R help - Oct 2007 - Forcing zero intercept in two predictor case - stat question not R question

When one is doing simple regression and needs to force a zero intercept
( for whatever reason. I realize it's a controversial issue ), 
then subtracting the means of the left hand side and the right hand side
from themselves does the trick. Does anyone know if there is a 
similar trick when the RHS has two variables ? Thanks.
--------------------------------------------------------

This is not an offer (or solicitation of an offer) to buy/se...{{dropped}}

Dear Mark,

--------------------------------
John Fox, Professor
Department of Sociology
McMaster University
Hamilton, Ontario
Canada L8S 4M4
905-525-9140x23604
http://socserv.mcmaster.ca/jfox 
-------------------------------- 
> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Leeds, Mark (IED)
> Sent: Wednesday, October 03, 2007 5:49 PM
> To: r-help at stat.math.ethz.ch
> Subject: [R] Forcing zero intercept in two predictor case - 
> stat questionnot R question
> 
> When one is doing simple regression and needs to force a zero 
> intercept ( for whatever reason. I realize it's a 
> controversial issue ), then subtracting the means of the left 
> hand side and the right hand side from themselves does the 
> trick. Does anyone know if there is a similar trick when the 
> RHS has two variables ? Thanks.
Yes, express all three variables as deviations from their means. (This
follows from the observation that the LS regression surface goes through the
centroid of the variables.)

Regards,
 John

On 4/10/2007, at 9:48 AM, Leeds, Mark (IED) wrote:
> When one is doing simple regression and needs to force a zero  
> intercept
> ( for whatever reason. I realize it's a controversial issue ),
> then subtracting the means of the left hand side and the right hand  
> side
> from themselves does the trick. Does anyone know if there is a
> similar trick when the RHS has two variables ? Thanks.
You are suffering from a bit of confusion here.  There is a great  
difference
between forcing the intercept to be 0 as in y ~ x1 + x2 - 1, and  
transforming
(centering) the variables involved so that the intercept ***is*** 0.   
The latter fits
exactly the same model as if you fitted it using the original  
variables (except
that it is generally a bit more numerically stable).  The former is  
what is
``controversial'' (i.e. likely to be misguided).

			cheers,

				Rolf Turner


######################################################################
Attention:\ This e-mail message is privileged and confidenti...{{dropped}}