# Transforming Variables

R

#### redckc

##### Guest
I have several different variables. The majority of the variables were positively skewed and therefore I did a log10 transformation to make them more normal. However, two of the variables were quite normal already and now after log10 are more negatively skewed. In order for me to do some parametric tests (possibly an ANCOVA) what do I do about those two variables?

Is it statistically correct to leave those two variables untransformed and compare them to the rest of the variables which have been transformed?

I would appreciate any help!!! ( I have looked at other threads and to my knowledge I could not see an answer).

thanks alot!

#### Dason

Are you talking about your independent variables or do you have multiple dependent variables?

R

#### redckc

##### Guest
I am talking about independent variables. However, I may intend to alternate it so that sometimes one of these independent variables may act as a dependent.

So just to clarify: Pretending all of the below are continous (I have categorical variables aswell which I may intend to use within the GLM but obviously these don't need transforming).

independent continous variables: a b c d e f g h + (unnamed categorical if GLM)
dependent: i

independent continous variable: a b c d e g h i + (unnamed categorical if GLM)
dependent: f

Does this make sense? I hope I made it clear for other users as well to read.

#### Dason

You realize you don't need your independent variables to be normally distributed right? You can leave them as is. It's really the residuals we want to be normally distributed. But if you're fitting a GLM even then you don't care about the residuals being normally distributed (there are other things to worry about though).

R