cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar

Prediction Profiler for Generalized Linear Models with Confidence Limits Formula

When I am using the Fit Model platform with Standard Least Squares, I can effectively save the prediction formula and standard errors to a separate column. When I do this for multiple responses, I can effectively use the Graph -> Profiler menu to overlay those models. This is very helpful, as each response may have a different set of meaningful predictors in it. Here is a reference to such a thread on the topic:

 

https://community.jmp.com/t5/Discussions/Confidence-Interval-in-prediction-profiler/td-p/10817

 

However, if I have a generalized linear model (GLM) such as a logistic regression, I have no such luck. Within the single model for fitting the GLM Personality, the results in that platform have confidence intervals on them. However, I am only able to export the prediction formula. If my most 2 important responses contains one normal response with Standard Least Squares, and one binomial response with a GLM logistic regression model, I cannot create a joint Prediction Profiler with effective predictions and confidence intervals. I am optimistic this can be done because we already have evidence that the confidence intervals can be created for GLMs. You just have to work out what the StdErr Pred Formula column would look like.

 

I brought this up as a ticket with JMP as well. The following series of responses from JMP support certainly brought attention to the issue, but perhaps if others find value in this enhancement, then it is more likely to be completed.

 

"Unfortunately, at the moment, I do not see a way to get a confidence interval around the predicted probability for the binary response in the combined standalone profiler."

 

"I see what you mean about the inability to save Standard Error Formulas for GLM's or for Least Squares models containing nested terms. I am not aware of any particular reason why this is the case. I will pass these items on to JMP R&D for their consideration in a future version of JMP."

 

5 Comments
gerd
Level V

Thank you for bringing this up. Having both continuous and binary responses is a very typical scenario for us, as well, and I miss the confidence intervals for the logistic model every time.

Status changed to: Acknowledged

Hi @jszarka, thank you for your suggestion! We have captured your request and will take it under consideration.

jszarka
Level V

Are there any updates? I saw that the Generalized Linear Mixed Model has some of this functionality. You can save the "Prediction and Interval Formulas." You get the better picture although you don't see the numerical values at certain combinations (see below). Will this be possible more broadly, say in Generalized Regression?

 

jszarka_0-1713366794556.png

 

hogi
Level XII

@jszarka wrote:

 I can effectively use the Graph -> Profiler menu to overlay those models.


Is there a trick ?

Overlay profiler from two different models 

hogi_0-1728127074849.png

 

jszarka
Level V

@hogi  - I'm afraid that my use of 'overlay' doesn't match what's in your thread. I meant in my case that you can have multiple profilers from different types of models, like logistic regression and ordinary linear regression. Having literal prediction equations on top of one another in the Profiler is an interesting request as well. The suggestion there on the residuals column seemed reasonable because concepts like desirability functions and the Simulator tool can be used to see how far away from zero the difference in models could be. Hope it works out!