[R] differences between aov and lme

[domerg]

Hi,

we have a split-plot experiment in which we measured the yield of crop 
fields. The factors we studied were:

B : 3 blocks
I : 2 main plots for presence of Irrigation
V : 2 plots for Varieties
N : 3 levels of Nitrogen

Each block contains two plots (irrigated or not) . Each plot is divided 
into two secondary parcels for the two varieties.
Each of these parcels is divided into three subplots corresponding to 
three ordered levels of nitrogen.

We found in Venables & Ripley  (Modern Applied Statistics with S-plus, 
3rd edition) the multistratum model for the same type of dataset but for 
three levels (without the "Irrigation" partition):
aov(Y~N*V+Error(B/V), qr=T)

which we adapted to our model:
aov(Y~N*V*I+Error(B/V/I))

In Pinheiro & Bates (Mixed-effect models in S and S-plus) and as we saw 
in the message "Re: lme and lmer syntax from Ronaldo Reis-Jr, Wed 26 Oct 
2005", we fitted also the mixed model :
lme(Y~N*V*I, random~1|B/V/I)

On a random simulated response Y, we didn't obtain similar results - 
only one factor with the same F-value - from the "aov" function
and the "lme" one (oppositely  to the  example used by Venables & Ripley 
and Pinheiro & Bates).

Is there a mistake in one of our two models or an explanation of this 
difference?

Thanks a lot in advance.


Caroline Domerg and Frederic Chiroleu
UMR 53 PVBMT (Peuplements Vegetaux et Bio-agresseurs en Milieu Tropical)
CIRAD
Pôle de Protection des Plantes (3P) - Saint-Pierre
Ile de la Réunion