MDS Features 

About Clustan
Cluster Analysis
User Support
What's New
White Papers
Contact Us
ClustanMDS Features
ClustanMDS is a Windows implementation of MDSCAL and KYST (M&K) that were developed by Bell Labs in the 1970s for non-metric multidimensional scaling.  The original programs were written in Fortran and have been translated and optimised for Windows, with the following features:

Produces MDS configurations for 1000 cases or more.  Whereas M&K were limited to 100 cases and 6 dimensions, by using dynamic storage allocation ClustanMDS is limited only by the memory available in your PC.  For example, ClustanMDS uses about 10MB memory for a proximity matrix on 1000 cases.

ClustanMDS searches for MDS configurations with minimum stress in a series of random trials, whereas M&K produced only one solution per run.  It will be evident when running ClustanMDS that "local" minimum stress configurations are easily found, and that a good "global" minimum stress solution is usually found only after a large number of trials with different random starting configurations (see example).

ClustanMDS obtains a continuous MDS configuration for proximity matrices derived from any data, including binary or mixed variables, and allows for missing proximities.  Thus it can generate configurations for attitudinal data, survey respondents, and binary or multistate strings such as DNA arrays.

By standardizing the main analytical options and simplifying program code, ClustanMDS has been optimised for efficient running.  For example, 100 trials on a proximity matrix for 25 cases was completed in under half a second on an Intel Pentium 4 PC (2.4 GHz) running Windows/XP (see example).

Hierarchical and non-hierarchical classifications can be colour-coded on MDS co-ordinates for full data sets or for summary cluster models (see example).

Final MDS configurations are rotated to principal components for consistent presentation of the main components having maximum variance in two or three dimensions.

Key cases can be plotted relative to a cluster model, to aid discussion of their relativities.

ClustanMDS is distributed free with ClustanGraphics.