no. 3645 PORTUNIDAE — STEPHENSON, WILLIAMS, LANCE 3 



type-species of the genus Portunus, P. pelagicus, is considered, a 

 general classical impression is gained that Callinectes species are 

 closer to P. pelagicus than are the bulk of the American species of 

 Portunus. Inclusion of P. pelagicus suggested to us that additional 

 Indo-West Pacific species should be considered, involving several 

 species close to P. pelagicus, P. macrophthalmus for comparison with 

 P. tuberculatus, and Scylla serrata exemplifying another related genus. 

 The number of species eventually compared (44) is sufficiently 

 large to give convenience to numerical techniques but not too large 

 for the conclusions from each technique to be checked against the 

 "common sense" of the classical background. With such comparison 

 possible, we found it not surprising that the overall outlook on the 

 group has not been changed materially. The important conclusions, 

 therefore, are in the field of methodology. It was hoped that a method 

 would be developed that could be applied to the very numerous 

 Indo-West Pacific species of Portunus, whose complex interrelation- 

 ships are difficult to determine by the traditional approach. 



Numerical Methods 



Form op data. — Basic taxonomic data normally are mixed: they 

 may, for example, comprise attributes that are qualitative ("yes" 

 or "no", "present" or "absent"), multistate (A, B, or C), ordered 

 multistate or ranked ("absent", "rare", "common"), and numerical 

 (measured). Few numerical models capable of accommodating all 

 these approaches are known; and, although computer programs 

 using such models exist, they are relatively inflexible and allow little 

 or no choice of alternative approaches in an exploratory situation. 

 There are, therefore, advantages in using simpler types of data if 

 this is practicable within the problem under study. When very 

 closely-related organisms are concerned — for example, in intra- 

 specific comparisons — the investigation normally involves measured 

 characters, and these alone may suffice. Interspecific comparisons 

 usually involve qualitative differences, and it may be advantageous 

 to reduce all the data to the qualitative form. The advantages are: 

 first, data can be tabulated in an extremely economical form, which 

 permits rapid intuitive assessment of taxonomic similarities; and 

 second, numerical systems for processing qualitative data are power- 

 ful, fast, and flexible, and their properties are well understood. 



Certain problems, nevertheless, remain to be resolved: (1) 

 decision must be taken as to whether double-negative matches are 

 to count as evidence of similarity (past experience of numerical 

 classifications suggests that they should do so, and the programs 

 at our disposal all make this assumption). (2) Provision must be 



