332 



Fishery Bulletin 92(2). 1994 



compassing part of Hawaiian Islands) at the right 

 of the diagram. A number of single blocks (0312, 

 0314, 0311) are separated from the others at upval- 

 ues of 3.10 and 2.87. Most of the blocks from the north- 

 eastern portion of the range were included in a single 

 cluster at lower upvalues, although some individual 

 blocks were separated by themselves (e.g. 0614, 0711). 

 In Figure 6B, a skyline diagram is included that 

 is based on five characters: length of rostrum (from 

 pterygoid); preorbital width; length of braincase; 

 length of temporal fossa; and width of temporal 

 fossa. These variables were identified by using ca- 

 nonical-variates analysis (reported below) as those 

 that in combination provided the greatest discrimi- 

 nation among blocks. The use of function-point clus- 

 tering on these selected characters provided clusters 

 that persisted through several clustering levels. At 

 the upvalue of 1.38, four clusters were formed. The 

 Hawaiian Island block (0802) is in a group by itself 

 (on left side of Fig. 6B). The blocks to the south, 

 southwest, and west are in a second assemblage (i.e. 

 0214 through 0506), while the northeastern blocks 

 form the other major cluster (i.e. 0513 through 

 0713). Block 0711 is separated into its own group. 

 At £<v-values less than 1.38, the second cluster is sub- 

 divided and the southern blocks are included with 

 some from the southwest. 



Canonical-variates analysis se- 

 lected five characters (Table 5). 

 Blocks are projected onto the first 

 two canonical variables resulting 

 from this analysis (Fig. 7), whereas 

 Figure 8B shows geographically the 

 distribution of geographic variation 

 projections onto canonical variable 1. 

 The eigenvalue for canonical vari- 

 able 1 is 1.00, while that for variable 

 2 is 0.19. The two together summa- 

 rize 82.4% of the variance in the five 

 characters. In the scatter diagram 

 (Fig. 7), blocks with only single 

 specimens are projected onto the 

 canonical variables generated by 

 using the other 29 blocks. As indi- 

 cated in Table 5, the most important 

 character in the canonical-variates 

 analysis was postorbital width. (Fig. 

 8A). It has relatively small values in 

 the northeastern portion of the 

 range, and considerably larger val- 

 ues in other areas to the south, 

 southwest, and west. The largest 

 postorbital width measurements are 

 found in specimens taken near the 

 Hawaiian Islands (0802). The geo- 

 graphic configuration of this character is very simi- 

 lar to the general pattern exhibited by the first ca- 

 nonical variable (Fig. 8B). Canonical variable 1 and 

 principal component I (Fig. 4) have similar geo- 

 graphic patterns. Canonical variable 2 does not re- 

 flect any particular geographic pattern. It basically 

 contrasts three blocks (0711, 0802, and 0311) with 

 the others, as indicated by these three being sepa- 

 rated in the upper portion of Figure 7 from other 

 blocks. 



Mantel test for geographic patterning 



Results of Mantel tests and matrix correlations com- 

 paring interblock character differences with geo- 

 graphic distances (in nautical miles) and with the 

 reciprocals of geographic distances are presented in 

 Table 6. Regional patterning, as indicated by a sig- 

 nificant association with geographic distance, was 

 found for 18 (60.0%) of the 30 characters. In these 

 characters, the localities the farthest apart tended 

 to exhibit the greatest morphological difference. 

 Local patterning, which is judged based on a signifi- 

 cant negative association between geographic dis- 

 tance-reciprocals and morphological differences, was 

 found in 22 (73.3%) of the characters. Not unexpect- 

 edly, virtually all characters (i.e. all but one) exhib- 



