[R-sig-eco] multiple factors in vegan:betadisper?

[Etienne Laliberté]

You're welcome. But thanks should not go to me but to Marti Anderson,
who was kind enough to take some time to answer the question very
clearly in the first place. If you could add some notes about this to
the betadisper help page, that'd be great!

Cheers

Etienne


Le mardi 21 septembre 2010 à 23:30 +0100, Gavin Simpson a écrit :
> On Wed, 2010-09-22 at 08:37 +1200, Etienne Laliberté wrote:
> > Dear Cameron,
> > 
> > A while ago, I had asked a similar question to Marti Anderson. Here's is
> > what she had replied:
> > 
> >         "I have actually moved away from doing 2-way PERMDISP types of
> >         analyses. The reason for this is that there are inherent
> >         difficulties in testing homogeneity of dispersions across the
> >         main effects in cases where there is an interaction between the
> >         two factors in their *locations*, when tested using PERMANOVA.
> >         Think of it this way - suppose there is no differences in the
> >         locations of samples in multivariate space among the levels of
> >         factor A and in level 1 of factor B, but there are such
> >         differences (once again, in *location*) in level 2 of factor B
> >         (which would correspond to an interaction in PERMANOVA). Then,
> >         clearly, this would be detected as a difference in dispersion,
> >         even though it is entirely due to the locations of cells
> >         (combinations of factors) in the two-way design. Thus, it seems
> >         to me that for the two-way design, to test dispersions, one can:
> >         
> >         (i) test for differences in dispersion among the individual axb
> >         cells (which is simply a one-way test where the individual cells
> >         are treated as levels of a single factor) and
> >         
> >         (ii) test for an interaction in a two-way PERMANOVA and if this
> >         is NS, then one can examine the test of dispersions (or
> >         locations, for that matter) separately and independently for the
> >         two factors."
> 
> Thanks for posting this Etienne. I purposely wrote betadisper() to
> accept a single grouping variable because I was struggling to see how
> you would do the analysis with multiple factors and so handled the
> simple, easy case first. It has been on my TODO list to revist this, but
> Marti's comments are good enough for me to tick this one off the list,
> and perhaps add a note to the help page for betadisper() summarising the
> above points.
> 
> Cheers,
> 
> G
> 
> >         
> > Hope this helps.
> > 
> > Cheers
> > 
> > Etienne
> > 
> > 
> > 
> > 
> > Le mardi 21 septembre 2010 à 13:24 -0700, Cameron N. Carlyle a écrit :
> > > Dear all:
> > > 
> > > I am wondering if it is possible to implement multiple factors in betadisper
> > > {vegan} when trying to run a Permutational Analysis of multivariate
> > > dispersions (PERMDISP)?  To run a PERMANOVA using adonis you are able to
> > > enter a formula, but the usage of betadisper seems to only allow a single
> > > group.
> > > 
> > > If it isn't possible, does anybody know a work around?
> > > 
> > > Thank you,
> > > Cam
> > > --
> > > Cameron N. Carlyle
> > > PhD Candidate
> > > Botany,
> > > University of British Columbia
> > > 
> > > 	[[alternative HTML version deleted]]
> > > 
> > > _______________________________________________
> > > R-sig-ecology mailing list
> > > R-sig-ecology at r-project.org
> > > https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
> > 
> > 
> 


-- 
Etienne Laliberté
================================
School of Forestry
University of Canterbury
Private Bag 4800
Christchurch 8140, New Zealand
Phone: +64 3 366 7001 ext. 8365
Fax: +64 3 364 2124
www.elaliberte.info

>From October 11:
School of Plant Biology
The University of Western Australia
35 Stirling Highway
Crawley, Perth
Western Australia 6009