29 



different properties of each data set. No objective benchmark has yet been able to delineate 

 strong from weak support [although the confidence intervals may be surprisingly large 

 (see Cavender 1978, 1981)], so all results can only be stated in relative terms (e.g., a clade 

 has stronger support than another, not strong support per se), and only for the data set in 

 question (Novacek 1991). Through the use of permutation, Källersjö et al. (1992) have 

 refined the concept of support to give it a more objective, statistical basis; however, as 

 the calculation of this total support is prohibitive using PAUP, it was unfortunately not 

 examined here. 



All trees in increasing increments of 69 steps (= one corrected step; see above) from the 

 most parsimonious tree length were retained using the KEEP command of PAUP in 

 conjunction with the heuristic search procedure described above. As a heuristic search 

 pattern was used, the number of trees retained at each step should be viewed only as a 

 rough estimate of the total number of trees of that length or shorter, rather than an exact 

 figure. Summaries at each length were viewed using both strict and majority rule consensus 

 algorithms (see above for advantages of each). Unfortunately, with the large number of 

 taxa examined here (and concomitant large number of possible trees), PAUP quickly ran 

 into memory limitations, and the support analysis could only be performed for solutions 

 up to and including four corrected steps longer than the most parsimonious length. 

 All these statistical tests are dependent on the power of the computer and/or the search 

 algorithm employed. In the case of all but skewness, it is important to minimally maintain 

 the searches as robust as the original so that the results will be roughly comparable. For 

 the PTP test, this is especially critical given that any less than optimal solution for the 

 random data sets will increase the probability of generating a significant result (Källersjö 

 et al. 1992). Similar errors can be anticipated for the remaining procedures as well. 



Comparative tools 



Relatively less attention has been paid to the various non-statistical means of inferring the 

 robustness of a cladogram. Of the comparative techniques described below, only the 

 constraint analyses really qualify as a (non-statistical) means of inferring the robustness 

 of a cladistic hypothesis. Although the remaining four "analyses" do indirectly indicate 

 the strength of the pattern of phocid phylogeny obtained herein, they are, more properly, 

 specific, interesting questions that arose in the course of this study. Each analysis is again 

 described in turn. 



Constraint analyses 



One invaluable feature of PAUP allows the user to constrain searches to satisfy (or not 

 satisfy) a given topology or range of topologies. Of its many possible uses (see Swofford 

 1993), topological constraints were used here to view how much less parsimonious a 

 desired set of alternative relationships forced the overall solution to be. In many ways, 

 this procedure is akin to support analyses, except that the shortest solution containing a 

 set of relationships not found in the most parsimonious solution(s) is desired here. 

 Major competing hypotheses of phocid phylogeny were identified from the literature and 

 tested here. These hypotheses apply to both outgroup and ingroup relations and the 



