210 



Fishery Bulletin 103(1) 



Table 1 



Number of age-0 walleye pollock (Theragra chalcogramma) collected near Kodiak Island, Alaska, September 1993, measured 

 for standard length, and examined in the laboratory to estimate condition, growth, and the weight and taxonomic composition of 

 stomach contents. Sample is the number of trawl hauls. 



; Whole wet weights from thawed fish. 



- Somatic wet weights from fish preserved in 95 f 4 ethanol after freezing at sea. 



3 Collective width of daily otolith increments 1-5; numbering begins with the most peripheral increment. 



4 Collective width of daily otolith increments 6-10. 



pylorus. Gut contents were dissected from the speci- 

 mens and weighed to the nearest 0.001 gram. Somatic 

 weight represented whole wet weight minus the gut 

 content weight. Three fish were omitted from further 

 consideration because of apparent regurgitation. Taxo- 

 nomic composition of age-0 diets was determined by 

 counting the organisms in the gut after sorting them 

 into broad taxonomic groups. 



Zooplankton was collected by using a 1-m Tucker net 

 (333-/mi mesh) to sample where age-0 pollock had been 

 collected. The net was fished through acoustic echo lay- 

 ers believed to be age-0 pollock in order to characterize 

 their immediate prey field. Potential prey items were 

 sorted into broad taxonomic groups and enumerated at 

 the Polish Plankton and Identification Center, Szezcin, 

 Poland. 



Temperature and salinity profiles (near surface to 10 

 m off bottom) were obtained by using a Seabird SBE- 

 911+ CTD system. Profile data were collected during 

 deployment at a descent rate of ca. 0.5 m/s. 



Statistically significant differences in age-0 condition, 

 growth, and feeding intensity among geographic areas 

 were detected with split-plot analysis of covariance (AN- 

 COVA) and post hoc multiple comparison tests (Proc 

 Mixed, SAS software, Littell et al., 1996). The covari- 

 ates were fish length or age (days since hatching). Fol- 

 lowing Milliken and Johnson (2002), we first tested for 

 covariate significance (H : all slopes = 0) and homogene- 

 ity of slopes (H () : equal slopes) to ensure appropriateness 

 of the following reduced, common-slope model: 



Y = a + (5x :j + Area ( + Sample, I Area 1 1 + e ljk , 



where Y = dependent variable; 



a = intercept parameter; 



P = slope parameter; 



x = covariate for sample i and area,/'; and 



e k = replicate error for sample /', area./', and fish k. 



A split-plot design was necessary to account for the 

 nesting of samples (trawl catches) within area, and 

 individuals within sample. To avoid pseudoreplication, 

 trawl catch was the sampling unit instead of individual 

 fish. Area was a fixed effect; sample was a random 

 effect. For body condition, lengths and weights were 

 log ( ,-transformed according to the method of Patterson 

 (1992); two points were omitted because of suspiciously 

 low length-specific, whole-body weight. For feeding in- 

 tensity, gut content weights (GCW) were fourth-root 

 transformed (GCW 025 ) to linearize the GCW-length 

 relationship and remove heteroscedasticity (Clarke and 

 Warwick. 2001). Significance of post hoc pairwise dif- 

 ferences was based on a Bonferroni-corrected, 0.05-level 

 of significance. The standardized catch data were not 

 incorporated into these tests; therefore the conclusions 

 pertain to the samples not weighted by catch. 



Nonmetric multidimensional scaling (NMS, PC-Ord, 

 McCune and Mefford, 1999) was used to ordinate the 

 diet and plankton samples according to taxonomic 

 composition. Each diet sample represented the aver- 

 age numerical composition of the diet of all fish in the 

 sample. This value was calculated by dividing the sum 

 of all items within each taxonomic category by the num- 

 ber of fish in the sample. The ordinations, one for diet 

 and another for plankton, were based on Bray-Curtis 

 similarity coefficients of fourth root-transformed data. 

 Differences among the four areas were statistically 

 tested by using a two-way nested analysis of similarity 



