[R-sig-eco] ΑΠ: constrained cluster analysis

[Gavin Simpson]

On Thu, 2011-07-21 at 00:00 +0300, Evagelopoulos Thanasis wrote:
> Thank you for your reply.

[ Please keep correspondence on list! ]

> Here are more details on the problem:
> 
> My data are macrobenthic invertebrates abundances from a transect of
> sampling sites across an environmental gradient (salinity gradient).
> 
> I tried both the CONISS and the CONSLINK linkage methods in the
> chclust function of the rioja R package.
> 
> I am used in having the y axis scaled according to the
> similarity/distance index used. I am not sure how to interpret the y
> axis of the dendrograms that chclust produces.
> 
> What does the dendrogram y axis measure in the case of the CONSLINK
> linkage method?

I haven't looked at the code - perhaps do so yourself or ask Steve
(Juggins, the maintainer), but it may well be the fusion distance in
terms of the actual distance coefficient used.

But does it matter what the units are? Aren't you really looking for
relatively large increases in fusion distance as indicators of
significant distances between groupings, regardless of how that fusion
distance is supplied?

See also the bstick() function in rioja which can help to guide the
number of groups to retain.

HTH

G

> Thank you again!
> 
> Athanasios Evagelopoulos
> ________________________________________
> Από: Gavin Simpson [gavin.simpson at ucl.ac.uk]
> Αποστολή: Τετάρτη, 20 Ιουλίου 2011 11:45 μμ
> Προς: Evagelopoulos Thanasis
> Κοιν.: r-sig-ecology at r-project.org
> Θέμα: Re: [R-sig-eco] constrained cluster analysis
> 
> On Wed, 2011-07-20 at 22:43 +0300, Evagelopoulos Thanasis wrote:
> > Hello,
> >
> > I want to do a constrained cluster analysis with R, if possible.
> >
> > However, the only function that I know, i.e. the chclust function of
> > the rioja package produces dendrograms with a y axis that is
> > apparently not in the scale 0-1 of the distance measure used.
> >
> > The "stratigraphically constrained clustering" in PAST software does
> > produce dendrograms with y axis scaled between 0 and 1. Can this
> > somehow be done with R?
> >
> > My data are from a series of stations across an environmental
> > gradient.
> >
> > Thank you!
> >
> > Athanasios Evagelopoulos
> 
> You don't provide even the bare minimum information to help you. The
> default method is CONISS, which forms constrained clusters that minimise
> within cluster sums of squares. That metric may be used on the axis, not
> the fusion "distance".
> 
> chclust has CONSLINK too which might work better for you, but it is hard
> to tell...
> 
> What code did you use?
> 
> G
> --
> %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
>  Dr. Gavin Simpson             [t] +44 (0)20 7679 0522
>  ECRC, UCL Geography,          [f] +44 (0)20 7679 0565
>  Pearson Building,             [e] gavin.simpsonATNOSPAMucl.ac.uk
>  Gower Street, London          [w] http://www.ucl.ac.uk/~ucfagls/
>  UK. WC1E 6BT.                 [w] http://www.freshwaters.org.uk
> %~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
> 

-- 
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%
 Dr. Gavin Simpson             [t] +44 (0)20 7679 0522
 ECRC, UCL Geography,          [f] +44 (0)20 7679 0565
 Pearson Building,             [e] gavin.simpsonATNOSPAMucl.ac.uk
 Gower Street, London          [w] http://www.ucl.ac.uk/~ucfagls/
 UK. WC1E 6BT.                 [w] http://www.freshwaters.org.uk
%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%~%