Book Description
For certain types of problems in multivariate data reduction, seriation and scaling may be reasonable approaches. Given a collection of n objects, seriation techniques try to order these objects on a one-dimensional scale in the sense of assigning a rank from one to n to each object. Scaling techniques attempt to do more by assigning a numerical value to each object so that not only is order achieved but also some quantitative measure of relative closeness is computed. Similarity functions are employed to measure the 'closeness' between pairs of vectors. Two general approaches are considered encompassing five methods. Lastly a section is devoted to several estimation problems that arise from considering the similarities between pairs of vectors as random variables having certain underlying mean and covariance structures.