Is it possible to compare the significance of a model effect across models of different systems?

[#1]

Hi,

I'm fairly new to JMP and statistics. I'm trying to model the effects of A,B, and C of percent recovery across different systems (X, Y, and Z). I'm using the Analyze-> Fit Model function and using the Response Surface Macro to generate the Model Effects. I'm generating models for each X,Y, and Z for using A,B, and C as the main models effects for each. Then I remove the model effects (main effects, two-way interactions, and quadratic terms) to optimize the adjusted R2 to generate the model.

Is it possible to comment on the significance of effect A in system X vs the significance of effect A in system Y on percent recovery? and if effect A has greater positive effect on percent recovery in system X compared to system Y (if effect B and C are held constant).

I know under the Prediction Profiler, there's the "Assess Variable Importance" function, but I wasn't sure if it were appropriate to use this function to compare variable importance across models for two different systems (even if the effects are the same).

Thanks,

[#2]

Commonly, different responses measured for each run of a designed experiment depend differently on the factor changes. Fitting each response separately and selecting the best model separately is often the best way to accurately explain or predict the effects.

You can define a common model in the Fit Model dialog window and then select the Fit Separately option before clicking Run.

[#3]

Thanks for the response. Just to clarify,  if I were to create models of how a set of inputs influences % recovery in two different systems, then it would not be possible to comment if one of those parameters in more important in one system than another? For example, I created independent models using inputs A-E trying to predict % recovery in two different systems using the Fit Model function and the Response Surface macro to account for interaction effects. Then, for the model for each system, the least significant effects (main, interaction, etc) were removed to optimize the R2. I also used the "Assess Variable Importance" function under Prediction Profiler and got the these tables.

System 1:

System 2:

Would it be accurate to say that Variable D has a more important effect in predicting the %recovery in System 1 than in System 2 because the “Total Effect” is larger in System 1 than in System 2?