Perrin et al.: Geographic variation in cranial morphology of Stenella attenuata 



329 



on standardized data. These environmental compo- 

 nents served as composite environmental variables 

 for comparisons with morphological variables. 



Matrix correlations and Mantel tests were used 

 to test for local and regional patterning of environ- 

 mental variables. Also, differences between each pair 

 of blocks for a given morphological variable were com- 

 pared with those for an environmental variable. 



Interspecific comparisons 



The predominant trends in the data sets for each of 

 S. attenuata and S. longirostris were summarized 

 with principal components and canonical variables. 

 Information is available for 16 blocks from which 

 both offshore S. attenuata and S. longirostris were 

 sampled (Fig. 1). These blocks are representative of 

 the total geographic range investigated in our stud- 

 ies. In order to compare general patterns of varia- 

 tion in the two species, we calculated product-mo- 

 ment correlations, Mantel tests, and matrix corre- 

 lations for individual morphological characters, prin- 

 cipal-component projections, and canonical-variable 

 projections of these 16 blocks. 



In our analyses, average distances based on mor- 

 phological characters were computed between each 

 pair of localities. To evaluate the extent of similar- 

 ity in geographic patterns, the original distance 

 matrices for each species were modified such that 

 only distances among the 16 localities common to 

 both species were included. These matrices were 

 then compared by using the Mantel test and com- 

 puting the matrix correlation. 



Results 



Sexual dimorphism 



Table 1 includes mean measurements for males and 

 females based on 11 blocks. For two-way ANOVAs 

 assessing geographic block and sex (df=21, rc=461), 

 all but 5 of the 30 were very highly significant 

 (P<0.001). The probability was 0.02 for number of 

 teeth (lower right) and 0.007 for orbital length. Two 

 characters showed no significant variation (upper 

 tooth counts) and one character (number of teeth 

 [lower left]) was close to significant (P<0.06). Sta- 

 tistically significant interactions (P<0.05) between 

 block and sex were found for five of the characters: 

 condylobasal length; width of rostrum (at 1/4 

 length); width of rostrum (at 1/2 length); width of 

 rostrum (at 3/4 length); and width of internal nares. 

 Interaction denotes that the degree of sexual dimor- 

 phism differs among blocks for these characters. 



Sexual dimorphism was significant for 22 of the 

 30 characters (Table 1). Females had longer ros- 



trums, which is reflected in a number of characters 

 (i.e. 2, 3, 22, 27, 30). In general, males had wider 

 skulls and tended to be larger for nonrostral por- 

 tions of the skull. Percentage differences between 

 sexes are presented in Table 1. The average abso- 

 lute difference (i.e. sign ignored) between the sexes 

 for the 30 characters was 1.78%. For 8 characters 

 where females were larger, the average difference 

 was 0.75%, whereas for 22 characters where males 

 were larger the average difference was 2.16%. The 

 greatest differences were found for width of premax- 

 illary (at 1/2 length) and width of rostrum (at 3/4 

 length) — 5.25% and 6.66%, respectively. 



Table 3 shows the results for the nested ANOVA 

 for cruise sets within blocks. Twenty-three of the 30 

 characters showed highly significant or very highly 

 significant block effects, whereas only three charac- 

 ters (those involving the temporal fossa and length 

 of braincase) reflected highly significant or very 

 highly significant effects for cruise set. Even in those 

 three cases, block effects were more pronounced. 

 Therefore, we conclude that combining cruise sets 

 into blocks did not have an important confounding in- 

 fluence on geographic patterns found among blocks. 



Correlation, ordination, and clustering 



Most character pairs had positive correlations. An 

 exception was tooth counts and temporal fossa mea- 

 surements, which tended to have negative correla- 

 tions with skull width measurements. The dendro- 

 gram in Figure 2 summarizes absolute correlations 

 (i.e. sign of correlation ignored) among characters 

 based on 29 blocks to provide an assessment of char- 

 acter covariation. The width of external nares was 

 the character with the least association with other 

 measures. Tooth characters join and are separated 

 from the remaining morphometric characters. Brain- 

 case measures and skull width (at parietals) clus- 

 ter in another relatively distinct group. The remain- 

 ing characters are arranged in two groups. The clus- 

 ter at the top of Figure 2 includes most length mea- 

 surements and height of ramus. Width measure- 

 ments along with length of antorbital process, or- 

 bital length, and length of temporal fossa are in- 

 cluded in the adjoining major cluster (Fig. 2). 



Table 4 includes character loadings on the first 

 two principal components based on data for 29 

 blocks. Component I explains 45.0% of the total 

 variance for the 30 characters, whereas component 

 II summarizes an additional 16.8% (cumulative to- 

 tal of 61.8%). Projections of blocks onto the two com- 

 ponents are depicted in Figure 3, and a map (Fig. 

 4) is included that renders geographic block projec- 

 tions onto the first component. Component I repre- 



