s-news
[Top] [All Lists]

Re: Multidimensional Scaling Techniques

To: "Bert, Jacobs" <b.jacobs@pandora.be>
Subject: Re: Multidimensional Scaling Techniques
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Fri, 29 Aug 2003 05:49:18 +0100 (BST)
Cc: s-news@wubios.wustl.edu
In-reply-to: <NGBBJAGJOLNNIEBEEJKCOELJCBAA.b.jacobs@pandora.be>
MDS normally works with dissimilarities, not similiarities.
This is true of isoMDS (sic) function (sic), as its help page says.
Remember MASS is support software for a book, and you do not appear to 
have consulted that reference material.

So you need to convert your input to a dissimilarity matrix.
It's possible that all you need is D = 1 - S.

On Thu, 28 Aug 2003, Bert, Jacobs wrote:

> I've found a good method to convert the cross-elasticity matrix into a
> similaritymatrix and if I use the Isomsd module from the MASS library I
> receive a normal result but not a usable one.
> 
> Because I'm working with cross-elasticities, the higher the value, the more
> similar brands should be and the closer they should be on a plot. On my
> plots they are very far away from each other.
> 
> How can I make believe the program that higher values result in higher
> similarities and eventually the closer they should be on the map.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>