51 
Rather, clustering analyzes the data on an ordinal scale 
(Hartigan, 1975). Cluster analysis has been recommended for 
studies of semantics (Miller, 1969; Rapopart & Fillenbaum, 
1972; Friendly, 1977). It is important to note that with 
cluster analysis, as with factor analysis, there are no best 
or unique solutions. 
For the data of each of the four judgement schemes, 
distance matrices were generated by averaging the absolute 
difference between the mean scores for each subject for all 
pairings of criteria. (See Appendix M for a listing of the 
program for generating the distance matrices.) Cluster 
analysis was done with BMDP sub-program P1M (Hartigan, 
1981). The clustering method used was complete linkage. 
With complete linkage, the distance between any two clusters 
is the largest distance between pairs of criteria in each of 
the CLGSLETT. Complete linkage produces compact clusters 
and avoids the chaining effects common with single linkage 
(Mardia, Kent & Bibby, 1979). In a study of semantics using 
cluster analysis, Miller (1969) found no significant 
differences between single and complete linkage. The 
results of the four cluster analyses appear as dendrograms 
in.Figures 10, 11, 12, and 13. 
Operationally, in this case, cluster analysis is a 
systematic examination of the mean absolute differences