28 



W. A. SANDS 



of clustering and incidentally to find out whether any adaptive features of the 

 same structures of one caste were expressed as dimensions of variation clearly 

 independent of the other. In addition, by using only a selected part of the data, it 

 provided a check on the validity of character weighting obtained from the other 

 analyses. 



In this combined analysis of the imago and worker characters the lowest correla- 

 tions occurred, as expected, between those of different castes and present in only 

 one of them. Imago antennal articles, for example, were weakly correlated with 

 worker fore tibia width. Since these probably represent adaptations to totally 



PRINCIPAL COMPONENT WEIGHTINGS 14 



CHARACTERS MEASURED 



eigenvector 



(TjWidth of head across compound eyes CO 



ye. greatest diameter 

 cellus. least diameter 

 cellus, greatest diameter 

 cellus to eye distance 

 ronotum, width 

 ronotum. length 

 ind tibia, length 

 ostclypeus, width 

 ostclypeus, length 



CANONICAL VARIATE WEIGHTINGS 

 eigenvector 



ntennal ar 

 ntennal ar 

 ntennal ar 

 ntennal ar 

 ntennal ar 

 ntennal ar 

 eft mandib 

 eft mandib 

 ft mandib 



coefficient levels: 



C one 

 e two 



c three 

 e four 

 e five 

 c nine 

 e- apical to 1st marginal 

 e- 1st to 3rd marginal 

 e- 3rd marginal to molar 

 ight mandible - apical to 1st marginal 

 \\\\2JjRight mandible- 1st to 2nd marginal 



g2)Right mandible- 2nd marginal to mola 

 23lMesonotum, length 

 24)Mesonotum, width at constriction 

 25jMetanotum, width at constriction 



2 5-3 5ZZT_ ! 3-5-4-5J 



over 4-5 



Fig. 14. Imago caste-weightings of measured characters obtained from two multivariate 

 analyses, principal components and canonical variates. The weighting co-efficients of 

 eigenvector 1 of the principal component analysis were all less than 2-5 and closely 

 similar in value. 



