USING DISTANCE INFORMATION IN THE DESIGN OF LARGE MULTIDIMENSIONAL-SCALING EXPERIMENTS

Studied which distances are most important in determining the recovery performance of a nonmetric multidimensional scaling algorithm. Using Monte Carlo methods, it is shown that the large distances are critical to satisfactory performance, whereas the small and the medium distances play a much less crucial role. This finding has been reliably demonstrated across a variety of conditions, although only for a single combination of dimensionality and number of points. Parallels between this work and previous results obtained using cyclic and other incomplete designs are noted. On the basis of these results some recommendations to experimenters regarding data collection procedures are presented; these represent a simple alternative to the methods advocated by I. Spence and D. W. Domoney (see record 1975-10755-001). (8 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved). © 1979 American Psychological Association.