R help - Jan 2004 - MANOVA power, degrees of freedom, and RAO's paradox

Hi,

I have a nested unbalanced data set of four correlated variables. When I 
do univariate analyses, my factor of interest is significant or 
marginally significant with all of the variables. Small effect size but 
always in the same direction. If I do a MANOVA instead (because the 
variables are not independent!) then my factor is far from being 
significant. How does that come about?

I have found a mention of a so-called Rao's paradox, which seems to deal 
with exactly this phenomenon. Does anyone know more about it, e.g. a 
reference?

The next strange thing is that if do the MANOVA in R, then both 
hypothesis and error degrees of freedom are multiplied by the number of 
variables. When I do it in SAS, however, only the hypothesis d.f. are 4 
x univariate, while the error d.f. are as in univariate, minus 3. This 
is irritating, in particular since no indication is given in the 
handbooks as to how degrees of freedom are calculated in a MANOVA? Can 
anyone tell me more about this? Are there different philosphies that are 
responsible for the differences between R and SAS?

I would be grateful for any help.

Regards, Oliver

R's manova is not intended to handle `nested unbalanced data': it
cross-references ?aov, which says the unbalanced part, and ?manova says
the nested part.

On Mon, 5 Jan 2004, Oliver Bossdorf wrote:
> Hi,
> 
> I have a nested unbalanced data set of four correlated variables. When I 
> do univariate analyses, my factor of interest is significant or 
> marginally significant with all of the variables. Small effect size but 
> always in the same direction. If I do a MANOVA instead (because the 
> variables are not independent!) then my factor is far from being 
> significant. How does that come about?
> 
> I have found a mention of a so-called Rao's paradox, which seems to
deal
> with exactly this phenomenon. Does anyone know more about it, e.g. a 
> reference?
> 
> The next strange thing is that if do the MANOVA in R, then both 
> hypothesis and error degrees of freedom are multiplied by the number of 
> variables. When I do it in SAS, however, only the hypothesis d.f. are 4 
> x univariate, while the error d.f. are as in univariate, minus 3. This 
> is irritating, in particular since no indication is given in the 
> handbooks as to how degrees of freedom are calculated in a MANOVA? Can 
> anyone tell me more about this? Are there different philosphies that are 
> responsible for the differences between R and SAS?
> 
> I would be grateful for any help.
> 
> Regards, Oliver
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
> 
> 
-- 
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