Help! repeating a within-subjects aov with lme

[Rob Campbell <rob@robertcampbell.co.uk>]

Hi,

I have a repeated measures data set where 5 animals are each tested on 5
sound durations and within each duration they are tested on 5 different
sound intensities (sound levels). The same 5 sound levels are used at 
each duration.
The data are at:
www.physiol.ox.ac.uk/~raac/ablated_level_dur.csv
abl.lev <- read.csv("ablated_level_dur.csv", header=T)
xyplot(prop.cor~level|duration,groups=animal,
       type="l",data=abl.lev)

I can analyse this in aov, correctly I think, but I can't quite get the 
same numbers to come out of a mixed-effects model with lme. I think I'm 
confused about specifying the random effects; or maybe I've done it 
right but lme just isn't going to return the same values. Can anyone 
help save my Thesis?
Many thanks,
Rob

Here it all is:


A.
I can do the within-subjects analysis in R using aov and I have an error
stratum for each main effect and interaction. Perhaps there is a simpler
way of specifying the error term and perhaps I want to do duration/level
rather then duration*level, but for now it's ok. The design is balanced.
animal and duration are factors. level is a continuous variable.

abl.lev.aov1 <-
  aov(prop.cor ~ duration*level +Error(animal/duration+animal/level+
       animal/duration/level), data=abl.lev)


I get these results back:
summary(abl.lev.aov1)
Error: animal
          Df   Sum Sq  Mean Sq F value Pr(>F)
Residuals  4 0.298673 0.074668

Error: animal:duration
          Df  Sum Sq Mean Sq F value    Pr(>F)
duration   4 1.24461 0.31115  74.244 4.008e-10 ***
Residuals 16 0.06705 0.00419

Error: animal:level
          Df   Sum Sq  Mean Sq F value  Pr(>F)
level      1 0.148720 0.148720  15.764 0.01653 *
Residuals  4 0.037737 0.009434

Error: animal:duration:level
               Df   Sum Sq  Mean Sq F value Pr(>F)
duration:level  4 0.031771 0.007943  3.9961 0.0196 *
Residuals      16 0.031803 0.001988


QUESTION: I want to repeat this analysis using a mixed-effects model 
with lme. I can get some of the way towards doing it but I'm stuck:
B.
If I run:
abl.lev.lm0 <- lme(prop.cor ~ duration*level,
                   random= ~level|animal/duration, data=abl.lev)

Then I get the same F-values for level and duration:level as the ANOVA
above. However, the duration:level p-value is different and the F value for
duration is different.
anova(abl.lev.lm0)
               numDF denDF   F-value p-value
(Intercept)        1    95 180.50491  <.0001
duration           4    16  70.77279  <.0001
level              1    95  15.76395  0.0001
duration:level     4    95   3.99606  0.0049

If I run:
abl.lev.lm1 <- lme(prop.cor ~ duration*level,
                   random= ~1|animal/duration/level, data=abl.lev)

The my duration F-value is ok but the level and interaction F-values are
shot. Does anyone know what I'm doing wrong. Shouldn't I be able to get 
the same numbers from lme as I got from aov?
Finally, why is "denDF" 16 for duration but 95 for everything else. 
Should level be 1 and 4?
anova(abl.lev.lm1)
               numDF denDF   F-value p-value
(Intercept)        1    95 151.92370  <.0001
duration           4    16  74.24458  <.0001
level              1    95 102.30594  <.0001
duration:level     4    95   5.46398   5e-04



--
  Rob Campbell   -   Research Student
                 -  in search of a Thesis

 Autistic Bacteriophage Research Group
 www.autisticBacteriophage.notlong.com

               Oxford

   ~  www.robertcampbell.co.uk  ~

                  *