Fitzgerald et al.: Elemental signatures in otoliths of larval Theragra cha/cogramma 



607 



merits for the MANOVA were also met by the data. 

 Among-location differences in the elemental composition 

 of larval pollock in specific regions of the otoliths were 

 compared by using one-factor multivariate analysis of 

 variance (MANOVA) and one-factor analysis of variance 

 (ANOVA). We treated location as a fixed factor in both 

 MANOVA and ANOVA tests. Because of difficulties col- 

 lecting pollock larvae, we were unable to achieve equal 

 replication of sites within locations. We therefore pooled 

 samples from collections within a location by randomly 

 selecting fish from each location for subsequent analysis. 

 However, the lack of replication at the within-location 

 level necessarily restricted our ability to draw general 

 conclusions concerning spatial variability in otolith 

 composition beyond the samples analyzed in the pres- 

 ent study. All a posteriori comparisons among locations 

 were performed by using Tukey's honestly significant 

 difference (HSD) test (experimentwise error rate = 0.05). 

 Multivariate differences in elemental signatures from 

 the MANOVA were visualized by using canonical dis- 

 criminant analyses (CDA). All analyses were conducting 

 by using the SAS statistical program (SAS, version 6, 

 1990, SAS Inst. Inc., Cary, NO. 



Comparisons of elemental profiles across otoliths were 

 made with repeated measures ANOVA. We tested the 



following null hypotheses: 1) there was no variation in 

 trace element profiles across individual otoliths (i.e., 

 from the nucleus to the edge), 2) there were no differ- 

 ences in mean element concentrations among locations, 

 determined by averaging data across individual otolith 

 profiles, and 3) there were no differences in the pat- 

 tern of element profiles across otoliths among locations. 

 Otolith profiles with missing values were removed, and 

 therefore we were able to use MANOVA for the repeated 

 measures analysis. The multivariate approach to re- 

 peated measures is generally more conservative than 

 univariate repeated measures analysis. However, the 

 multivariate test does not assume sphericity of orthogo- 

 nal components, requiring only that the data conform to 

 multivariate normality with a common covariance ma- 

 trix for individual larvae at each location (Littell et al., 

 1991). The approach still requires that adjacent points 

 on the trajectories be equidistant. Therefore samples 

 from EPMA were assigned to a distance category at 

 intervals of 15 ftm (0 //m, 15 fim, 30 /.im , 45 /jm, 60 f/m, 

 75 /im, and 90 j/m) across the otolith, to a distance of 

 90 ^m from the nucleus. Samples were averaged when 

 more than one measurement was available within a 

 distance category. Laser ablation ICP-MS samples were 

 assigned to a distance category at intervals of approxi- 



