Transforming variables

[John Stokes]

Hi, I had a doubt regarding the transformation of variables.
Someone told me that I should transform every variable into something to improve the results.
For example, if I transform my dependent variable into log function and some of my independent variables into square root, some into cubes, some inverse, does that work?
If I transform my variables like that, how do I interpret them?
First of all, is it right to transform the variables in so many different ways?
Please guide.

[Carlo Lazzaro]

John:
never heard about such a weird advice.
There are situations when transforming makes sense, as variable transformation is a ststistical tool like many others.
Please also consider that converting transformed variables back to their original metric(s) is not an easy task at all.
In addition, without knowing what you're after, I find difficult not to provide a trivilal reply.

[Nick Cox]

There are situations in which variables are transformed differently -- an easy example would be one predictor logged and the other left as is -- but there is absolutely no principle whereby that in itself is a good thing.

Back-transformation is really only ever an issue for the outcome or response variable (dependent variable).