Lowry et al.: Phylogeny of Oreosomatidae 



695 



ware (Swofford, 1991). Two phenetic methods of 

 analysis of genetic distance obtained from the allele 

 frequency data were examined with BIOSYS-1. First, 

 Nei's (1978) unbiased genetic distance measure be- 

 tween each species was calculated, and trees were 

 constructed by cluster analysis and the unweighted 

 pair-group method with arithmetic averaging 

 (UPGMA). Second, Rogers' (1972) distance measures 

 were calculated and trees constructed by the dis- 

 tance-Wagner procedure (Farris, 1972) with outgroup 

 rooting. The Wagner procedure, unlike the UPGMA 

 analysis, does not assume a constant rate of evolu- 

 tion. For the cladistic maximum-parsimony analy- 

 sis (PAUP), the loci were coded as characters and 

 the most common alleles as unordered character 

 states. When two common alleles were at equal fre- 

 quencies (0.5), they were treated as multiple states 

 and interpreted in the analysis as a polymorphism. 

 The "branch and bound" and "exhaustive" routines 

 were applied to search for the most parsimonious tree. 



Results 



Several measures of genetic variation were exam- 

 ined (Table 4). Four of the seven oreo species had 

 average sample sizes per locus exceeding 200, but 

 the average sample size for three species was less 

 than 30. Two of the genetic variation parameters, 

 mean number of alleles per locus and percentage of 

 variable loci, are clearly dependent on sample size: 

 both will increase as increasingly rare alleles are 

 detected. Percentage of polymorphic loci, especially 

 with use of the 0.95 rather than 0.99 criterion, will 

 be less sample-size dependent, and mean heterozy- 

 gosity per locus is little affected by rare alleles. In 

 fact, estimates of heterozygosity (and genetic dis- 

 tance) are influenced more by numbers of loci than 

 by numbers of individuals (Nei, 1978; Gorman and 

 Renzi, 1979), and all estimates of heterozygosity 

 given here are based on the same 26 loci. All oreos 

 showed high levels of variation, with average 



