370 
Fishery Bulletin 109(4) 
study, we provide data on Steller sea lion diet from 
the southern extent of the EDPS range based on 1416 
scat (fecal samples) collected from five sites in Oregon 
and northern California from 1986 through 2007. We 
tested for seasonal, annual, and spatial differences in 
diet composition and discuss our results in relation to 
findings from Alaska. 
Materials and methods 
Field and laboratory 
We collected scat from four locations off Oregon and one 
location off northern California from 1986 through 2007 
(Fig. 1; Table 1). Three of the five locations were occu- 
pied seasonally as rookeries (Orford Reef, Rogue Reef, 
St. George Reef), whereas the other two were strictly 
nonbreeding haul-outs (Columbia River South Jetty, Cas- 
cade Head). Scats collected from May through August 
were classified samples from the “breeding season” and 
scats from the remainder of the year, as samples from 
the “nonbreeding season.” Scats were collected opportu- 
nistically as part of other research activities or during 
dedicated food habit collection trips. 
Scat samples were collected and processed accord- 
ing to the method described in Lance et al. 1 Collec- 
tions made after 2003 were processed with a standard 
washing machine according to collection and processing 
procedures described in Orr et al. (2003). Recovered 
hard parts were examined with a dissecting micro- 
scope and identified to the lowest possible taxonomic 
level by comparing all identifiable prey remains (e.g., 
bones, otoliths, cartilaginous parts, lenses, teeth, and 
cephalopod beaks) with a comparative reference collec- 
tion of fish from the northeastern Pacific Ocean and 
Oregon estuaries. Individual samples that contained 
both identified prey and remains too eroded to be iden- 
tified (unidentified fish) were included in this analysis, 
whereas samples with only unidentified remains (n=ll) 
or no remains (n=22) were not. 
Data analysis 
We summarized the relative importance of prey in sea 
lion diet by calculating the frequency of occurrence 
(FO) of each prey type. Frequency of occurrence was 
defined as the number of scat containing a given prey 
type divided by the number of scat with identifiable 
prey. Although other summary statistics are possible, 
FO is a simple calculation, widely used, and probably 
least affected by differences in prey recovery (Tollit et 
al., 2010). We calculated exact 95% confidence intervals 
1 Lance, M. M., A. J. Orr, S. D. Riemer, M. J. Weise, and J. 
L. Laake. 2001. Pinniped food habits and prey identifica- 
tion techniques protocol. AFSC (Alaska Fisheries Science 
Center) Proc. Rep. 2001-04, 36 p. Alaska Fisheries Science 
Center, NMFS, NOAA, 7600 Sand Point Way NE, Seattle, 
WA 98115. 
46° Nl— 
42°N- 
Pacific 
Ocean 
A 
Columbia R .if 
South Jetty A 
\ 
0 50 100 km 
i 
Orford Reef • 
Rogue Reef • 
St. George Reef •' 
128°W 
Bonneville 
Cascade Head A Dam 
124°W 
OR 
CA 
Figure 1 
Locations (solid circles=rookeries, solid triangles = 
haul-outs) where Steller sea lion ( Eumetopias jubatus) 
scat was collected off Oregon and northern Califor- 
nia, 1986-2007. See Table 2 for detailed information 
on sampling locations and effort. 
for FO by assuming that the number of scat in a collec- 
tion containing a given prey was binomially distributed. 
In addition to univariate summaries, we were also 
interested in testing whether multivariate diet compo- 
sition differed between collections. Wright (2010) and 
Lemons et al. (2010) noted that the common practice 
of using chi-square tests to compare diets violates the 
assumption of independence for that test by ignoring 
the nesting of multiple prey items within a scat. Vio- 
lation of the assumption of independence results in 
psuedoreplication and biased chi-square statistics. More 
appropriate alternatives for comparing multivariate diet 
composition between groups include distance-based per- 
mutation methods (e.g., Luo and Fox, 1996; Anderson, 
2001; Berry and Mielke, 2003); multiple-response cat- 
egorical variable methods (e.g., Agresti and Liu, 1999; 
Bilder and Loughlin, 2009; Nandram et al., 2009); and 
mark-recapture methods (Lemons et al., 2010). We chose 
the distance-based Mantel test (Mantel, 1967; Luo and 
Fox, 1996) because it could be formulated to address our 
questions of interest, was easy to implement in exist- 
ing software, and has been used by other researchers 
studying animal diets (e.g., Hudon and Lamarche, 1989; 
Green and Burton, 1993; Jones and Barmuta, 1998). 
We implemented Mantel tests, using package “vegan” 
(Oksanen et al., 2009) in R (R Development Core Team, 
2009). We tested whether diet composition differed by 
month (after controlling for year and site), year (after 
controlling for month and site), or site (after controlling 
for month and year). Distances among scat samples 
were computed using the Jaccard coefficient which is 
an asymmetrical binary coefficient commonly used to 
compare sampling units using species presence-absence 
