[R] binomial errors in split-plot design

Prof Brian Ripley ripley at stats.ox.ac.uk
Thu Mar 4 16:44:43 CET 2004


First define what you mean by residuals.
Then extract them.
Then use qqnorm and qqline as usual.

Note that the first step is not clearcut even for a binomial glm, nor for 
a Gaussian mixed-effects model.

On Thu, 4 Mar 2004, Christoph Scherber wrote:

> Thanks!
> 
> And how can I then plot  a Q-Q line for model checking? qqnorm works 
> fine, but I couldn´t find how to use qqline for mixed effects models of 
> this type
> 
> so far, I have tried (e.g.)
> 
> qqnorm(glm1,~resid(.)|TREATMENT)
> 
> but I don´t know how to specify qqline for this
> 
> the full model is 
> glm1_glmmPQL(cbindarea~BLOCK+targetweight+TREATMENT+SOWNDIV+GRASS+LEG+SHERB+THERB,random=~1|PLOTCODE/TREATMENT,family=binomial)
> 
> where categorical variables are in capital letters
> 
> Best regards,
> Chris.
> 
> 
> Prof Brian Ripley wrote:
> 
> >There are several possibilities, including glmmPQL (MASS) and GLMM (lme4).
> >Be careful with the interpretation, as you don't have the advantages of 
> >balance that the classical AoV gives you.
> >
> >On Thu, 4 Mar 2004, Christoph Scherber wrote:
> >
> >  
> >
> >>I have proportion data with binomial errors. The problem is that the 
> >>whole experiment was laid out as a split-plot design.
> >>
> >>Ideally, what I would like is having a glm with an Error term such as 
> >>glm(y~x+Error(A/B)) but I fear this is not possible. Would using lme be 
> >>an alternative? How could I state the variance structure, then?
> >>    
> >>
> >
> >  
> >
> 

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595




More information about the R-help mailing list