1978 Volume 20 Issue 1 Pages 7-17
(compatible with EndNote, Reference Manager, ProCite, RefWorks)
(compatible with BibDesk, LaTeX)
A maximum likelihood estimation procedure is developed for nonmetric multidimensional scaling (MDS) which applies to the situation in which all empirical pairwise orderings of dissimilarities are assumed to be independent. The proposed method, while formulated within Thurstonian framework, does not presuppose initial unidimensional scaling of “observed” distances. Rather, the original nonmetric data (which are the set of empirical ordinal relations on the dissimilarities between stimuli) are directly related to the representation model (which is a distance function of some form) through a single optimization criterion based on the maximum likelihood principle.