174 



form of the phocid pelvis (de Muizon 1 982a ) is not sufficient to assure the monophyly of 

 the phoeids over the w hole of the post-cranial features (Fig.25F). However, the apparently 

 distinctive nature of the monachine post-cranial morphology as a whole isee Hendey & 

 Repenning 1972; Wyss 1988a) is indicated here, with the monachines. together with the 

 otarioids. being clearly separated from the remaining arctoid carnivores and the phocines. 

 This apparent conflict between the overall and regional solutions, and between the 

 individual regional solutions, can be attributed to one of two causes: 1 1 sampling error, 

 or 2) the presence of more than one signal within our data set. The potential effects of 

 sampling error have perhaps been underestimated in phylogenetic analysis. It is possible 

 that man\^ competing hypotheses of phylogenetic relationships stem from a biased selection 

 of characters from the universe of all possible characters. With the use of appropriately 

 designed tests, such competing hypotheses may be shown to be statistically equivalent. 

 Such may also be the case here. Although sampling error can never be eliminated, it can 

 be minimized by selecting a random set of characters (or. at least, a wide range of 

 characters) from the universe of all possible characters. The selection of a wide range of 

 characters is the more practical option at this point in time: however, it is susceptible to 

 the second, more serious error source, that of different signals w ithin the process partitions. 

 The possibility of such additional, non-phylogenetic processes being a potential source of 

 character covariation has been mentioned by Faith & Cranston (1992) and Hillis &. 

 Huelsenbeck (1992). An obvious example here is for the teeth, where the determining 

 signal is likely a "'functional" one. derived from the demands of food specialization within 

 the phoeids (see Chapskii 1955a: McLaren 19~5>. Within the more restricted and more 

 localized regional character sets, these various signals are apparently sufficient to swamp 

 the single (phylogenetic?) signal that predominates at the level of all characters. In all 

 cases, these "regional signals" are also quite strong, indicating a less homoplasious solution 

 (as indicated by the four goodness-of-fit statistics: see Fig. 25) than does the "overall 

 signal" (compare with Fig. 5). This is especialh" true of the basicranial. teeth, post-cranial, 

 and miscellaneous data sets. (The dramatic increase in CI for some regions is an artifact 

 of there not being an}' uninformative characters included in their partitions of the entire 

 data matrix. I 



Given such strong "regional signals", we can see how the "overall signal" might have 

 surfaced within the larger data set by examining a closely analogous situation. Wilkinson 

 (1991) suggested that increasing numbers of homoplastic characters (which might be 

 derived from the "regional signals" here) could be accommodated within a data set. so 

 long as they are random!}' distributed. Farns I 1969) made an even stronger statement in 

 that, given certain conditions (including the random distribution of homoplasy). the 

 number of homoplastic characters could outnumber the number of informative ones (and 

 by a considerable amount) without being detrimental. It would appear then that the various 

 "regional signals" within our data matrix are so localized as to become "insignificant" at 

 the level of the whole matrix. However, the conflicts between these "regional signals", 

 and between each and the "overall signal", are still sufficiently strong to make the overall 

 solution more homoplasious than an}" of the regional ones. The "overall signal" would 

 appear to be a more widespread, but slight!}" more dilute signal than the regional ones, 

 which are apparent!} very strong in certain restricted anatomical regions. 



