Working in a multiple linear regression setting I am attempting to fit a general model to a data set that consists of 7 predictor. To do so I have run the powertransform() function and received the output pictured. I am under the impression that the powertransform() function suggests transformation for the predictors. Under a single parameter transformation the general form falls into a box-cox transformation, does the form utilizing the suggested lambdas differ from that of the box-cox transformation when using the powertransform() function, further what can we concluded from the given output? I was under the impression, that if the values were close enough to certain values they could be rounded down or up.enter image description here


1 Answer 1


My experience with the Box-Cox transformation, which I actually self-programmed into a spreadsheet, suggests first, use the indicated power (like 1.79), do not feel obligated to round.

The primary reason of suggesting rounding is to possibly explain the rationale of the transformation to others. Better is to focus on the best model fit and reverse transform the confidence intervals when done.

One can modify the transform to allow the addition of a constant to the data, which, at times, does assist in the model fitting exercise.

In addition to improving normality, it does appear in practice to also improve the homogeneity in the variance of the residues.

This online educational reference may be of assistance.

  • $\begingroup$ That was my major concern, the model before transforming was extremely normal, however upon applying the suggested transformation the model becomes not normal. Under this instance does it not seem that the powerTransform should not be applied? $\endgroup$ Jun 7, 2020 at 2:37
  • $\begingroup$ Possibly correct, it depends on what also happen to the variance. More homogeneous variance can mean that the prediction error is probably more accurate. Loss of normality is more of a concern for the correct interpretation of significant levels. $\endgroup$
    – AJKOER
    Jun 7, 2020 at 22:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.