Error with mdscale

I was trying to use nonclassical MDS function 'mdscale'. I have read that non classical MDS allows incomplete distance matrix to use.I have used a dissimilarity matrix (dist_mat) with few missing entries, replaced by NaN.This is in upper triangular form for five subjects as shown below.Thus 'dist_mat' has 10 distances with unknowns represented by NaN. For Ex: dist_mat=[3.9714;NaN;NaN;3.6492;NaN ; 2.8870 ; NaN ; NaN ; NaN ;NaN] I am getting error with mdscale(dist_mat,2) as- "Cannot initialize with 'cmdscale' when there are zero weights or missing data. Initialize with 'random' or provide explicit starting points" I did not understand what this means, why cmdscale here? How to assign weights and how to initialize with 'random'? Please help, Thanks

Answer 1

[Peter Perkins]

>> help mdscale
 mdscale Non-Metric and Metric Multidimensional Scaling.
[snip]
'Start' - Method used to choose the initial configuration of points
          for Y.  Choices are:

'cmdscale' - Use the classical MDS solution.  This is the default.
            'cmdscale' is not valid when there are zero weights.
'random'   - Choose locations randomly from an appropriately
             scaled P-dimensional normal distribution with
             uncorrelated coordinates.
matrix     - An N-by-P matrix of initial locations.  In this
             case, you can pass in [] for P, and mdscale infers P
             from the second dimension of the matrix. You can also
             supply a 3D array, implying a value for 'Replicates'
             from the array's third dimension.

[cvp]

Thanks Mr. Peter Perkins,

I have seen its help but I could not understand. Actually here I want to know whether we can use an incomplete distance matrix? If yes how? Please give a small example.

According to my understanding cmdscale is used in classical MDS and mdscale in nonclassical MDS.

In mdscale where to use random,please explain with a small example