[R] Decomposing tests of interaction terms in mixed-effects models

[Andrew Robinson]

On Mon, Aug 04, 2008 at 02:51:48PM +0200, Peter Dalgaard wrote:
> Andrew Robinson wrote:
> >On Mon, Aug 04, 2008 at 10:17:38AM +0200, Peter Dalgaard wrote:
> >  
> >>Andrew Robinson wrote:
> >>    
> >
> >That is a neat idea, thanks, Peter, but it doesn't quite fit the bill.
> >The summary provides t-tests but I am hoping to find F-tests,
> >otherwise I'm not sure how to efficiently test A (3 levels) at the two
> >levels of C.  
> >
> >The anova.lme function doesn't help, sadly:
> >
> >  
> >>anova(lme(Y ~ M2 - 1, random = ~ 1 | Block, data = example))
> >>    
> >   numDF denDF F-value p-value
> >M2     6    25 23.0198  <.0001
> >
> >so I'm still flummoxed!
> >
> >Andrew
> >  
> You do have to peek into M2 to know that the test is for whether the 
> last two coefs are zero, but how about
> 
> > M3 <- M2[,2:4]
> > M4 <- M2[,5:6]
> > anova(lme(Y ~ M3+M4, random = ~ 1 | Block, data = example))
>            numDF denDF  F-value p-value
> (Intercept)     1    25 10.66186  0.0032
> M3              3    25 55.31464  <.0001
> M4              2    25  1.27591  0.2967

Marvelous, many thanks, Peter.

> Also, check out estimable() in the gmodels package.

Will do.  Actually had done, but will do again.

Cheers

Andrew

-- 
Andrew Robinson  
Department of Mathematics and Statistics            Tel: +61-3-8344-6410
University of Melbourne, VIC 3010 Australia         Fax: +61-3-8344-4599
http://www.ms.unimelb.edu.au/~andrewpr
http://blogs.mbs.edu/fishing-in-the-bay/