R help - Apr 2009 - Kruskal's MDS results

Dear List,

I'm trying to interpret the results of the Kruskal's Non-metric
Multidimensional Scaling algorithm (isoMDS, MASS package).

The 'goodness of fit' is reported as "The final stress achieved (in
percent)".

What does this mean exactly? I've tried to google for an answer but I've
not come up with a definitive answer.

Regards,
Dieter


--
Dieter Vanderelst
PhD Student

Active Perception Lab
University of Antwerp
http://batbits.webnode.com/

Postal Address:
Prinsstraat 13
B-2000 Antwerp
Belgium

Hi Dieter,

I'll take a shot at this.
As I understand it, the stress is telling you how the ordination distances
compare with original dissimilarities that you calculated.

It is a measure how well your ordination has done in representing the
relationship of your sites. Note that the stress will differ depending on how
many dimensions are used. I believe the default is k = 2 in isoMDS.

Hope this helps,
Michael

> 
> Dear List,
> 
> I'm trying to interpret the results of the Kruskal's
> Non-metric Multidimensional Scaling algorithm (isoMDS, MASS
> package).
> 
> The 'goodness of fit' is reported as "The final stress
> achieved (in percent)".
> 
> What does this mean exactly? I've tried to google for an
> answer but I've not come up with a definitive answer.
> 
> Regards,
> Dieter
> 
> 
> --
> Dieter Vanderelst
> PhD Student
> 
> Active Perception Lab
> University of Antwerp
> http://batbits.webnode.com/
> 
> Postal Address:
> Prinsstraat 13
> B-2000 Antwerp
> Belgium
Michael Denslow

Graduate Student
I.W. Carpenter Jr. Herbarium [BOON]
Department of Biology
Appalachian State University
Boone, North Carolina U.S.A.

-- AND --

Communications Manager
Southeast Regional Network of Expertise and Collections
sernec.org

Dieter,

You could always try the "Classification, clustering, and phylogeny 
estimation"  list which often includes posts regarding MDS: 
http://lists.sunysb.edu/index.cgi?A0=CLASS-L

regards

Bob

Hi Dieter,
> I understand that the stress is a measure of how good the
> algorithm managed to represent the ordinal distances between
> items. And I also see why it's dependent on the number of
> dimensions. 
> I was hoping someone could tell me exactly what the formula
> for the percentual stress is. To me it's not clear how this
> metric is calculated.
If you have a copy of:
Venables, W. N. and Ripley, B. D. (2002) Modern Applied Statistics with S.
Fourth edition. Springer.
There is a brief section on the stress calculation. Also, I believe that the
methods was first outlined in Kruskal 1963.
http://www.springerlink.com/content/010q1x323915712x/

The you can see 
MASS::isoMDS
or
MASS::Shepard
to get the exact R calculations.

Hope I helped this time,
Michael
> 
> Regards,
> Dieter
> 
> 
> Michael Denslow wrote:
> > Hi Dieter,
> > 
> > I'll take a shot at this. As I understand it, the
> stress is telling
> > you how the ordination distances compare with
> original
> > dissimilarities that you calculated.
> > 
> > It is a measure how well your ordination has done in
> representing the
> > relationship of your sites. Note that the stress will
> differ
> > depending on how many dimensions are used. I believe
> the default is k
> > = 2 in isoMDS.
> > 
> > Hope this helps, Michael
> > 
> > 
> >> Dear List,
> >> 
> >> I'm trying to interpret the results of the
> Kruskal's Non-metric
> >> Multidimensional Scaling algorithm (isoMDS, MASS
> package).
> >> 
> >> The 'goodness of fit' is reported as "The final
> stress achieved (in
> >> percent)".
> >> 
> >> What does this mean exactly? I've tried to google
> for an answer but
> >> I've not come up with a definitive answer.
> >> 
> >> Regards, Dieter
> >> 
> >> 
> >> -- Dieter Vanderelst PhD Student
> >> 
> >> Active Perception Lab University of Antwerp
http://batbits.webnode.com/
> >> 
> >> Postal Address: Prinsstraat 13 B-2000 Antwerp
> Belgium
> > 
> > Michael Denslow
> > 
> > Graduate Student I.W. Carpenter Jr. Herbarium [BOON]
> Department of
> > Biology Appalachian State University Boone, North
> Carolina U.S.A.
> > 
> > -- AND --
> > 
> > Communications Manager Southeast Regional Network of
> Expertise and
> > Collections sernec.org
> > 
> > 
> > 
> 
>