Subscribe Bookmark RSS Feed

Advice Needed to Select Modeling Platform

robot

Community Trekker

Joined:

Feb 27, 2012

Hi,

I am a long-time user of JMP, but I am relatively new to the Modeling platforms in JMP.  I have a response with several known effects with a roughly linear relationship.  The response will also vary with its position within XY space of a wafer.  Please see image for more details.

What is the best platform to use to develop a model for this response?  I am using JMP10.

Thanks!

10080_2015_09_30 Effect Response Example.jpg

1 ACCEPTED SOLUTION

Accepted Solutions
Solution

Robot,

I am guessing you want to know the how the 3 main effects affect your response independently and in combination (interactions).  Based on your scatterplot above the Fit Model platform is a good place to start.  Open the Fit Model platform and set your response as Y and then select the 3 effects and add them to the Construct Model Effects.  Select Run and review the Summary Statistics to get an idea of how these 3 main effects fit the model.  If what you show above is a true indication of your data the fit should be pretty good based on the R-Square and R-Square adjusted. Look at Prob > F in the Whole Model fit.  You want that to be closer to zero.  This term lets you know whether your model is better than the mean of your data would be better at explaining your variation.  Look at the Prob > |t| in the Parameter estimates table as well.  The JMP default is 0.05 to decide whether or not an individual variable is important to your fit model or not.  There is more behind explaining the importance of this P value and how you should interpret it, but this is just about getting you started.  Plot Resdiual by Predicted and look for obvious patterns in the data.  If these residual points look fairly random your model form is probably good enough.

That will get you started.  I would also suggest that you watch the on-demand webcast at the l

ink below.  This will walk you through step-by-step for using the Fit Model platform.

Specifying and Fitting Models | JMP

Best,

Bill

3 REPLIES
Solution

Robot,

I am guessing you want to know the how the 3 main effects affect your response independently and in combination (interactions).  Based on your scatterplot above the Fit Model platform is a good place to start.  Open the Fit Model platform and set your response as Y and then select the 3 effects and add them to the Construct Model Effects.  Select Run and review the Summary Statistics to get an idea of how these 3 main effects fit the model.  If what you show above is a true indication of your data the fit should be pretty good based on the R-Square and R-Square adjusted. Look at Prob > F in the Whole Model fit.  You want that to be closer to zero.  This term lets you know whether your model is better than the mean of your data would be better at explaining your variation.  Look at the Prob > |t| in the Parameter estimates table as well.  The JMP default is 0.05 to decide whether or not an individual variable is important to your fit model or not.  There is more behind explaining the importance of this P value and how you should interpret it, but this is just about getting you started.  Plot Resdiual by Predicted and look for obvious patterns in the data.  If these residual points look fairly random your model form is probably good enough.

That will get you started.  I would also suggest that you watch the on-demand webcast at the l

ink below.  This will walk you through step-by-step for using the Fit Model platform.

Specifying and Fitting Models | JMP

Best,

Bill

robot

Community Trekker

Joined:

Feb 27, 2012

Thanks Bill.  I will definitely watch the webcast.

For the model, I am hoping that given:

Effect_A = a

Effect_B = b

Effect_C = c

Wafer_X = x

Wafer_Y = y


What is the predicted response?


The response varies roughly linearly with effects A, B, and C, but spatially with X and Y.  Is the Fit Model platform able to account for the spacial variation in X, Y?  I will read and watch the webcast to learn more on my own.  Thank you for the input!

robot

Community Trekker

Joined:

Feb 27, 2012

Thanks Bill.  The webcast was very helpful.