A comparison of MDS, Cluster Analysis and Similarity Trees

for the representation of consumer judgments of similarity and reference data

Abstract

Research has been carried out to investigate the abilities of different models in mapping consumers' cognitive structures. The abilities of different judgment methods in capturing proximity data have also been examined and discussed. The research, using 15 kinds of drinks as stimuli, engaged four models, the Euclidean distance multidimensional scaling, the average linkage cluster analysis, and two newer models, proposed mainly by Amos Tversky, Shmuel Sattath and James E. Corter, called the additive similarity tree and extended similarity tree models. It also employed three proximity judgment methods, simple order ranking, pairwise similarity judgments and pairwise preference judgments.

The multidimensional scaling and cluster analysis models, although frequently used in representing consumers' cognitive structures, were generally found to be less informative and interpretable, when compared to the additive similarity tree and the extended similarity tree models.

The simple ranking judgments, although simple and easy to answer and analyse, was found to be not reliable, when compared with the more detailed pairwise judgments. The preference judgments have long been believed to be based on similarity judgments, but no evidence of such can be found in this research.

The underlying theories of the models, their limitations, the drawbacks and benefits of using each of them have also been discussed at the beginning of the dissertation so that the readers can also interpret and comment on the results of the research.

Acknowledge

I would like to show my gratitude to my dissertation supervisor, Dr. Anne Tomes of the University of Sheffield for her kind and patient guidance throughout the time this dissertation was being written. Special thanks are given to her advice and encouragement from time to time.

I would also like to thank Mr. Gerard Hodgkinson and Dr. Jo Padmore, especially for their help in making clear many psychological and statistical concepts. Their advice in using English, using the computer and the programs are much appreciated.

Last but not least, I want to thank my beloved wife, my parents and all those who have taken part in this dissertation and research project.