Multidimensional scaling plot: point grouping/clustering

[#1]

I'm using multidimensional scaling on a 91 x 91 distance matrix. When the 2D MDS plot is displayed, the points are plotted using various different symbols and colors assigned by JMP, evidently clustering points togethers based on their proximity in the multidimensional distance space (see attached screenshot). Questions:

1. What is the criterion that JMP is using to decide how to group or cluster points?
   
   
2. Is there any way to adjust the criterion or threshold for how points are clustered?
   
   
3. Is there any easy way to get a report on how many clusters have been found (other than manually counting them on the plot; not practical for large data sets)?

I have searched JMP documentation and Google and have not been able to find anything on this.

I'm looking (preferably) for answers that are not dependent on scripting.

Thanks in advance for any assistance.

[#2]

I suggest that you select in JMP: Help > Books > Multivariate Methods. A chapter is devoted to MDS. The last section of the chapter presents statistical details.

[#3]

Thanks for the suggestion@markbailey, but the book chapter sheds no light at all on these questions.

[#4 · Accepted solution]

There are a couple of understandable misconceptions in your questions.

The colors and markers are not generated by the Multidimensional Scaling platform. They are the row states already assigned in your data table.

MDS doesn't do any clustering or grouping. (That's why you can't find it in the documentation). To paraphrase Wikipedia: MDS translates the pairwise 'distances' from your data into an abstract cartesian space.

If you want a clustering algorithm you can use the K-means Clustering platform. It will allow you to adjust the number of clusters and report on which rows belong to which cluster.

[#5]

Ah, now I see what happened. I had in fact already used hierarchical clustering with this same data set (many months ago). Recently a colleague suggested I try MDS as an alternative approach for the type of analysis I needed. I reopened my saved JMP file and ran the distance matrix through MDS, not realizing that the cluster labels generated by the cluster analysis were still in the file. So I was seeing the 2D plot from MDS, but labeled (without my realizing it) according to the output from the previous cluster analysis. So, thanks to your response, I now understand what happened. When I closed all files and re-imported the distance matrix, the labels and color coding all disappear, as expected.

Thanks very much-- it all makes sense now.