[R] Redundancy contrasts

[John Fox]

Dear Robert,

What you suggest is certainly meaningful (that is the hypothesis tested 
makes sense), but raises an issue of simultaneous inference. There are 
simultaneous comparison procedures for dealing with this kind of problem. 
See, for example, the multcomp package.

John

At 12:05 PM 2/18/2003 -0200, you wrote:
>Is it meaningful to run 2 different sets of contrasts on a model , or is
>there some redundancy somewhere ?
>For example,I have a model :
>tcons ~ group
>
>where group is a factor with 3 levels ( A, B, C)
>I first run the model with the default contrasts (treatment),
>so I  tested   (A vs B)   and (A vs C);
>but is it meaningful to also carry on a 2nd analysis with an other set
>of
>contrasts, to  test B vs C  ?  ie   c(0,-1,1)  ,  in fit.contrasts
>gregmisc notation.
>
>I'v heard  that the  first  set of contrast ( ie treatment default in
>the example)
>  extracts all the "information"
>in the model, and that  a second analysis with an other set of contrasts
>was not meaningful.

-----------------------------------------------------
John Fox
Department of Sociology
McMaster University
Hamilton, Ontario, Canada L8S 4M4
email: jfox at mcmaster.ca
phone: 905-525-9140x23604
web: www.socsci.mcmaster.ca/jfox
-----------------------------------------------------