Hi,
If you want to compare between levels of the fixed effect " treatment" ,
I guess you can do as this:
glht(model, linfct=mcp(treatment="Tukey"))
I think that you might need to recode your treatment levels as : A1, A2, B1,
B2.
I am not sure it is legitimate to compare across different fixed effects
using Tukey.
I thought it is only for the comparison of multiple levels within a single
fixed effect.
I hope it will help a bit,
good luck
yuanye
2011/5/26 matteo dossena
> Dear all,
>
> I have to analyse the outcome of an experiment of this kind:
> subjects (1-20) assigned to a treatment with two levels: A and B (ten to A
> and ten to B).
> the response variable (resp) has been measured twice, however the time
> interval between the two measurement and the biological meaning of the
> experiment allow me to consider the two repeated measures as not temporally
> auto correlated. Therefore the only random effect would be subject, and I’m
> using the two measures as blocking factors called "date".
> The design of the experiment is as follow:
>
> treatment A B
> date 1 10 10
> 2 10 10
>
> What I’m interested to is the mean value of resp.
> after the model selection procedure i came up with the best following
> model:
>
> model<-lme(resp~treatemt*date, random=~1|subject,data)
>
> I'm happy with this and all make sense, but now i what to make a pairwise
> comparison.
> To do so i'm trying this way:
>
> K<-cbind(0,diag(length(fixef(model))-1)
> rownames(K)<-names(fixef(model))[-1]
> test<-glht(model,linfct=K)
>
> which gives:
>
> treatmentB==0
> date2==0
> treatmentB:date2==0
>
> but this tests if the differences between the estimated value of the fixed
> effects are equal to zero, and, correct me if i'm wrong, it gives the same
> kind of information of the lme summary table.
> what i actually want is to know if the difference between all the possible
> pair combinations are significant, something that look like the outcome of
> the same TukeyHSD()
> which would be:
>
> treatmentAdate1=treatmentBdate1
> treatmentAdate2=treatmentBdate2
> treatmentAdate1=treatmentBdate2
> treatmentAdate1=treatmentAdate2
> treatmentBdate1=treatmentAdate2
> treatmentBdate1=treatmentBdate2
>
> I’ve been looking for but i couldn't find anything helpful. Any suggestion?
> And an another question which maybe not appropriate for this list. When it
> comes to present the results is it correct make a box plot of the predicted
> value of the model?
>
> ps. as i moving the first steps in either statistical modelling and in this
> list, please accept my naiveness, and of course i'm very open to any kind
> of corrections..
>
> [[alternative HTML version deleted]]
>
>
> _______________________________________________
> R-sig-mixed-models@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
[[alternative HTML version deleted]]