The first part dealt with scaling of
uj from data
djk corresponding to (
uj-
uk),
j,
k=1, 2, ...,
n, where
djk are not available for all possible pairs. The generality of the procedure in the sense that
djk can be obtained in various ways and the importance of incomplete data in practice were discussed. It was shown with an example that
uj obtained by the method proposed by Gulliksen (1956) is sometimes critically affected by choice of the origin and a method to overcome the instability was suggested. In the second part, a simple method was described to convert
djk into a scale of common unit when
djk are originally given with different units according to row
j. The method is applicable even when the data matrix is incomplete and the obtained
djk with a unified unit, complete or incomplete, can be used for unidimensional scaling as stated in the first parts, and also for multidimensional scaling of the set of
n stimuli.
View full abstract