65 



(including the overall parsimony analysis) being based on the same data matrix, or some 

 sample(s) thereof. Combined with all of the procedures using some form of parsimony 

 criterion, it is not too surprising that they all indicate roughly the same solution, as they 

 are merely summarizing the underlying distribution of the matrix in slightly different ways. 



As exemplified by the results of the bootstrap analysis, the common finding is for relatively 

 strongly supported outgroup relations, with support dropping off markedly within each 

 phocid subfamily (some fairly robust species pairs therein notwithstanding). However, we 

 would postulate that the region of weaker signal is limited even further to that portion of 

 the cladogram near the polytomy within the Phocini (plus Erignathus). Despite the 

 comparably weak bootstrap frequencies generally present in both subfamilies, the pattern 

 advocated in the previous section for the monachines appears to be remarkably robust and 

 survives largely intact in both the support and successive approximations analyses. A 

 monophyletic Monachus, in particular, seems to be very robust. In contrast, the pattern 

 within the Phocini (plus Erignathus) is more labile, with almost every analysis holding 

 for a slightly different set of relationships. Although the membership of the group in 

 question {Erignathus, Histriophoca, Pagophilus, Phoca spp.. and Pusa spp.) is constant, 

 as is its monophyletic status, only the Erignathus, Histriophoca, and Pagophilus clade 

 appears to have any consistent support. Overall, this set of conclusions could also be 

 reached by merely examining the number of synapomorphies supporting the various nodes 

 within each phocid subfamily (Fig.5C). The nodes within the Monachinae are more 

 strongly supported in this respect than are those within the Phocinae, and especially those 

 within the Phocini (plus Erignathus). 



Therefore, it was somewhat surprising that a test aimed directly at elucidating this weak 

 region (constrained skewness) did not identify it as such (Fig. 12). Constraining the 

 stronger, and therefore supposedly more informative, "anti-Phocini*' (Fig.l2A) did not 

 eliminate a significant left-hand skew in the distribution as expected [g, = -0.483 

 (ACCTRAN) or -0.368 (DELTRAN); critical g, = -0.29 or -0.22 at the 0.05 level for nine 

 taxa and 250 binary or four-state characters respectively]. Thus, there would appear to be 

 greater support (i.e.. character covariation) within the Phocini (plus Erignathus) than the 

 remaining tests indicate, as skewness seems to be very sensitive to minute amounts of 

 covariation (Hillis & Huelsenbeck 1992). As well, the g { s for the "anti-Phocini'" test are 

 approaching their respective critical values to a greater extent than we have ever witnessed 

 in a skewness test, indicating some reduction in the level of character covariation, but not 

 to non- significant levels. Finally, the constrained skewness test does appear to be working 

 properly (within the suspect nature of PAUP's RANDOM TREES subroutine), as the 

 reciprocal constraint of the weaker "E-Phocini" (Fig.l2B) produced the expected signifi- 

 cantly left-hand skewed distribution [g, = -0.540 (ACCTRAN) or -0.541 (DELTRAN); 

 critical g, = - 0.08 at the 0.05 level for 25 taxa and 250 binary or four-state characters]. 

 However, in order to more rigorously test this last supposition, random clades of a fair 

 size (say six or seven taxa) should be constrained, and the skewnesses of the resulting 

 distributions analyzed. 



