54 



Character covariation within the data set 



Both the PTP and skewness tests produced highly significant results, indicating very strong 

 character covariation within the data set. Using 99 randomly permutated data matrices, 

 the PTP statistic (or a) was calculated as 0.01. The more accurate estimates proposed by 

 Källersjö et al. ( 1992) showed even lower values: 6.0 x 10~ for a" and 2.2 x 10~ 1S for a*. 

 In accordance with Källersjö et al. (1992), this last, lowest value is held to be the best 

 estimate of the level of significance here. The critical PTP length value corresponding to 

 the 0.05 level was determined to be 79,619 steps, some 9,785 steps (142 corrected steps) 

 longer than the overall parsimony solution. The shortest and longest trees derived from 

 the permutated matrices were 79,227 (+ 9.393 steps / 137 corrected steps) and 80.605 

 steps (+ 10.771 steps / 157 corrected steps) respectively. 



Skewness tests echoed the findings of the PTP test. A frequency distribution of tree lengths 

 from a random sample of 1.000.000 trees possessed a value of -0.503 for g, (Fig. 7). This 

 was judged to be significant at the 0.05 level based on a critical value for g l (25 taxa and 

 250 binary or four-state characters) of -0.08 (Hillis & Huelsenbeck 1992). The shortest 

 random tree obtained, 85.170 steps, was 15.336 steps (223 corrected steps) longer than 

 the overall solution. As suggested by Källersjö et al. (1992). the sample size used here 

 was apparently not sufficient to sample trees from the attenuated left-hand tail of the 

 distribution. 



Regional support within the overall solution 



While the previous subsection demonstrated significant character covariation throughout 

 the data set as a whole, not all regions within the resultant solution will necessarily be 

 equally supported by this set of covarying characters. The remaining statistical tests show 

 considerable agreement as to the regional localization of stronger and weaker signal within 

 the data set. This more localized signal, it appears, is sufficiently strong and/or widespread 

 to override regions of weak support and be manifested at the level of the entire solution 

 (see above). 



Bootstrap analysis (Fig. 8) 



The majority rule consensus tree obtained from a bootstrap analysis of 1.000 replicates 

 (Fig.8A) agrees quite strongly with the overall parsimony solution. The various goodness- 

 of-fit statistics are virtually identical between the two solutions, and the bootstrap solution 

 at 70.006 steps is only 172 steps (= three corrected steps) longer than the overall solution. 

 Only two major topological differences were observed, one in each phocid subfamily. In 

 the monachines. Leptonychotes moved to a more terminal position to form a clade with 

 Lobodon. This clade now becomes the sister group to Monachus spp. In the phocines. the 

 clade composed of Erignathus, Histriophoca, and Pagophilus moved basally to form the 

 sister group to Phoca spp. plus Pusa spp. Within this latter clade, Pusa is indicated to be 

 monophyletic, with Pusa caspica again being held to be basal to the remaining species of 

 the genus. Phoca remains paraphyletic and related by symplesiomorphies. with Phoca 

 largha maintaining its more basal status within the genus. However, as the bootstrap 

 solution is based on statistical considerations and not parsimonious ones (as well as being 



