Message-ID: <efb536d50812300438q4e492256y26486d12e8421043@mail.gmail.com>
Date: 2008-12-30T12:38:43Z
From: Sarah Goslee
Subject: why stress value remains so high after invoking of metaMDS
In-Reply-To: <2f5200a00812292206n45e29e90q9b8510c59b3de602@mail.gmail.com>
If cm is a similarity matrix, why are you taking its Euclidean distance?
(The usage I'm familiar with has similarity as a pairwise measure of
association.)
Otherwise, if you feel the stress is too high, that implies that a 2-dimensional
solution is inadequate for your data and you should consider more dimensions.
Sarah
PS It's not necessary to repost; not everyone is here 24 hours a day, especially
over what for most of Europe and the US are holidays. A question may sit
for a few hours (or *gasp* days) until someone with the right expertise checks
in.
On Tue, Dec 30, 2008 at 1:06 AM, Wu Chen <geminiwhu at gmail.com> wrote:
> Hello everyone!
>
> metaMDS(cm, distance = "euclidean", k = 2, trymax = 50, autotransform
> =TRUE, trace = 1, plot = T)
> (cm is a similarity matrix, in which values are positive integers or 0)
>
> I use this command to run NMDS on my matrix "cm". But the stress is very
> high after analysis. About 14.
> Actually, there is no improvment comparing with using isoMDS.
>
> cd<-dist(cm,method="euclidean")
> loc<-isoMDS(cd,tol = 1e-10,trace=T)
>
> Is there parameters that I can change to improve the performance? Or is
> there any other better methods to do MDS?
> --
> Wu Chen
> Information Management School, WHU
>
--
Sarah Goslee
http://www.functionaldiversity.org