Choose Language Hide Translation Bar

Box-Cox transformation and "real data" graph

I needed to perform a Box-Cox transformation on a responds of a DOE. The data look acceptable after this transformation. I am mostly interested in the profiler data (extreme settings).
What is the easiest way to convert those Box-Cox transformed profiler graph back to a "real data" graph (back transformed to actual human readable data points)?
How can I get the 95% confidence interval into this graph?

Thanks in advance.

0 Kudos
Staff (Retired)

Re: Box-Cox transformation and "real data" graph

You may want to check out Box, Hunter, & Hunter (2005) "Statistics for Experimental Design, Innovation and Discovery 2nd Edition, pp 503-506.

After your transformation if the minima is located at...
0 = log
2 = square
1/2 = sq. rt
-1 = reciprical

Also if you double click the column the formula for the transformation is shown

From a chemist:)

Message was edited by: Lou V
0 Kudos

Re: Box-Cox transformation and "real data" graph

Lou, thanks for your responds. Your input is correct and by manipulating the formula is exactly how I am currently doing this back transformation. Just thought there might be a more elegant way to do this.
0 Kudos
Super User mpb
Super User

Re: Box-Cox transformation and "real data" graph

If it is appropriate to use one of the 4 transforms mentioned, then there *is* a more elegant way to proceed. Return to the Fit Model dialog (or recreate it), click on the Y variable and then on the red triangle Transform menu. Choose the desired transform and run the analysis. Then the profiler will display using the original metric and if you save the prediction formula, it will be created in the original metric, meaning that the transform will be unwound in the formula created. The confidence and prediction intervals also will be saved in the original metric but there is no straightforward way to get formulas for them though it is possible with some effort.

0 Kudos