topic Re: Model Validation in JMP Statistical Software as compared to MODDE in Discussions

Model Validation in JMP Statistical Software as compared to MODDE

PieAnteater793 — Mon, 05 Feb 2024 18:04:56 GMT

Hi there,

I am reaching out to seek some clarification on a feature in the JMP statistical software.

In MODDE, we find a valuable capability for model validation, allowing us to scrutinize and validate the model used to run and analyze a Design of Experiments (DoE). This functionality is presented in an intuitive graphical user interface (GUI) manner, facilitating a seamless user experience Please refer the attached image or to minute 21:15 in the linked video for further details - https://www.youtube.com/watch?v=UTYRpStJDE8

However, I've been exploring JMP for similar functionalities, and I am yet to discover a straightforward method for model validation. While JMP is excellent for executing DoE analyses, it appears to lack a built-in mechanism to validate the model it employs during the analysis.

Could you kindly shed some light on whether JMP offers a feature similar to MODDE for validating the DoE model? If so, I would greatly appreciate guidance on how to access and utilize this feature.

Thank you for your time and assistance. I look forward to your insights on this matter.

Re: Model Validation in JMP Statistical Software as compared to MODDE

peng_liu — Tue, 06 Feb 2024 04:18:09 GMT

The plot is a graphical representation of "Summary of Fit". I cannot think of any particular places in JMP that present "Summary of Fit" graphically, but "Summary of Fit" reports (or under related names) are widely available among JMP's analytical platforms.

Rigorously speaking "summary of fit" is not the only way that one can validate a model. Please check out general subjects under the terms such as "Model Comparison", "Model Selection", "Variable Selection", "Cross Validation" among what JMP provides. They are all relevant knowledge for one to build a useful model.

Re: Model Validation in JMP Statistical Software as compared to MODDE

Victor_G — Tue, 06 Feb 2024 12:25:55 GMT

Hi @PieAnteater793,

Most of the responses can be found in the answer from @peng_liu.

You can explore the different report options in the red triangle from your Fit Model windows to display residuals plots, actual by predicted plot, etc...
Note also that depending on the "emphasis" used in "Fit Model", you may have access to some of the plots you would like to see. For example, when the model emphasis is "Effect screening", you can display "Sorted Parameters Estimates" (accessible in the red triangle options of "Estimates"), which offers a synthetic view of statistical significance and effect size :

Concerning your specific plot, you can recreate it with the Graph Builder, as R² and adjusted R² are available in the "Summary of Fit" panel (righ-click on it, and you can directly create a datatable from it), and R² and R² PRESS (or Q²) are available in the "Press" panel (in the red triangle, go to "Row Diagnostics", then "Press") so you can also create a datatable from it.
From then, combine the tables and you'll have a similar plot using Graph Builder :

You can calculate Validity and Reproducibility from MODDE in JMP based on the formula from : https://www.sartorius.com/download/544636/modde-12-user-guide-en-b-00090-sartorius-data.pdf

You may not be able to calculate these metrics if you are not able to conduct this test : Lack of Fit

I hope this complementary answer will help,

Re: Model Validation in JMP Statistical Software as compared to MODDE

PieAnteater793 — Tue, 06 Feb 2024 15:16:30 GMT

Hi @Victor_G ,

Thank you for hitting the nail on the head with your solution. The hyperlinks really helped. Much appreciated!