[R-sig-eco] TukeyHSD error from split plot model (Chris Habeck)

[Philip Dixon]

  Chris,

The problem is more fundamental than finding/using an R function, if by 
'interaction term' you mean comparison of all pairs of cell means..  
Statistically, Tukey;s HSD doesn't exist for that situation.  Tukey's 
pairwise comparison method assumes there is a one and only one standard 
error for the difference of two means.  There are two standard errors 
for comparisons of cell means in a split plot, one for comparisons 
within a main plot treatment and one for comparisons between main plot 
treatments.  I recommend folks use a Bonferroni adjustment to the 
p-value, a sequential Holm adjustment, or FDR.  That is assuming that 
you have calculated the appropriate s.e. for each pair of differences.  
This can be done, but is not easy, in R.

I have been told that Doug Bates and the LME4 group are working on an 
effects package that will simplify these calculations, but I don't know 
its current status.  I suggest you look at the R-mixed models forum for 
the most recent information.

Philip