UNIDIMENSIONAL SIMILARITY REVISITED

It is argued that the use of direct scaling methods to derive the perceptual magnitudes in the similarity function is inadvisable on two grounds. (1) Attention may shift from one way of rating to another and thereby affect the perceptual magnitude so that what is rated may depend on the method of rating. (2) Direct methods of scaling seldom give consistent data and present knowledge of the reason for this is limited. The preferred procedure then is to study solely the structure of the similarity matrices themselves. When this is done on four sets of published data it is found that the similarity mechanism of Eisler and Ekman gives a close fit.