22 



Although some authors indicate that both missing data and inapplicable characters (e.g., 

 feather size for mammalian taxa) be coded as "missing" (represented by a question mark) 

 (e.g., Swofford 1993), a distinction was made here between these two cases. Inapplicable 

 characters were instead assigned to a discrete state (state 9), as advocated by Maddison 

 (1993). Largely, this ties in with how PAUP (and other computer algorithms) treat missing 

 data. PAUP will initially treat the missing datum as if it were almost entirely absent from 

 the tree (at least with respect to that character), and then later attempt to infer an 

 appropriate state for any missing data based on parsimony (Maddison 1993). While this 

 latter step is valuable when the state is unknown due to ignorance (creating a valuable 

 hypothesis to be tested in the future), it is clearly inappropriate for inapplicable characters 

 in that PAUP may infer a state that clearly does not apply to the taxon in question (e.g., 

 "large feathers" in mammals when it should really be "feathers absent") (Platnick et al. 

 1991). 



Assumptions concerning taxa 



In dealing with the large number of polymorphic taxa, PAUP's multistate taxa option was 

 set at "polymorphism", forcing PAUP to account for all but one of a polymorphic taxon's 

 states in the most parsimonious way possible by invoking changes within this terminal 

 taxon (Swofford 1993). Although the underlying assumption of this setting is that the 

 multistate taxon is a heterogeneous group (i.e., a higher level cluster of morphologically 

 variable taxa), this setting comes the closest to treating the indicated polymorphisms as 

 real and important. The alternative setting, "uncertainty", selects only the most parsimo- 

 nious state out of the set provided, ignoring the remaining states, and thus the 

 polymorphism, altogether. However, one limitation of "polymorphism" is that PAUP will 

 not form a polymorphic ancestral taxon, even if all of its descendants are identically 

 polymorphic (Swofford 1993). Although this results in the loss of much potential grouping 

 information, it should be noted that the other major phylogeny inference packages (i.e., 

 Hennig86 vl.5 and PHYLIP v3.5) will not handle polymorphic data at all (Sanderson 

 1990). 



The taxa Canis lupus, Enhydra lutris, Lutra canadensis, Martes americana, Odobenus 

 rosmarus, Procyon lotor, Ursus americanus, and Zalophus calif ornianus (hereafter referred 

 to solely by their generic appellations, as are the monotypic phocid genera) were assigned 

 as outgroup taxa. In so doing, we assumed that each taxon is a representative member of 

 a higher level taxon: canids, lutrines, lutrines, mustelids minus lutrines, odobenids, 

 procyonids, ursids, and otariids respectively. This is almost certainly not the case, but we 

 deemed the alternative, using the presumed ancestral state for each higher level taxon, as 

 less desirable. Such an assessment requires at least some tacit assumptions about both the 

 internal phylogeny and ancestral affinities of the higher taxon. As well, the use of ancestral 

 states may conceal the presence of some potentially important derived subgroups with 

 which the true affinities of the ingroup may lie. In any case, trees were rooted such that 

 the collective outgroup used here was forced to be paraphyletic with respect to the phocids 

 (which were forced to be monophyletic) in accordance with the current views on caniform 

 phylogeny (see Tedford 1976; Flynn et al. 1988; Wyss & Flynn 1993; Vrana et al. 1994). 



