How do I get prediction formula as DOE design using Fit model in analyze

Question

Hello,

I made a experimental design manually and paste it into JMP table instead of using DOE design tool available in JMP.

After adding all required response, I ran fit model to analyze the data and get a model formula. I constructed model effects using response surface in Macros.

After running, I checked the formula but that is not in the form as I expected.

For example, I set temperature range between 100 and 140 with the center point 120. So this level can be coded between -1 and 1 by (temperature-120)/20.

I expected that I will get the prediction equation in the coded form but what I got is the formula with different constant. Could you please let me know how to solve this issue? Thank you for all your help!

Best regards,

statman · Accepted Answer

First, welcome to the forum.  Second, if you can attach your JMP file to your post, that is helpful to us (you can anonymize it).

When you run fit model, JMP will analyze the data (coded in the background) and then transform the beta coefficients back to the original level setting in your table.  If you want a coded model, you should code the levels before you run fit model.  Coding should be equi-distant centered on 0 e.g., -1,1 for a 2 level factor, -1,0,-1 for a three level factor)

statman · Accepted Answer

Thanks for adding your JMP table.  I don't understand your question...The model will be written in whatever levels were used for analysis.  So if the left one (screenshot) was with actual values and the right one was with coded levels, the models will, of course, be different.  Realize the beta coefficients get multiplied by the terms and so they must be adjusted for the actual levels when writing the model.  They are coded to normalize the slopes so they can be compared for significance.

A great test is to save the prediction formulas to the table (Red triangle next to Response Yield)>Save Columns>Prediction formula.  Once you have done this, you can put in values for the factors that are significant (just add rows to the table) and JMP will predict the results using the model saved.  Do the results make sense?

I looked at your data set, and I'm not sure how you got to the model(s) you did?  Did you analyze the effects to determine which ones were significant and then reduce the model to the significant terms? I can see some of the terms you included have very little effect. There are some funky residuals as well.  Looks like biomass and temp have the biggest effects. IL and Time may interact with biomass, but their main effects are negligible.

I ran fit models for both uncoded and coded (after reducing the models) and saved the prediction formulas to the table.