[R] Schmera: a question on R software

[Torsten Hothorn]

On Wed, 18 Oct 2006, Douglas Bates wrote:

> On 10/18/06, Torsten Hothorn <Torsten.Hothorn at rzmail.uni-erlangen.de> wrote:
>>
>> On Wed, 18 Oct 2006, SCHMERA Dénes wrote:
>>
>>> Dear All,
>>>
>>> I would like to run a generalized linear mixed model with the software R (one categorical predictor, one random factor, the distribution of the dependent variable is binomial, and the link is logit). Thereafter, I would like to perform multiple comparisons (post hoc test) among the groups of the categorical predictor.
>>>
>>> Is it possible with the software R?
>>>
>>
>> yes and no. There is the `multcomp' package available but the CRAN version
>> is limited to linear models (more or less). However, I'm currently working
>> on a re-implementation which has solutions for your problem and much
>> more :-)
>>
>> For the time being, the development version is available from
>>
>>         http://www.imbe.med.uni-erlangen.de/~hothorn/multcomp2
>>
>> The basic steps would be as follows:
>>
>> ### fit logistic regression model
>>> my.glm <- glm(response ~ myfactor, data = <mydata>, family = binomial())
>> ### set up linear hypothesis for coef(my.glm) corresponding to all-pair
>> ### comparisons
>>> library("multcomp2")
>>> gh <- glht(my.glm, linfct = mcp(myfactor = "Tukey"))
>> ### simultaneous confidence intervals with plots
>>> confint(gh)
>>> plot(confint(gh))
>
> The original question was about generalized linear mixed models which
> is going to be even more difficult. :-(

oh, oups, didn't read the question carefully enough. The framework is 
general enough that one can feed in `lmer' objects into `glht'. It is 
assumed that the fixed effects coefficients are asymptotically joint
normal with covariance as reported by `vcov()' and confidence intervals 
and (adjusted) p-values are valid only asymptotically.

However, if this is really appropriate is a question I'm not able to 
answer :-(

Torsten

>
>>> Are traditional methods of multple comparisons (for instance Tukey test) adequate for this purpose or not? Does Tukey test consider the binomial dsitribution of the dependent variable and the logit link?
>>>
>>> I am waiting for your answer,
>>> Denes Schmera
>>>
>>> ______________________________________________
>>> R-help at stat.math.ethz.ch 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.
>>>
>>
>> ______________________________________________
>> R-help at stat.math.ethz.ch 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.
>>
>>
>>
>
> ______________________________________________
> R-help at stat.math.ethz.ch 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.
>