creating multiple models and comparing them using combinations of variables.

Question

Hi,

I have found that parsimonious logistic models with 3-4 variables work very well for my data (more than that and the model is unstable). But my variable pool (after selection) is somewhere between 5 to 10 to 20 variables depending on how strict I am on the selection process. I don't know a priori which variables are the most important ones, and of course there's the issue of interactions... So I have a large-ish pool of possible variables, and what I want to do (or at least explore) is a "brute force" approach, testing all the different combinations of variables. i.e. n=20 r=3 or 4. it's a lot! But that would then give me a distribution of models, from which I'd choose the ones that perform best and then move to further validating those ones.

Is there a way to semi-automate this in JMP pro (15.2.1) without scripting? I was wondering of how I could use the bootstrapping option but can't seem to be able to choose only 3 or 4 variables from the pool.

If this can only be done with JSL, if there's a kind spirit who's willing to help with the code I'd be very grateful. I can do basic adaptations using my coding background (from my remote past), but can't code JSL from scratch...

Thanks,

Uriel.

Mark_Bailey · Answer

My apologies! I was thinking of another course

You can use the link on that page to suggest a training date upon which to build a public class. (At least, that purpose is how I understand it.)

utkcito · Answer

Hi Mark,

actually the link shows that there's no date for this course. I would be happy for that to be a mistake!

uriel

Mark_Bailey · Answer

You should know that JMP offers training called "Finding Important Predictors." We have a class starting on November 30. See this page.