30 W. A. SANDS 



mandibles of both castes are closely correlated, there is a good deal of independent 

 variation in the proportions of marginal and molar parts of the worker mandibles. 

 These sections of the mandible are not merely more variable than those of the 

 imago, but proportionately more variable than their own apical sections. It is 

 not yet clear how to interpret this in terms of adaptation to feeding or other be- 

 haviour, and more detailed studies need to be made. 



The length of the worker postclypeus is similarly less strongly correlated with 

 that of the imago than might be expected. Since the postclypeus houses the 

 cibarial dilator muscles it is probably associated with the pumping or manipulation 

 of moistened soil particles. It is conceivable that this is associated in imagos with 

 tunnelling to found a new colony and otherwise little used, whereas in the worker 

 it must be used in a variety of ways throughout its life. The independent variation 

 could be a response to different situations. 



The independent dimensions of variation indicated by the correlation coefficients 

 were also shown in the elements of the eigenvectors corresponding to the first five 

 latent roots that together accounted for 89% of the trace of the correlation matrix. 

 The first vector as usual consisted of largely similar elements, associated with a 

 general size factor, and represented 76% of the variation. The second vector 

 had its largest elements contrasting the apical teeth of the imago mandible with the 

 molar measurements of workers. Other features given major importance by this 

 vector were the worker apical teeth and the length of the imago postclypeus. The 

 third and fourth vectors were mainly concerned with variation in characters only 

 present in one caste or the other. It was not until the fifth vector that the small 

 independent variation of the worker postclypeus received a weighting coefficient 

 of any size, and here it was involved in a contrast with several characters of imago 

 and worker mandibles. This vector corresponds to a latent root of less than 10 

 (07) representing only i-8% of the variance. This brief account serves to show 

 that the method may yield useful comparisons of homologous structures in different 

 castes of social insects. The detailed figures of the analysis are not appropriate 

 in this context, and the full data are retained for future reference at the BMNH. 



The canonical variates analyses carried out separately on imago and worker 

 castes demanded at least two representatives of each of the species included, and 

 some consequently had to be left out. Instead of the 60 putative species represented 

 in the foregoing analyses, the analysis of the imago covered 50, and that of the workers 

 34 species. Thus although the canonical variates gave some indication of the 

 grouping of species into genera so far as this is shown by measurements they were 

 necessarily incomplete. The larger weightings of these two analyses, although 

 naturally with different numerical values, were attached to essentially the same 

 set of characters as in the principal component analysis. This probably arose 

 partly from the fact that many species were represented by small numbers of 

 specimens and the dispersion of the ends of the mean vectors was thus closely 

 related to that of the individuals. However, the combinations of character weight- 

 ings in the eigenvectors differed slightly, suggesting differing emphasis on characters 

 involved in clustering species and on those of greater value for generic discrimination. 



The character weightings obtained from both principal component analyses and 



