524 



Fishery Bulletin 102(3) 



occurring, robust, and diagnostic pollock structures 

 were removed from all scats containing pollock (see 

 Tollit et al., 2004, this issue). All were from the cra- 

 nium region (see Zeppelin et al., 2004, this issue) and 

 included the sagittal otolith (OTO), as well as the inter- 

 hyal (INTE), hypobranchial 3 (HYPO), pharyngobran- 

 chial 2 (PHAR), angular (ANGU), quadrate (QUAD), 

 and the dentary (DENT). Each individual fish element 

 was assigned one of three condition categories (good, 

 fair, or poor) and was measured three times (±0.01 mm) 

 at a specific location to calculate a mean estimate (see 

 Tollit et al., 2004, this issue). 



Fork lengths of pollock eaten by Steller sea lions in 

 Southeast Alaska were first estimated by applying al- 

 lometric regressions (Zeppelin et al., 2004, this issue) 

 to otolith lengths (OTOL) without correcting for par- 

 tial digestion (see Pitcher, 1981; Merrick and Calkins, 

 1996). We also measured and substituted otolith width 

 (OTOW) when otoliths were broken lengthwise. We then 

 applied appropriate DCFs and regression formulae to 

 otoliths assigned in good and fair condition (Tollit et 

 al., 2004, this issue). Finally, we applied allometric 

 regressions (Zeppelin et al., 2004, this issue) to all ele- 

 ments of the remaining six cranial structures (bones) 

 assigned to good or fair condition categories to provide 

 estimates of fish size across structures both with and 

 without applying the appropriate DCFs (Tollit et al., 

 2004, this issue). Structures in poor condition were 



excluded because of large intraspecific size variation 

 noted from feeding experiments with captive sea lions 

 (see also Sinclair et al.. 1994; Tollit et al., 1997; Tollit 

 et al., 2004, this issue). 



To incorporate the major sources of error in our 

 method, we calculated confidence intervals (95^r) for 

 fork-length estimates. First, we applied a random 

 bootstrapped regression equation, followed by a boot- 

 strapped correction factor applicable to each selected 

 structure (see Tollit et al., 2004, this issue). For the 

 five structures (INTE, HYPO. PHAR, ANGU, and OTO) 

 in good condition for which Tollit et al. (this issue) 

 recommended a DCF of 1.0 (no correction), we drew 

 bootstrapped values from a discrete declining triangu- 

 lar probability distribution (hj ranging from 1.0 to 1.05 

 (to simulate a limited degree of digestion). Finally, we 

 bootstrapped individual scats at random, by selecting 

 n scats with replacement from the original sample size 

 n (to account for resampling variability across scats) 

 and included only selected elements within those ran- 

 domly bootstrapped scats. Bootstrap randomizations for 

 these steps were done 1000 times and 959! confidence 

 intervals were taken as the 25 th and 975 ,h values of the 

 sorted bootstrapped values. 



Finally, consideration was also given to whether 

 an individual fish might be represented by different 

 structures within a single scat. We therefore compared 

 length estimates using all structures with those esti- 



