Fit Two Level Screening Different outputsI find when running a 'Fit Two Level Screening' script for my DoE, I obtain slightly different numbers each time I run it. The overall ranking of significance for each parameter remains the same but the p-values and t-ratios change, such that some parameters move in and out of significant (p < 0.05). My data is somewhat noisy and I only have a 1-2 parameters that are just marginally significant. I'm running a classical screening design for 7 two-level factors (5 continuous, 2 categorical) with one output response. I believe the screening uses a simulation method so is not deterministic. Is this due to there being many solutions to fit the data?
Re: Fit Two Level Screening Different outputsThere is good info on Fit 2-level screening in the JMP documentation. I think this is a really useful platform in situations like the one you describe. The help documentation says: "A screening design often provides no degrees of freedom for error when the model of interest includes interaction terms. Consequently, classical tests for effects are not available. In such cases, the Fit Two Level Screening platform is particularly useful."
I have found it to be a useful way to guide me as to what are likely to be important effects to include in my model. I would then use that guidance and fit a model using the Fit Model platform. This is easy to do from the 2-level screening platform: with your effects selected (I would select all effects that are clearly or border-line significant) hit run or make model as per these help instructions.
The Fit Model report then gives you a more definitive answer about the significance of effects. You can also edit the model in this platform from the Effect Summary to take out any effects you find to be unimportant. You can see the effect on the model to help you find the most useful model.
