53 



from very different sources (e.g., morphological versus molecular data), or whether or not 

 one can even justifiably pool data matrices of very different signals (see Bull et al. 1993), 

 total evidence does, in principle, provide the advantage of a more varied data set that 

 presumably provides a better and/or wider representation of the suite of all possible 

 characters. However, we are left with the same dilemma, now only one step further 

 removed. So long as the set of characters is non-random, the tests we currently have will 

 not place confidence intervals on the true phylogeny. 



Therefore, we- will interpret our results according to what the various tests maximally 

 indicate: character covariation (PTP and skewness tests), or how strongly our data present 

 some underlying distribution which may or may not be the true phylogeny (bootstrap and 

 support analyses, as well as successive approximations and constraint analyses). The true 

 "test" lies in the Character Analysis section, where the set of characters producing this 

 distribution are individually presented and described. Implications as to the overall 

 accuracy of our solution (with respect to the one true phylogeny) are not intended. But, 

 since we naturally feel that our data set is one of the best available in terms of taxonomic 

 rank examined, number of taxa (both ingroup and outgroup), range of morphological char- 

 acters (with the acceptance and inclusion of polymorphic data), and general inclusiveness, 

 we feel that it provides one of the better estimates of the true phylogeny of the phocid 

 seals. However, with no knowledge as to how random our character set is, we make no 

 pretense as to the accuracy of our solution. 



150001 



3 



o 



10000- 



5000- 



0-1— 



85000 



95000 100000 105000 



90000 



110000 



Length of randomly generated tree 



Fig.7: Frequency distribution of tree lengths for a random sample of 1 ,000,000 trees generated from 

 the inversely weighted data matrix. Skewness statistic (g,) = -0.503. 



