L9 



present in museum collections. The vast majority (191 characters) were osteological, 

 originating from both the cranial (153 characters) and post-cranial skeleton (38 characters). 

 This disposition towards osteological cranial characters reflects both the high information 

 content of the skull in mammals, and the tendency of museums to preserve large mammals 

 as skulls only. The osteological characters were divided according to their general region 

 as follows: snout, 21; orbit and zygomatic arch, 35; palate and ventral side of snout 

 (excluding teeth), 18; basicranial region, 43; bony tentorium and bony falx, 5; dorsal 

 braincase, 4; teeth, 23; mandible (excluding teeth), 3; miscellaneous skull, 1; forelimb, 17; 

 pelvis, 8; hind limb, 12; and miscellaneous post-cranial, 1. Twenty-eight of the originally 

 selected and recorded characters were excluded from the analysis for various reasons (see 

 Character Analysis), leaving a functional total of 168 characters. The fact that a character 

 was autapomorphic (including those multistate characters with autapomorphic states) was 

 not considered sufficient grounds for its exclusion (see Yeates 1992). Although such 

 characters do not provide grouping information, their inclusion here reveals cases of 

 unusual and previously undocumented morphologies, or of when our observations do not 

 accord with those of the literature, calling the value of the particular character into doubt. 

 Specific descriptions of all individual characters, including those deleted from the analysis, 

 are found in the Character Analysis section. 



Data collation 



For the 27 taxa used in this study, a total of 286 specimens were examined (see Appendix 

 A). The data editor of MacClade 3.0 (Maddison & Maddison 1992) was used to input the 

 character states for each individual specimen and to generate a consensus set of character 

 states for each species. Polymorphic data (i.e., when a specimen simultaneously possessed 

 two or more states or, more commonly, was intermediate between two supposedly discrete 

 morphologies) were maintained. 



Although all variation is important and potentially informative, the large amount of 

 intraspecific variation, primarily among the phocids, required some manner of resolution. 

 Retention of every state indicated for a species by its representative specimens would 

 unnecessarily clutter the analysis (and thereby possibly decrease resolution) with what 

 amount to statistical outliers. Clearly, some states were more predominant than others 

 within a species, and it was these presumably more informative states that needed to be 

 retained. We accomplished this with a modified majority rule algorithm which would 

 hopefully maintain only the more predominant character state(s). For a given taxon and 

 a given character, the consensus state was ordinarily the most frequent state among all 

 specimens for that taxon. Note that polymorphic data, such as when a specimen possessed 

 both states 0 and 1, were treated as a discrete state (the state "01"), rather than independent 

 occurrences of the singular states. However, if the next most frequent state(s) possessed 

 the same frequency, or the same frequency minus one observation (i.e., highest frequency 

 -1), then the consensus state was a combination of these "equally" most frequent states 

 (i.e., the taxon was counted as being polymorphic for that character). 

 The only exception to the above formula occurred if one or more of the "equally" most 

 frequent states was polymorphic to begin with. In this case, the specimen polymorphisms 

 were "broken", the frequencies for each singular state were counted, and the above 



