Skip to content

aov and lme differ with interaction in oats example of MASS?

3 messages · Karl Knoblick, Peter Dalgaard, Brian Ripley

Original
Karl Knoblick
# 
Dear R-Community!

The example "oats" in MASS (2nd edition, 10.3, p.309) is calculated for aov and lme without interaction term and the results are the same. 
But I have problems to reproduce the example aov with interaction in MASS (10.2, p.301) with lme. Here the script:

library(MASS)
library(nlme)
options(contrasts = c("contr.treatment", "contr.poly"))
# aov: Y ~ N + V
oats.aov <- aov(Y ~ N + V + Error(B/V), data = oats, qr = T)
summary(oats.aov)
# now lme
oats.lme<-lme(Y ~ N + V, random = ~1 | B/V, data = oats)
anova(oats.lme, type="m") # Ok!
# aov:Y ~ N * V + Error(B/V)
oats.aov2 <- aov(Y ~ N * V + Error(B/V), data = oats, qr = T)
summary(oats.aov2)
# now lme - my trial!
oats.lme2<-lme(Y ~ N * V, random = ~1 | B/V, data = oats)
anova(oats.lme2, type="m")
# differences!!! (except of interaction term)

My questions:
1) Is there a possibility to reproduce the result of aov with interaction using lme?
2) If not, which result of the above is the correct one for the oats example? 

Thanks a lot!
Karl


      __________________________________  Alles was der Gesundheit und Entspannung dient. BE A BETTER MEDIZINMANN! www.yahoo.de/clever
Peter Dalgaard
#
Karl Knoblick wrote:
The issue is that you are using marginal tests which will do strange
things when contrasts are not coded "right", and in particular treatment
contrasts are not. Switch to e.g. contr.helmert and the results become
similar. Marginal tests of main effects in the presence of interaction
is not necessarily a good idea and they have been debated here and
elsewhere a number of times before. People don't agree entirely, but the
dividing line is essentially whether it is uniformly or just mostly a
bad idea. It is essentially the discussion of type III SS.
Some of us feel that type III sum of squares and so-called ls-means are
statistical nonsense which should have been left in SAS.
   -- Brian D. Ripley
      s-news (May 1999)

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On Thu, 28 Jun 2007, Karl Knoblick wrote:

        

            

      
      
      
    

        
That is the problem.  You need true contrasts, so use contr.helmert.  When 
I did so I got the same results from lme and aov for the anovas.

The question of what a 'marginal' AoV means without orthogonality is moot.
The sequential version is fine here.