450 



-8- 



2. Factor AnalytU 



The nethod of deriving feccors vaa hy roc«ciag a telected number 

 o£ principal coapoacntt uaing the well-known VArimax criteria, and thea 

 obtaining an obliqua tolution b«sed upon the Variaax loading* uaing the 



y , 



Promax method (7). McKeimell'r anAljraia was based upon factors obtained 

 by Variaax rotation; tb« Proaax method has the ability to define more 

 specific factors (i.e. the varLsnce of the factor loadings is increased) 

 but the factors so obtained are obllqua. This means that the scores 

 obtained on any one Promax factor may be correlated to some extent with 

 the scores on other factors. A range o£ factor solutions iron 2 factors 

 CO \M factors was obtained; by visual inspection the one which aost 

 closely matched HcKensell's 10 factors was the L2-f actor solution, and 

 this was carried forward to the next stage of cluster analysis. 



3. Cluster Analysis (_/^ 



A general description of the cluster analysis is givea in Appendix IV 

 (page 54). The method used gave all 12 factors resulting from the 



r 



factor analysis an equal opp6^Cunity of contributing to the cluster 



/ 

 solution, and took into account tbe loadings of all kl variables on each 



of the 12 factors. This represents' a deliberate departure from the 



method adopted by McKennell (6) who only, carried forward 8 of his 10 



factors to the clustering stage and ignored all factor loadings apart 



from those shown in Table 1. Tbe total sadt>le w»a randomly divided into 



two halves, and cluster analysis was carried out on each half independently 



producing solutions ranging from 2 clusters to 12 clusters. For each ^ 



c 

 solution in turn the similarity between the cluster profiles in one half ^ 



IS* 



I 

 of the sample and those in the other half was computed by means oE a o 



m 



