20 



algorithm was reapplied. This was necessitated as the normal polymorphic consensus 

 between the "equally" most frequent states 0 and 01, for instance, is meaningless (i.e.. the 

 state "001"), and probably reflects a greater preponderance of state 0 in that particular 

 taxon. However, note that a polymorphic consensus could still result if two or more 

 singular states happened to be "equally" frequent. 



The overall effect of this algorithm was to produce many polymorphic taxa, something 

 fairly uncommon in phylogenetic analysis. It is unclear to us exactly why this is the case, 

 but it is likely done (whether through the selection of characters that yield monomorphic 

 taxa, through the algorithms employed to arrive at consensus states for the taxa, or by 

 simply coding polymorphic data as missing) to simplify the overall analysis. However, we 

 believe that the large amount of polymorphism that we observed to be natural and 

 important, with its undue restriction resulting in the loss of a great deal of potential 

 information. This same procedure was employed to collapse species into a higher level 

 taxon for the condensed analysis (see below). The final data matrix appears in Appendix C. 



Cladistic analysis 



A cladistic analysis (sensu Hennig 1966) of the final data matrix was conducted using the 

 parsimony program PAUP 3.1.1 (Swofford 1993). PAUP was also used to conduct the 

 many statistical tests and comparative tools employed in this study to judge the robustness 

 of the overall solution (see below). 



Despite its supposed increased objectivity over other systematic methods, a cladistic 

 analysis still entails a large number of assumptions, both about how the data are to be 

 treated and how the actual analysis is to be conducted. The numerous assumptions we 

 have made concerning the data (both characters and taxa), the implications thereof, and 

 their apparent advantages over alternative assumptions are described first. This is followed 

 by an explanation of both the search criteria and methods of summarizing the output that 

 were used. 



Assumptions concerning characters 



All characters were assumed to be of equal weight, and multistate ones were held to be 

 unordered. Although either case requires assumptions equal in magnitude to weighted or 

 ordered characters (Sober 1988; Barrett et al. 1991), they were resorted to out of simplicity 

 and/or ignorance. In the first case, equally weighted characters typically imply independ- 

 ence among characters (as co-dependent characters are accordingly down-weighted), 

 and/or characters of roughly equal importance, reliability, or quality [see Underwood 

 (1982) and Bryant (1989) for other uses of weighting]. However, this is not implied here. 

 As we could not objectively determine the degree of character independence, nor relative 

 character importance a priori, we adopted the simplest solution, that of equally weighted 

 characters. 



Indeed, we make no pretense as to the independence of our characters. By all being drawn 

 from the same organism, all characters will be correlated with one another to some degree. 

 However, to our knowledge, there has never been a test devised that quantifies the level 

 of character independence or correlation, nor has it ever been explicitly stated what level 



