Add VIFs to DOE diagnostics

[lynnehare]

Currently, we see VIFs only after using the Make Table button and simulating a response, then examining Fit Model output, assuming users have added the VIF option to Preferences or other means of revealing them.  But VIFs are design diagnostics, and as such should appear in the design diagnostics section.  Making this change or addition will speed user design generation especially for second and higher order models.

[jszarka]

This is what I saw as JMP's take on this in the past -- Include VIF in Evaluate Design 

[jszarka]

I see this is being piloted in JMP 19 EA. I have submitted a ticket (Case 00174254) - it is not the same VIF you will see in the Fit Model output. It seems closely related to the values in the Estimation Efficiency tab in the Evaluate Design output. I've generated orthogonal Central Composite Designs where the VIFs all equal 1, but the Evaluate Design VIFs indicate most terms with VIF > 2. VIF is also hidden by default - you have to know to right-click on the table to uncover it. I'd also recommend showing it by default.

Hopefully this will be remedied by the production release.

[jszarka]

I see the results delivered, but it isn't great per Case 00174254. Below is a 3-factor CCD with n=20 and the CCD-Uniform Precision design type chosen. With simulated data the VIFs are all close to 1, but the Evaluate Design platform looks a bit worse, with no results equal to 1.

[mia_stephens]

Thank you @jszarka , we have captured this issue. Moving this back to acknowledged.

[jszarka]

My hope would be that the VIFs would end up matching. After that is implemented, I would also suggest this show up by default in the Estimation Efficiency tab as it would be unlikely that most users would know that it is hidden there.

[Sarah-Sylvestre]

If you are an authorized JMP user, you can try this out in JMP 20 EA1. To request to be a part of the EA, please visit JMP.com/earlyadopter. 

In JMP 20, there is a new (hidden) column in the Estimation Efficiency table called 'Fit Model VIF' that is calculated consistently with Fit Model. In JMP 19, there was a hidden column called 'VIF' that used an alternate definition than Fit Model. The new column in JMP 20 uses the same formula from Fit Model, so the metrics should always be consistent. There is also a new platform preference to always show this column so that users that want to see the metric don't have to unhide the column.

 

[jszarka]

Yes, I do see this feature @Sarah-Sylvestre - thank you! It is much more intuitive than the "DOE VIF" term. I will continue to evaluate and add feedback. My first attempts were very successful, so this looks to be a great add.

The addition of a new platform preference is really good to have as well. We can deploy JMP with this preference built in, which provides easier use!