1
$\begingroup$

Let's say I have a year's worth of magazine issues (January, February, March, etc), and I want to visualize the differences among them. The classic example of multidimensional scaling (MDS) would have a triangular matrix of subjective difference scores between each possible pair of magazine issues. MDS is then applied to these difference scores.

However, what if instead of simple difference scores, I have a collection of objective sub-features for each issue to use for differentiation. For example:

  • Number of pages

  • Number of advertisements

  • Mean words / page

    and so on...

I ultimately wish to acquire some sort of distance between each magazine pair while accounting for each of these features.

How can I use MDS while also taking into account these features? Or is there an alternative technique that is more appropriate?

Improve this question
$\endgroup$
1
  • $\begingroup$ I don't understand what "sub-features" are, as opposed to you simply have some relevant variables. In addition, I don't see why you need to use MDS at all. Generally, you use MDS b/c all you have is a dissimilarity matrix & you want to build a kind of approximation of what the underlying feature space might look like (eg, its minimum dimensionality). In your case, you already know what the features are, so you can use them directly. If you want to visualize the locations of the issues in the (eg) 2D space w/ minimal distortion for some reason, you can do PCA & plot in the 1st 2 PCs. $\endgroup$
    – gung - Reinstate Monica
    Feb 16, 2014 at 0:47

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.