Constraining Nonmetric Multidimensional Scaling Configurations.

Abstract

The interpretation of multidimensional scaling outputs is usually based on the identification and labeling of geometric structures in space. Some of the most commonly used structures are reviewed. Interpretation of the scaling outputs requires many psychological and mathematical assumptions including the assumption that the configuration with the lowest stress is the output desired. Unfortunately, little is known about the uniqueness of a configuration generated from fallible data and this non-uniqueness also affects the interpretation of the spatial outputs. A scaling method incorporating information in addition to the dissimilarities is proposed and the implications of this approach for the interpretation of a configuration are discussed. (Author)