R help - May 2008 - AMOVA results from ade4 different than in the reference publication

Hello,

I am trying to run some AMOVA analyses with the amova function in the 
package ade4.

When running the example dataset provided in ade4, I noticed a 
difference between the published results from the same data (Excoffier 
et al. 1992) and what ade4 calculates.

Below are the data for "within sample/population" from ade4 and from
the
haplotypic distance matrix in the paper:

source	name	Variance	% total variance	p-value	phi
ade4	within samples	0.478	75.390	1.000	0.246
paper	within populations	0.478	75.390	0.000	0.246

In the R output, it seems that the Std.Obs for "within samples" in
ade4
is very high and may be the reason why the difference is not 
significant. Std.Obs is not given in the publication.

Is the high p-value an error in ade4 or am I missing something?

Thanks already for your help!

Kim
-- 
___________________________________________
Kim Milferstedt, PhD
Postdoctoral Researcher
University of Illinois at Urbana-Champaign
Department of Microbiology
C207 CLSL
601 S. Goodwin Avenue
Urbana, IL 61801

phone: 001-217-244-0721
email: milferst at uiuc.edu

Kim Milferstedt <milferst <at> uiuc.edu> writes:
> 
> Hello,
> 
> I am trying to run some AMOVA analyses with the amova function in the 
> package ade4.
> 
> When running the example dataset provided in ade4, I noticed a 
> difference between the published results from the same data (Excoffier 
> et al. 1992) and what ade4 calculates.
> 
  A generic answer to this type of question is that you should contact
the package maintainers (see help(package="ade4") for their
e-mail addresses) -- especially for
these kinds of specialized questions (most R-helpers are not population
geneticists and wouldn't know an AMOVA if it bit them) ...

  On the other hand, running example(randtest.amova) [which is
what I assume you did, although you don't say] I do confirm
your results, and furthermore the plot that is produced looks
like the p-value *should* be very low -- and the std. obs.
is large (which I think is a *good* thing from the point of
view of rejecting null hypothesis -- seems like this is the
analogue of a t- or z-score), but negative.  Perhaps the package
authors did an upper 1-tailed test by mistake? (I would follow
through with this but it looks like the important stuff goes
through into C code and I can't take the time to dig that
deep right now.)

  Bottom line: looks like a mistake, take it up with
the package maintainers.

  good luck,
    Ben Bolker