Absolute value of fit model

[#1]

I've run an experiment and am doing a fit model. I am only concerned with the absolute magnitude of the response, but to get a good linear fit I have to include the sign of the response.

My question is if it's possible to run the fit model with the sign, but then have jmp display the absolute value in all the charts (interaction profiles, prediction profiler, etc.). That way it would be easier to see when the response is being minimized rather than trying to keep track of how close to zero it is.

[#2]

If I understand your question, you want to fit the model to a response that ranges from negative to positive values but then plot the absolute value of the predicted response. If that interpretation is correct, then follow these steps:

• Fit the model as usual.
• Click the red triangle next to Response and select Save Columns > Prediction Formula.
• Select the new column in the data table and select Cols > Column Info... .
• Select the Formula column property.
• Click Edit Formula.
• While the red box surrounds the entire formula (your model), click the Numeric function group and select Abs.
• Click OK to dismiss the formula editor.
• Click OK to dismiss the column information dialog.

Now use this column in your plot.

[#3]

Thanks for the reply. You understood the problem correctly, but editing the column formula only solves half of it. I would like to be able the use the factor profiler and interaction plots with the absolute values already applied. In a model that doesn't cross zero you can easily see how to minimize or maximize the response by the trends on the y-axis. But with a zero crossing response there isn't a clear direction to move factors without also studying where the zero line is (a point that is easy to miss and might be confusing when discussing with others).

[#4]

I found a decent workaround for the prediction profiler (but not interaction plots, etc). I want to minimize the response, so using desirability functions I can set high desirability at 0 and low desirability on both positive and negative values. Now I can watch the desirability profile and drive things to a maximum that way while looking for interactions between factors.

Not a perfect solution, but fits 90% of my needs for now.