Using Log in a Box-Cox Transformation

[Fabien Gilles]

Hi everyone,

I have a question regarding statistics in general, rather than STATA- and was hoping I could get some help.
I am currently working on multiple regression and am trying different transformations, as my data isn't linear.
Would it be okay for me to use the boxcox function on the logged versions of my dependent and independent variables? i.e. boxcox lnDepdendent lnIndependent1 lnIndependent2 (and so on...)
I feel like it would be overdoing it, but just wanted my suspicions confirmed. Thank you in advance!

[Nick Cox]

That makes no sense. The point of the boxcox command (not a function) is to choose a transformation. By working with ln of your variables, you already chose a transformation. Also, dependent on your units, logarithms can be zero or negative, but boxcox is about entirely positive variables.So for example ln ln as a double transformation requires all original values to be greater than 1; it's hard for me to think of a substantive example where that is natural.

So, your misgivings have the right sign but the wrong magnitude.

[Hemanshu Kumar]

your misgivings have the right sign but the wrong magnitude

this is a classic!