Background: I have an experiment which has a linear constraint. For example, X1 may represent a temperature set-point to which I am heating my material, and X2 may represent a cooler temperature which the material will see that allows it to cure after it was heated.
For this (contrived) example a reasonable linear constraint may be X1 - X2 >= 0. JMP DOE makes this really easy to set.
Now, I run the experiment, collect the data, and fit the model. I use Stepwise (let's say AICc, although it doesn't really matter) and come to the conclusion that one of the constrained factors, say, X2, is not needed in the final model.
Problem: If I fit a model that does not include ALL of the columns listed in the Constraint property, JMP will give an error "There was a problem loading the constraint(s) in the JSL constraint property."
What are my options here? Accept an inferior model that keeps all of the factors that are in the Constraint in the model? Delete the Constraint property (fine if I don't have other constrained variables involving X3, X4, etc.)? Save the model equation to the data table, then edit the formula and add a term for 0*X2 (if X2 is the insignificant term) and display the model with the Profiler tool? Other options?
Is this a JMP bug or has JMP always functioned this way? I don't remember this being a problem in older versions of JMP - I'm on 14 now.
In JMP 13, I see that warning when you try to fit the model (using Fit Model/Fit Least Squares) and one of the terms in the constraint is missing, however I can click through the warning and the model fit will still run. I also get the same warning if I've saved the prediction formula to the data table and try to create the Profiler of the prediction equation, but again, I can click through and the profiler runs ok.
I am getting different behavior in situations where I have constraints on many X's with some of these constraints acting on one Y and other constraints acting on another Y, leading to situations like this (top profiler is on Y2 only, second profiler is the Y2 row from a profiler on both Y1 and Y2). Notice that if I don't save both Y1 and Y2's prediction equations and profile them (bottom profiler), the constraints are not respected on Y2 alone.