Some Missing Data Patterns for Multidimensional Scaling

Most tasks used to gather information for multidimen sional scaling analysis are quite difficult for people to perform. An experiment was run to determine if sys tematic limits existed in such data collection situations and to determine the form that these limitations as sumed. The solution obtained from a complete order ing of stimuli to targets, using the conditional rank or der paradigm, was compared to solutions obtained from certain partial orders, constructed from the com plete orders by setting certain rankings equal. The par tial orders were found to reproduce the complete order solution quite accurately when about one half of the information was eliminated. The information elimi nated about similar items produced more differences in the obtained solutions than did the information about dissimilar stimuli. Suggestions about efficient techniques for gathering information for multidimen sional scaling purposes are discussed.