178 



that Ray (1976b) regards any conclusions based on a limited number of skulls as being 

 extremely tentative (also Davies 1958b). This potential source of error impacts primarily 

 on such species as Mirounga angustirostris and Monachus monachus, where only a limited 

 number of specimens were available for study (see Appendix A). Fortunately, Kesner 

 (1994) indicates that beyond some minimal number of specimens, it is more advantageous 

 (with respect to error and statistical power) to increase the total number of characters 

 rather than to examine more specimens (to reduce the probability of including an incorrect 

 character state). As well, the modified majority rule algorithm used to determine the 

 consensus character states for each species should reduce the presence of any outright 

 erroneous or trivial states in the data matrix, while hopefully retaining the more important 

 polymorphisms. 



As well, the type of data employed in this study might have had some effect on the overall 

 result obtained. The discordance between phylogenies derived from morphological versus 

 biochemical or molecular data has long been noted (see Hillis 1987; Patterson et al. 1993). 

 This conflict has been attributed to the different assumptions and methods of analysis 

 inherent for each data type (Hillis 1987), but may also derive from the fact that 

 morphological and molecular data are apparently the most effective at phylogenetic 

 reconstruction at different taxonomic levels. In assessing attempts to resolve the 

 phylogenetic relationships of the even-toed ungulates (Mammalia: Artiodactyla), Novacek 

 (1993) pointed out that molecular data have produced good resolution for the internal 

 (intra-ordinal) relationships of the group, but not for its outgroup (inter-ordinal) 

 relationships, where morphological data have performed better. 



Apparently, molecular data appear to work better at the lower taxonomic levels (see 

 Irwin & Arnason 1994: 53). At progressively higher levels, the accumulation of neutral 

 changes should tend to obscure any phylogenetic signal, being visualized either as a 

 reasonably high amount of homoplasy, or as a general lack of resolution (although 

 conservative molecular regions may be more immune to this potential problem). 

 Conversely, morphological data appear to work better at higher taxonomic levels. As well 

 as being potentially afflicted by some or all of the problems mentioned by Arnold (1981), 

 morphological data do not appear to be discriminatory enough to pick out some of the 

 fine differences required for a species level analysis. Largely, this seems to stem from the 

 traditional use of fairly obvious, but simple morphological characters (e.g., presence or 

 absence of various processes, foramina, ...), which, when combined with the general 

 predisposition towards binary characters, will ignore a vast assemblage of more complex, 

 and potentially more finely discriminating, features (e.g., shape features as elucidated by 

 some form of morphometric analysis). Apart from the increased effort required to acquire 

 these more complex characters, their acceptance has been hindered by the perception that 

 the features are somehow not "real" or discrete, and therefore not subject to natural 

 selection in the same way. This latter point arises from our supposed functional and 

 evolutionary "understanding" of morphological features, allowing us to a priori eliminate 

 potential characters that might be unsuitable for various reasons (e.g., too homoplastic, 

 too trivial). 



This conflict between the two data types may have been manifested somewhat in this 

 study with the acknowledgement that the pattern of outgroup relationships was generally 



