[R] how to specify a particular contrast

[array chip]

I didn't set the variable "treat" as ordered factor,
just ordinary factor. In my example, the 3 levels of
"treat" are of equal distance in the order of from a
to b to c. So my understanding is that a contrast in
the form of "a+c-2*b=0" was to test a linear trend of
the response variable among the 3 levels of "treat".
However, this linear trend also includes a special
situation where a=b=c, i.e. the response variable
among the 3 levels are the same, which is contrary to
the linear trend conclusion and is what I want to
exclude. Hope my explanation helps. Thanks


--- Ross Darnell <r.darnell at uq.edu.au> wrote:

> I don't quite follow what you are trying to do but
> the second contrast
> has  a few interpretations with the same meaning in
> your case 
> 
> 1) are the 2-1 and 3-2 differences  equal
> 2) lack of fit of a linear trend 
> 3) is there a quadratic response
> 
> If you declare your factor to be "ordered" then the
> default contrasts
> will be poly()nomials.
> 
> Ross Darnell  
> 
> -----Original Message-----
> From: r-help-bounces at r-project.org
> [mailto:r-help-bounces at r-project.org]
> On Behalf Of array chip
> Sent: Friday, 18 January 2008 10:22 AM
> To: r-help at stat.math.ethz.ch
> Subject: [R] how to specify a particular contrast
> 
> Hi, I am running a simple one-way ANOVA with an
> independent factot variable "treat" (3 levels: a, b
> and c) and a response variable "y". I want to test a
> linear relationship of the response among the 3
> levels
> of the variable "treat" (ordered a->b->c). I used
> glht() from multcomp package. Later I found out I
> need
> to exclude the situation where the response at the 3
> levels of "treat" are equal. I can do separate
> contrasts to test them separately:
> 
> obj<-aov(y~treat,data=dat)
>   ### testing a=b=c
> summary(glht(obj, linfct= mcp
> (treat=c('a-b=0','a-c=0','b-c=0'))),test=Ftest())
>   ### testing linear relationship among a,b and c
> summary(glht(obj, linfct= mcp
> (treat=c('a+c-2*b=0'))),test=Ftest())
> 
> Is there anyway to build one contrast that tests
> both
> at the same time, i.e. just generate one single p
> value. Because the ultimate purpose was to test the
> linear relationship among the 3 levels of the
> variable
> "treat". Or I am asking something that is
> non-sensible
> to do?
> 
> Thanks
> 
> John Zhang
> 
> 
> 
>  
>
________________________________________________________________________
> ____________
> Looking for last minute shopping deals?
> 
> ______________________________________________
> 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.
> 



      ____________________________________________________________________________________
Looking for last minute shopping deals?