What inspired this wish list request?
When doing model selection for a pharmaceutical process step, we often do smaller designs that have many partially confounded effects. What I commonly do is fit a full RSM or factorial to degree 2 model with GenReg, and use a model selection method (Stepwise or Lasso). Often specifying the use of heredity is not possible due to JMP reporting it cannot maintain the heredity in the model selection process. Turning off heredity will select a model, but often with higher order terms that do not have their main effects included. In GenReg you can request to relaunch with all active effects and also that with interaction or RSM terms, but if main effects were not in the initial selected model, main effect will not be in those relaunched models.
Simple example:
There are 10 factors, X1-X10, and staring with a full RSM model in these factors, Lasso in Genreg selects this model:
Y =a X1 + b X2*X3
X2*X3 also happens be highly correlated with X4
I want to be able to relaunch a staring model in X1, X2, X3, X4 that has either just main effects, main effects + 2FI, or a RSM in those factors.
What is the improvement you would like to see?
For a fitted model from a model selection in GenReg, add some way to:
a) for the higher order effects selected, identify the main effects that may be highly correlated to that interaction term
b) add a way to relaunch with active effect (with higher order model terms, potentially) , including the main effects for the selected parameters and the list of parameters found in a)
Why is this idea important?
It would provide efficiency and consistency in how we do DOE analysis.