R help - Nov 2011 - variable transformation for lm

Hello,

I am doing a simple regression using lm(Y~X).
As my response and my predictor seemed to be skewed
and I can't meet the model assumptions. Therefore
I need to transform my variables.

I wanted to ask what is the preferred way to find out
if predictor and/or response needs to be transformed
and if yes how (log-transform?).

I found a procedure in "A modern approach to Regressoin
in R" (Sheather, 2009): There they suggest an approach
with the function bctrans from alr3...but it seems that it
is deprecated. So what is the best way (box-cox test) find the best
transformation for predictor and response simultaneously?
AFAIK boxcox from MASS is used only used for transformation
of the predictor?

Thank you very much
Johannes

--

See the acepack package or the transace function in Hmisc for ACE or AVAS.

-- Bert

On Thu, Nov 3, 2011 at 8:55 AM, Johannes Radinger <JRadinger@gmx.at>
wrote:
> Hello,
>
> I am doing a simple regression using lm(Y~X).
> As my response and my predictor seemed to be skewed
> and I can't meet the model assumptions. Therefore
> I need to transform my variables.
>
> I wanted to ask what is the preferred way to find out
> if predictor and/or response needs to be transformed
> and if yes how (log-transform?).
>
> I found a procedure in "A modern approach to Regressoin
> in R" (Sheather, 2009): There they suggest an approach
> with the function bctrans from alr3...but it seems that it
> is deprecated. So what is the best way (box-cox test) find the best
> transformation for predictor and response simultaneously?
> AFAIK boxcox from MASS is used only used for transformation
> of the predictor?
>
> Thank you very much
> Johannes
>
> --
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>


-- 

Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: 467-7374
Website:
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm

	[[alternative HTML version deleted]]

On Nov 3, 2011, at 11:55 AM, "Johannes Radinger" <JRadinger at
gmx.at> wrote:
> Hello,
> 
> I am doing a simple regression using lm(Y~X).
> As my response and my predictor seemed to be skewed
> and I can't meet the model assumptions. Therefore
> I need to transform my variables.
The presence of skewness in either or both the response or predictors does NOT
imply failure to meet model assumptions. The assumptions of linear regression
regarding normality only apply to the residuals after the estimation of the
model.

-- 
David.
> 
> I wanted to ask what is the preferred way to find out
> if predictor and/or response needs to be transformed
> and if yes how (log-transform?).
> 
> I found a procedure in "A modern approach to Regressoin
> in R" (Sheather, 2009): There they suggest an approach
> with the function bctrans from alr3...but it seems that it
> is deprecated. So what is the best way (box-cox test) find the best
> transformation for predictor and response simultaneously?
> AFAIK boxcox from MASS is used only used for transformation
> of the predictor?
> 
> Thank you very much
> Johannes
> 
> --
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.