[R] use aov or lme for split plot design?

[C.Schaedel at unibas.ch]

Dear all

I'm not sure if I did the right analysis for my specific split splot  
design. We are
studying biomass increase with different CO2 concentrations with four  
different
functional plant groups (e.g. grasses, herbs, broad-leafed trees and  
conifers). Of each
functional plant group we have four species. The design is orthogonal.

The design is:

Blocks: 2 (climate chambers, called Gruppe)

Plot: 3 (3 CO2 concentrations)

whole plot treatment is CO2

in each chamber we have the same 4 functional plant groups (funktGr),  
each represented by
four species (the species are nested in the functional plant groups,  
because if I take
species as a single factor I get 16 species, which is not true.

in each chamber I have four replicates (pseudoreplicates)

half of the plants got fertilisation (Fert), that means each plant got  
it's own
fertilisation

the response variable is biomass (g)

Randomisation was done to plot and subplot level.

I tried this:

aov(log(g)~CO2*funktGr/Species*Fert+Error(Gruppe/CO2),data=biomass)

the output is:


Error: Gruppe
           Df  Sum Sq Mean Sq F value Pr(>F)
Residuals  1 0.11923 0.11923

Error: Gruppe:CO2
           Df  Sum Sq Mean Sq F value   Pr(>F)
CO2        2 14.0297  7.0149  872.65 0.001145 **
Residuals  2  0.0161  0.0080
---
Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1

Error: Within
                          Df  Sum Sq Mean Sq   F value    Pr(>F)
Fert                      1   0.042   0.042    1.3379 0.2503649
funktGr                   3 249.764  83.255 2639.6236 < 2.2e-16 ***
CO2:Fert                  2   0.030   0.015    0.4830 0.6184632
CO2:funktGr               6   1.415   0.236    7.4747 1.501e-06 ***
Fert:funktGr              3   0.696   0.232    7.3551 0.0001777 ***
funktGr:Species          12  16.813   1.401   44.4214 < 2.2e-16 ***
CO2:Fert:funktGr          6   0.480   0.080    2.5386 0.0254449 *
CO2:Fert:funktGr:Species 60   2.045   0.034    1.0809 0.3636501
Residuals                93   2.933   0.032

Questions:

- Can I nest the factor Species in funktGr the way I did? It doesn't  
make sense to
analyse Species as a main factor in the whole analysis, I would need  
to do four separate
analysis within a functional plant group, is that correct?

- Do I need to average over the four pseudoreplicates for each chamber  
or is the
pseudoreplication taken out by the Error term?

- I often found in the R-helps that people use the lme functiona  
instead of aov. Which
one is more appropriate?

- with the aov function can I also eliminate the non significant  
factors to simplifiy the
model?

Thanks a lot for your answers

Christina
-- 
Christina Schädel
Institute of Botany, University Basel
Schönbeinstrasse 6
CH-4056 Basel
ph. +41 61 267 35 06
fax +41 61 267 29 80
E-Mail C.Schaedel at unibas.ch