No quadratic effects in Response surface experiment

[#1]

JMP Pro 

So for an assignement an optimilisation problem has to be solved but i can't seem to enter quadratic effects. First i thought i didn't have enough runs but even after gathering data from 80 runs the quadratic effects stayed on zero( see screenshot). Even though you can an estimate for the effect Mf*Mf , the effect remains very small and doesnt getting noticed by JMP after coding. The data comes from an I-optimal screening design but it could be there are some mistakes. I added the I-optimal design and gathered data below. I used all possible models and forward substitution to get the most signifcants effect ( Alpha,Beta,Aq, Mf, Pin,Dzul) for the responses ( Range , Consumption) . I believe that Alpha should have a significant effect on the Range.

[#2 · Accepted solution]

Welcome to the community, @Giel .  It is difficult to respond to your inquiry as you have not attached the data set from which you are having questions.  Having lots of data points is not what is needed to estimate quadratic effects.  What you need is continuous factors at more than 2 levels (or you can estimate design space simple curvature with center points).  From the screen shot, it appears you are doing stepwise (additive) model building.  If the degrees of freedom are "consumed" before you can estimate quadratic effects, then they will appear to be 0.  Usually, this is because you lack the degrees of freedom to estimate these effects or they have already been explained by other terms in the model.  Have you tried analyze>fit model>standard least squares personality and effect screening?

[#3]

Hey @statman  first off all, thank you for taking your time for this problem.  I thought I added my first data in a text file below , to be sure i added it here below in a jmp file. I did indeed make a model using continous factors with 2 levels, now i tried to make a different I optimal design using discrete numeric factors with 3 levels. I however dont know if its theoretically right to just change the continous into discrete numeric factors, what do you think?  The data with 3 levels indeed gave me the necessary quadratic effects . 

[#4]

Some important fundamentals...What questions you can answer, what conclusions you can draw, what tools/techniques you use for analysis, interpretation of the results, confidence in your ability to extrapolate conclusions ALL DEPEND ON HOW THE DATA WAS ACQUIRED!  While you can certainly "play games" of analysis by changing factors data types, ultimately the analysis should be appropriate for the data type.  For example, let's take a continuous factor at 3-levels vs. a categorical factor at 3-levels: it certainly makes sense to estimate a quadratic effect for the continuous variable, but it makes no sense for a categorical factor.  That given, since your data is from an experiment, discrete numeric will give you the same model as continuous variables tested at the same specified number of levels.