Hi community,
I have two questions two; one jmp specific implementation question and the other on DOE strategic question.
I vaguely remember that there is a way to look at Scatterplot and data table more interactively.
For instance, below I selected 4 rows that contain the variables on data table. I would like to have these conditions highlighted in the scatter plot on the right.
Another question is validation of model built from RSM. I was able to built pretty good predictive model through RSM.
I still want to validate this. Do you suggest to just pick random data points from design space to validate the model or use Augmentation?
Any pros and cons and something to consider on it?
Thank you all!
Created:
| Last Modified:
I will illustrate the method using the Coffee Data data table in the Sample Data folder. I selected a model predicting coffee Strength from the Time and Charge factors.
Click the red triangle at the top of Fit Least Squares (the window you show at the start of your discussion) and select Save Columns > Indiv Confidence Limit Formula.
Determine the factor settings for which you want to verify the prediction. I chose settings predicted to yield the optimum and too strong responses. I enter those levels in two new rows of the original data table:
Note the target and tolerance for each condition.
| Condition | Target | Tolerance |
| Optimum | 1.30 | 0.175 |
| Too Strong | 1.47 | 0.175 |
The target is the predicted mean response from the Prediction Profiler or mid-point of the individual confidence interval. The tolerance is the half-width of the individual confidence interval.
I conduct 5 runs for each condition and save the observed responses in a new data table:
Select Analyze > Distribution and cast both data columns in the Y role. Click the red triangle next to Optimum and select Test Equivalence. Enter the target in the first box and the tolerance in the second box:
Click OK.
The mean is simultaneously significantly greater than the lower limit and less than the upper limit. These two one-sided tests (TOST) demonstrate that the 5 replicates under optimum conditions confirm the model's prediction. Do the same test for the other response.