502 



Fishery Bulletin 102(3) 



take into account only the loss of size to the relevant 

 feature being measured to estimate fish length (Figs. 

 1 and 2). 



The grading criteria for otoliths (OTO) were based 

 on the condition categories developed by Sinclair et 

 al. (1996) to investigate prey selection by northern fur 

 seals (Callorhinus ursinus). As seen in Sinclair et al. 

 (1996) and other studies (Frost and Lowry, 1986; Tollit 

 et al., 1997), external features such as lobation and the 

 general shape and definition of the sulcus were found in 

 our study to be good indicators of the degree of otolith 

 digestion. For the remaining cranial bones, digestion 

 indicators included the loss of definition or breakage 

 of defined structural features such as the horns and 

 ridge (QUAD), hammerhead and stock (DENT), swan 

 neck, notch and ridge (INTE), honeycomb and crown 

 iPHAR), cap, neck, and head (ANGU) and tube and 

 cone (HYPO). We used changes in the described condi- 

 tion-category criteria (see Table 1 for full details) in 

 tandem with photo-reference material (Figs. 1 and 2) to 

 classify all structures into one of three digestion grades 

 or condition categories: "good", "fair," or "poor." 



Hard parts recovered from feeding experiments were 

 sorted, and all selected cranial structures were as- 

 signed a condition category and measured with cali- 

 pers to within ±0.01 mm. Because otoliths were often 

 chipped or partly broken lengthwise, both length and 

 width were measured. To test our grading technique, an 

 independent observer (T.Z.) reassigned a random sub- 

 sample of each condition category of pollock structures 

 (n = 158) in a blind test. 



On initial investigation, high intraspecific variation 

 was observed within the selected structures assigned in 

 poor condition in our feeding study with captive Steller 

 sea lions. Consequently, structures in poor condition 

 were not used to calculate DCFs for this category. Our 

 basis for exclusion was supported by the work of Sin- 

 clair et al. (1994) and Tollit et al. (1997). Captive sea 

 lions in our study occasionally regurgitated prey in the 

 swim tank. Recovered structures that we considered to 

 have been regurgitated were excluded from DCF calcu- 

 lations (i.e., vertebrae still articulated, bones that had 

 flesh attached or that were of a size to exclude passage 

 through the pyloric sphincter). 



Mean reduction (MR) in the metric of each structure 

 (s) recovered from our feeding experiment was estimated 

 for each remaining condition category (c) according to 



MR 



T 



xlOO, 



where the mean size of egested structures (E) of each 

 condition category was calculated from measurements 

 of those recovered from the captive feeding experiments. 

 and the mean size of each ingested structure (/.) was 

 estimated from the fork length of fish fed by using 

 inverse predictions of the regression formulae derived 

 from fresh material (Zeppelin et al., 2004, this issue). 

 Mean ingested size was estimated by using bootstrap 



simulations (1000 runs) that randomly sampled with 

 replacement and selected the median (500 th value I from 

 the sorted bootstrapped values (Reynolds and Aebischer, 

 1991). 



For pollock, mean reduction for each condition cat- 

 egory was compared across size ranges by using a Krus- 

 kal-Wallis analysis of variance. A significance level of 

 P<0.0056 was set based on the Bonferroni adjusted 

 probability for nine multiple comparisons (Siegel and 

 Castellan, 1988). Failing to find any significant differ- 

 ences resulted in pooling the data from each size range 

 to calculate specific condition category MR values. Con- 

 dition category DCFs were calculated for each selected 

 structure as / JE SC except for PHAR structures of Atka 

 mackerel because too few elements were recovered from 

 the scats of captive animals. 



Estimating confidence limits around digestion 

 correction factors 



We used a bootstrap simulation to estimate upper and 

 lower bounds of the 95% confidence interval (CI) given 

 that the DCF is a ratio of two means (Reynolds and 

 Aebischer. 1991 1. This technique allows different sources 

 of error to be combined or partitioned. There were two 

 major sources of error associated with calculating DCFs 

 (Tollit et al., 1997). The first were those associated with 

 the regression formulae used to calculate the mean size 

 of structure ingested from the original fish fed. and the 

 second were those associated with the errors around the 

 mean size of egested structure (i.e., resampling errors). 



We assessed errors associated with the regression 

 formulae using a parametric bootstrapping procedure 

 (Manly, 1997) that involved regressing structure size 

 against fork length. This was repeated 1000 times and 

 95% confidence intervals were taken as the 25 th and 

 975 th values of the sorted bootstrapped regression coef- 

 ficient values. Results were compared to those computed 

 analytically by using the resultant standard error (Eq. 

 17.23 in Zar, 1984) and were found to be consistent (see 

 Zeppelin et al., 2004, this issue). 



We estimated resampling errors related to the vari- 

 ability in digestion of egested structures by repeatedly 

 selecting n structures, at random, with replacement from 

 the original sample set of n egested structures. Mean 

 egested size was recalculated in this way 1000 times, 

 as were a mean DCF and 95% CI as described above. 

 Both regression and resampling errors were combined in 

 sequence to derive overall 95% CIs around DCFs. 



Our recommended procedure for applying our DCFs to 

 cranial structures recovered from scats collected in the 

 wild has four steps: 1) measure the recovered structures 

 and grade the extent of digestion using defined criteria 

 and photo-reference collection; 2) exclude structures 

 graded in poor condition; 3) multiply measurements of 

 structures in good and fair condition by their appropriate 

 digestion correction factors to derive their original size; 

 and 4) calculate the size of prey from allometric regres- 

 sions relating corrected structure measurements to fish 

 fork lengths (see also Tollit et al.. 2004, this issue). 



