Multidimensional Scaling and Dimension Reduction
Multidimensional scaling is a statistical (or more properly data-analytic) technique that allows us to "squeeze" high-dimensional data down into a small number of dimensions, usually two or three. In two dimensions we can then map our data points in the plane, and this often gives us greater insight into the data, allows us to observe (possible) clusters, and suggests directions for further, more formal analysis. MDS is quite different from principal components analysis (PCA), but in some of the results of the two methods can be similar or analogous. I will discuss MDS and its variants, and its relationship to PCA and other dimension-reduction techniques, with real-world data examples and graphics.
Fee: None to members of the Cornell community, but registration is required. Since space is limited, early registration is encouraged.
Instructor: Professor Bunge, ILR Social Statistics
For times and locations of upcoming workshops, please see the Workshop Schedule.