[R] factor differences in anova

[wouter buytaert]

Well, it is a quite specific design: we want to see if there is
influence of sample location, topography and soil horizon (depth) on the
water retention of a set of soil samples. We have water retention points
of every sample at different pF tensions, and want to compare the
complete pF-curves. As there is of course a strong relationship between
tension (pF) and water content, my PhD director proposed this type of
Error() design.

Thanks for the help!

Wouter Buytaert
PhD student
Katholieke Universiteit Leuven
Belgium



On Mon, 2003-04-14 at 15:47, Prof Brian Ripley wrote:
> On 14 Apr 2003, wouter buytaert wrote:
> 
> > Hello,
> > 
> > (maybe a quite basic statistical question, but I'm just struggling with
> > it)
> > 
> > I'm doing an anova:
> > 
> > > Res1<-aov(H2O~location+topo+horizon+pF+Error(location:topo:horizon))
> > 
> > (water retention of soils)
> > 
> > which gives a significant difference at factor "location".
> > 
> > Which function should I use to now which locations (there are 3: A, B
> > and C) differ significantly and which do not?
> 
> Possibly TukeyHSD or package multicomp.  However, we would need to know a 
> bit more about the design, as that's a rather peculiar Error() 
> specification and you would need to be careful to apply these functions in 
> the right stratum (assuming the location effect only occurs in one).
> 
> Function se.contrast would also be useful, and as there are only three
> pairwise differences, multiple comparison issues are going to be minor
> (especially as you have tested for an overall difference first).