Browne et al.: Improving pinniped diet analyses 



425 



clarki, O. kisutch, O. nerka, and O. mykiss less than 30 cm 

 in length and O. tshawytscha less than 35 cm in length 

 were considered seaward migrating juveniles (Groot and 

 Margolis, 1991). All distinguishable salmon otoliths were 

 identified to species and all identifications were verified 

 by W. Walker. 



Annual and seasonal variations in frequency of occur- 

 rence (FO) were examined with generalized linear models 

 (Venables and Ripley, 1994). We limited our analyses to 

 prey taxon with FO >5% during one or more seasons. Fre- 

 quency of occurrence on each sampling date was modeled as 

 a binomial random variable and for each prey taxon, we fit- 

 ted five models: constant, season (S), year (Y"), season-i-year 



tS-i-y), and season-i-year with interactions iSxY). To account 

 for overdispersion, we scaled Akaike's information criterion 

 (AIC) using 6=residual deviance/degrees of freedom from 

 the SxY model (Venables and Ripley, 1994). The model with 

 the smallest scaled AIC was considered the best descriptor 

 of seasonal and annual variation in FO. 



Results 



Over 1500 scats were collected from March 1995 through 

 October 1997. Sample sizes varied among years and 

 within season (Table 2). Frequency and number of indi- 



