[R-sig-ME] Generalized randomized block design
leverkus
leverkus at ugr.es
Mon Jan 28 14:57:07 CET 2013
Dear R users,
I am struggling with the formulation in lme of a generalized randomized
block design with subsampling, and I would very much appreciate some
help. The experiment consists of 3 plots (of ca. 20 ha each) located at
different altitudes on a mountain slope. In each plot there are 9
subplots, which are 3 replicates of 3 post-fire wood management
treatments. In each subplot we sampled 8 transects for plants (except in
one subplot, where only 5 transects were sampled), and my response
variable is species diversity. In order to take account for the
experimental design and get the correct number of denominator degrees of
freedom, I am using (1|Plot/Subplot) in the random effects. Subplot is a
categorical variable which joins treatment names (treatments are "SL",
"NI", "PCL") and replicates (1,2,3): SL1, SL2, SL3, NI1... This gives me
the correct replication: 3 plots and 27 subplots. As for now, my model
looks like this:
lme(diversity~Treatment,random=~1|Plot/Subplot)
However, treatment effects are likely to vary with altitude, so I wish
to test for the treatment x plot interaction. This is where I am stuck.
By including plot as a fixed effect (diversity~Treatment*Plot) I have no
df to calculate the plot effect and this looks weird to me. Besides, I
want to have plot as a random effect. Could anyone give me some
suggestions? (I don´t mind using lmer instead.)
Thanks in advance,
alex
More information about the R-sig-mixed-models
mailing list