Does Box Cox transforming my regression model display the transformed or untransformed data?

[#1]

Simple question (I Believe). I had to perform a Box-Cox transformation to obtain a better regression fit. Does the prediction profiler transform back to the untransformed values (e.g. the original) without me having to manual transfer back myself?

Thanks!

[#2 · Accepted solution]

That depends.

I used the Refit with Transform command and the Prediction Profiler in the new Fit Least Squares platform still shows the original variable. That is to say, it back-transforms the prediction of the fitted model. This behavior is captured if you save a script for the platform at this point.

On the other hand, if I save the transformed response to a data column and use it in the Y role, the Prediction Profiler shows the transformed response. JMP does not remember where the variable in the Y role came from.

You could have tried it yourself. JMP makes it easy to try things without breaking anything.

[#3]

I really like @Mark_Bailey 's statement about trying things in JMP. One of my pet missions was to always encourage people to NOT try and pick the 'best', whatever 'best' actually means, modeling approach BEFORE any modeling work is done. Try a bunch and see which one works for your particular questions, data and problem at hand. Standard Least Squares, General Linear Modeling, Tree Methods, Penalized Methods (if you have JMP Pro), Neural Networks, and on and on. Try them all, and to paraphrase Mark, "...you can't break anything."

I used to tell my JMP users, it's so easy, your just moving around electrons in the software...go for it. It would be a shame to leave something untried for so little investment in effort.

[#4]

A great approach and something I continue to strive towards!