THE SOLDIERLESS TERMITES OF AFRICA 25 



negative or low positive values of their corresponding elements of vectors 3 and 5. 

 Anenteotertnes (45-55) occupies the 'upper' half of the cluster having largely positive 

 or low negative values in these vectors. Adaiphrotermes forms a line of three points 

 (58-60) to one side. The nearest neighbours of some species in each genus are in the 

 adjoining genera, and this is accentuated by the close intermingling of the species 

 of all three that occurs in other dimensions all of which contribute to the taxonomic 

 similarities on which the cluster analyses are based. In these three genera which 

 are generally similar, the intestinal characters clearly separate them on a con- 

 ventional basis. Astalotermes has virtually no mesenteric overlap with the procto- 

 deum, or mixed segment. In Anenteotertnes there is a long mixed segment, also 

 present in Adaiphrotermes, but in the latter the mesenteron is dilated at its posterior 

 end and the fore tibia has only two apical spurs. The relationships of the genera 

 are discussed in more detail in the section on phylogeny. 



Multivariate analyses of measurements 



The shortage of other characters has led taxonomists to make increasing use of 

 measurements in studying termites. However, this has developed very largely 

 on a trial and error basis and a large number of differing measurements and ratios 

 (indices) have been employed. Roonwal (1964) noted that 66 measurements and 

 34 indices had been published and suggested a further 22 and 19 respectively bringing 

 the totals to 88 measurements and 53 ratios. He recognized that such numbers 

 were impractical and advocated a selection of the 'more useful and sensitive characters' 

 numbering 32 measurements and 18 indices 'for general use'. The method of 

 assessing sensitivity was not indicated. There has been some published discussion 

 on the numbers of individuals of any species or colony series that should be measured 

 to obtain a representative sample for statistical purposes. Bouillon (1966) con- 

 cluded that 6-10 specimens were sufficient. Variation within colonies of the soldier- 

 less termites was soon found to be slight, about 5% being the maximum range for 

 either sex or caste. In these small termites accurate measurements of small features 

 such as mandibular teeth are difficult to take. It was concluded from preliminary 

 tests with common, well represented species, that for the purposes of the multi- 

 variate analyses it was adequate to measure one or two specimens of each caste 

 from each colony. 



One of the first applications of multivariate analysis to a taxonomic problem 

 was made by Stroud (1953) using the complete centroid method of factor analysis 

 (a close approximation to principal component analysis) to sets of measurements 

 of the imago and soldier caste of Kalotermes species. Since then these methods 

 have been applied in many other plant and animal groups, but not in termite 

 systematics. 



The purposes of principal component analysis used in taxonomic problems may 

 be summarized as the extraction from a large body of primary data the smallest 

 number of meaningful dimensions; to interpret these where possible in terms of 

 recognizable characteristics, and to derive from their objective weightings a set of 

 taxonomic indices giving the best discrimination of individuals. The clustering 

 of individuals into groups can be examined, and misplaced specimens recognized. 



