[R] unbalanced effects in aov

[Brooke LaFlamme]

Thank you for the suggestions from Professor Ripley and Steve Elliot. I see now why my data are unbalanced even though I don't have any missing data. 

I think I should use other methods designed for unbalanced data, but does using lme with plate as a random effect also help to fix this problem? I am still very new at this type of analysis. 

Thank you for the help.

Brooke 



-----Original Message-----

> Date: Tue Sep 18 12:38:37 EDT 2007
> From: "Prof Brian Ripley" <ripley at stats.ox.ac.uk>
> Subject: Re: [R] unbalanced effects in aov
> To: 
>
> On Fri, 14 Sep 2007, Brooke LaFlamme wrote:
> 
> > Hi, I have been having some trouble using aov to do an anova, probably 
> > because I'm not understanding how to use this function correctly. For 
> > some reason it always tells me that "Estimated effects may be 
> > unbalanced", though I'm not sure what this means. Is the formula I am 
> > using written incorrectly? Below is the code I am using along with the 
> > data:
> 
> [...]
> 
> > I am treating all the variables as factors (except for response, obviously).
> >
> > formula<-response~species+line%in%species+replicate%in%line+sex%in%species+plate
> > model<-aov(formula, data=my.data)
> >
> > This is the output:
> >
> >> model
> > Call:
> >   aov(formula = formula, data = my.data)
> >
> > Terms:
> >                     species        plate species:line line:replicate
> > Sum of Squares  0.0026469288 0.0000945202 0.0003320255   0.0002008000
> > Deg. of Freedom            2           11           27             10
> >                 species:sex    Residuals
> > Sum of Squares  0.0001383116 0.0006315465
> > Deg. of Freedom            3           66
> >
> > Residual standard error: 0.003093362
> > Estimated effects may be unbalanced
> >
> > Any help would be greatly appreciated as the R help documentation for 
> > aov does not address this issue.
> 
> For the benefit of those who are unable to appreciate 
> fortunes::fortune("WTFM"), the help page actually says
> 
>       'aov' is designed for balanced designs, and the results can be
>       hard to interpret without balance: beware that missing values in
>       the response(s) will likely lose the balance.  If there are two or
>       more error strata, the methods used are statistically inefficient
>       without balance, and it may be better to use 'lme'.
> 
>       Balance can be checked with the 'replications' function.
> 
> So let's do as it suggests:
> 
> > replications(formula, data=my.data)
> $species
> [1] 40
> 
> $plate
> plate
>   1  2  3  4  5  6  7  8  9 10 11 12
> 11 11 11 10  9 11 10 11 10 11  9  6
> 
> $`species:line`
> [1] 4
> 
> $`line:replicate`
> [1] 6
> 
> $`species:sex`
> [1] 20
> 
> and the problem will be clear to those who have read ?replications.
> 
> -- 
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595