[R] Orthogonal polynomial contrasts

[Frank E Harrell Jr]

Peter Dalgaard wrote:
> Chris Bennett wrote:
>> How do you remove one of the terms from an ordered polynomial contrast in
>> your linear model.  For example, I have significant terms for linear and
>> cubic but not quadratic, how would i remove the quadratic term from
>> lm(response~treatment)
>>
>> Cheers,
>> Chris
>>   
> Are you _sure_ you want to do that? The interpretation can get tricky, 
> but if you insist, just set the contrast matrix for the term 
> accordingly, e.g.,
> 
> C(myterm, contr.poly(4)[,-2])
> 

Related to Peter's note of caution, using statistical significance as a 
basis for this decision is asking for trouble.  See
   author = {Grambsch, P. M. and {O'Brien}, P. C.},
   year = 1991,
   title = {The effects of transformations and preliminary tests for
           non-linearity in regression},
   journal = Stat in Med,
   volume = 10,
   pages = {697-709}

Frank

-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University