Replicating DOE analysis from a paper

PolygonBison420

Hey guys,

very very new to the JMP/DOE/stat world and I have some questions.

I'd like to replicate ANOVA and RSM analysis I've found in a journal.

What I know:
CCD design
20 treatments including 5 center points
3 factors and 3 responses

Table of factors and their responses is attached.

What I learned from the table:

- rotatable CCD (value of 1.682)
- it's actually 6 center points, not 5 (am I correct with this one? Value combination of 60/1/3.5 appears 6 times)

I am not sure how to incorporate axial values to the analysis. Do I just go Fit Model and then add factors via Macros/Response Surface?

Thanks!

Data1.png

Accepted answers

Victor_G

Hi @PolygonBison420,

Welcome in the Community !

In order to create a similar design, you can go in "DoE" menu, then "Classical" and "Response Surface Design".

You can then specify the 3 continuous factor and 3 responses.

Then you have the choice between different designs :

The CCD with Uniform Precision does contain 6 centre points, like in your example.

Then, select the axial value corresponding to rotatable design (1,682 for 3 continuous factors) :

And you'll end up with a similar design to the one you have shown. Here is the script to generate the design :

DOE(
	Response Surface Design,
	{Add Response( Maximize, "Capacity factor (k)", ., ., . ),
	Add Response( Minimize, "Retention time (Rt2)", ., ., . ),
	Add Response( Maximize, "Resolution (RS12)", ., ., . ),
	Change Factor Settings( 1, 55, 65, "A: Organic Solvent" ),
	Change Factor Settings( 2, 0.8, 1.2, "B: Flow rate" ),
	Add Factor( Continuous, 3, 4, "C: pH", 0 ), Set Random Seed( 31406552 ),
	Make Design( 2 ), Set Axial Choice( 1 ), Center Points( 6 ),
	Simulate Responses( 0 ), Save X Matrix( 0 ), Set Run Order( Randomize ),
	Make Table}
)

For more infos about Response Surface Designs, you can look here : Build a Response Surface Design (jmp.com)

In order to analyze each model linked to the responses, you can have a look here : Standard Least Squares Models (jmp.com)

I attach the design datatable with a script "Fit Least Squares" that provides the analysis of the responses (please take note that I am not sure that the target for each response is correct, i.e Capacity factor should be maximized, Retention time should be minimized, and Resolution should be maximized ?).

You can try it yourself by launching the script "Model", selecting the responses (model is already defined, but if you have to do it manually you can select the factors and use the macro "Response Surface") and check "Fit Separately". You can then compare the models in the publication with yours.

Hope this answer will help you,

Example_Central-Composite-Design.jmp

All comments

Victor_G

Hi @PolygonBison420,

Welcome in the Community !

In order to create a similar design, you can go in "DoE" menu, then "Classical" and "Response Surface Design".

You can then specify the 3 continuous factor and 3 responses.

Then you have the choice between different designs :

The CCD with Uniform Precision does contain 6 centre points, like in your example.

Then, select the axial value corresponding to rotatable design (1,682 for 3 continuous factors) :

And you'll end up with a similar design to the one you have shown. Here is the script to generate the design :

DOE(
	Response Surface Design,
	{Add Response( Maximize, "Capacity factor (k)", ., ., . ),
	Add Response( Minimize, "Retention time (Rt2)", ., ., . ),
	Add Response( Maximize, "Resolution (RS12)", ., ., . ),
	Change Factor Settings( 1, 55, 65, "A: Organic Solvent" ),
	Change Factor Settings( 2, 0.8, 1.2, "B: Flow rate" ),
	Add Factor( Continuous, 3, 4, "C: pH", 0 ), Set Random Seed( 31406552 ),
	Make Design( 2 ), Set Axial Choice( 1 ), Center Points( 6 ),
	Simulate Responses( 0 ), Save X Matrix( 0 ), Set Run Order( Randomize ),
	Make Table}
)

For more infos about Response Surface Designs, you can look here : Build a Response Surface Design (jmp.com)

In order to analyze each model linked to the responses, you can have a look here : Standard Least Squares Models (jmp.com)

Hope this answer will help you,

Example_Central-Composite-Design.jmp

Phil_Kay

Hi,

In addition to what @Victor_G has shown you about designing the experiment, I think you are interested in how to repeat the analysis in JMP.

The starting point is obviously to have the data in JMP. If this comes from a pdf, you should be able to import the data using the option in JMP to import from pdf.

Or you can manually input the data into a JMP table.

Or you might have the data as .csv or .xlsx, which can also be imported directly into JMP.

I suggest that you share the data as a JMP file first. Then the community will be able to help you to replicate the analysis.

I hope this helps,

Phil

PolygonBison420

Victor, that works brilliantly. I was just confused since authors talked about 5 center points which didn't make sense when looking at the options suggested by JMP. Center points are always treatment of all variables at the middle level/0, right?
It seems it's also harder to recreate RSMs from the papers since almost no authors list response limits.

Since you've been so kind to recreate the design in JMP I also want to ask you what's the most effective way to do so.

For instance, I recreate the design via DOE/Clasical/RSD. My result is a data table with the variables however what's the best way to add response data from the paper?

Is there a quick way for me to match the pattern order of the runs of the paper (via random seed perhaps)? Or do I just have to manually seek for it ("aha --- treatment is in line four, let's copy the response to the first response column of my table..." ).

Hope the question makes sense.

Phil, thanks for the suggesting that.

Victor_G

Hi @PolygonBison420,

Is the paper accessible online ? Do you have a link (it could help to better know how the centre points were defined, what were their expected goals regarding the different responses, and their conclusions/interpretation/optimum found).

Centre points are points that have all factors at their middle level 0, right.

Maybe the authors meant that they had replicated centre point 5 times (or did 5 replicate runs of the centre point) ?

In my case, I generated a design through JMP, and had to "manually" match the responses values from your screenshot to the correct rows of the design. Not an optimal solution, I know, there might be a possibility to have the same order of rows, but as you mention, that would mean knowing the random seed (from the publication) and having access and use it in JMP during design creation : Output Options (jmp.com) (option in the red triangle of the DoE window creation "Set Random Seed").

As @Phil_Kay suggested, you might also directly import the table from the pdf or .csv/.xslx into JMP (this way you don't have to manually copy-paste some values), but before running the analysis, you may have to add some column properties, so that JMP better understand some properties linked to your factors and responses :

For factors, column properties "Coding" (specifying values for low (-1) and high (+1) levels), "Design Role" (specifying the type of factor in your design, here "Continuous" for the 3 factors) and "Factor Changes" (by default on "Easy", but for specific design like split-plot design, it is changed to hard or very hard in order to constrain the randomization of runs),
For Responses, column property "Response Limits" is necessary so that the optimization goal is set (maximize, minimize, match target or None), and the values and associated desirabilities are known for the Prediction Profiler.

I hope this answer will help you,

PolygonBison420

Victor,

thanks again.

The paper (open access):
https://aip.scitation.org/doi/pdf/10.1063/1.5112268

I did look more carefully in the center points but all I can find is the following quotes:
"Twenty experiments along with 5 center points were studies with three factors"
"A three-factorial, CCD was taken with fifteen experimental runs and 5 center points"

Your process of matching responses was similar to mine then.