48 



1994). It is unknown what the extent of this is here, as these factors act in opposition to 

 one another, but it would be prudent to rely on other, more robust, indicators of resolving 

 power. 



The relatively high values of selected goodness-of-fit statistics (CI = 0.456, RI = 0.629, 

 and RC = 0.407) likewise point to a high resolving power. Benchmarks for evaluating 

 these statistics are rare, and, as these indices estimate the degree of homoplasy, they may 

 be specific for the group under examination (see Methods and Materials). However, in 

 the case of CI, the value obtained here is about on a par with the expected value for 27 

 taxa, 0.461 (Sanderson & Donoghue 1989). These relatively high values are somewhat 

 surprising, as the phocids as a group, and especially the phocines, have been characterized 

 as possessing a reasonably high number of reversals within a monophyletic pinniped 

 framework (Wyss 1988a; Berta & Wyss 1994). Regardless of whether this apparent 

 preponderance of reversals derives from a singular use of ACCTRAN optimization (see 

 above also), the fairly homoplastic nature of the phocids is reflected by the high value 

 obtained for the HI (0.770). However, this value is likely inflated to an unknown extent 

 due to PAUP's failure to designate polymorphic ancestral nodes (see Methods and 

 Materials). 



Thus far, we have only presented a preliminary assessment of the support for the overall 

 solution. This will be built upon by the results of specific, statistical tests designed to more 

 objectively quantify the level of support (both for the solution as a whole and for the 

 specific clades within it) and of various comparative tools, which are presented in the 

 following two sections respectively. Although the comparative tools are not tests of support 

 per se, their output very often will indicate the robustness of a solution [= how resistant 

 it is to further change; Maddison et al. (1984)], and can be used to corroborate the findings 

 of the true tests of support. 



STATISTICAL TESTS 



Interpreting statistical results 



While the influx of numerous statistical tests has been a great boon to the practice of 

 phylogenetic analysis, the results of these tests seem to be frequently misinterpreted. The 

 case for both the PTP test and skewness is clear and has been mentioned in the Methods 

 and Materials section: the degree of character covariation that these tests really indicate 

 is held to equate with the degree of phylogenetic signal in a given data set. Likewise, 

 analyses such as the bootstrap and Bremer support have been, or could be, taken to provide 

 some form of confidence interval on how well a data matrix estimates the one true 

 phylogeny. In reality, these tests merely indicate how well that data matrix presents its 

 own underlying distribution (= hierarchical pattern of relationships), which may or may 

 not coincide with the real distribution. The extension towards how well this underlying 

 distribution estimates the real phylogeny, again, requires additional assumptions. 

 Most cladists believe that the one true phylogeny is represented by a pattern of shared 

 derived characteristics in organisms and can be reconstructed by interpreting this pattern 

 through some criterion (e.g., parsimony, maximum likelihood). Thus, we attempt to gather 



