Lee and Sampson: Dietary variations for three rockfish species off the Pacific Northwest 
517 
Axis 1 (43.5%) 
Figure 5 
The principle component analysis (PCA) plot for diet composi- 
tions of the seasonal quarterly collections of three rockfish 
stomach species samples: Sebastes flavidus (crossed hatch), S. 
entomelas (open circle), and S. pinniger (closed triangle). Each 
point represents an individual stomach sample. In parentheses 
on the axis labels are the percentages of the total variance 
in the data explained by each axis. The prey species groups 
(large closed circles) are overlaid in the same ordination space 
according to a weighted average of the individual PCA scores: 
Deca = decapods, Eupha = euphausiids, Hetero = heteropods, 
Jelly =jellyfishes, Misc=miscellaneous. 
the species and changed differently over the 
seasons. 
The factor “predator type” accounted for most 
of the variability in the PCA axis 1 scores in 
terms of mean squares (MS = 10.152, df=2), and 
season accounted for most of the variability of 
the axis 2 scores (MS = 1.643, df=2). In other 
words, PCA axis 1 mostly measured differences 
between the fish species and PCA axis 2 mostly 
measured quarterly changes in their food habits. 
In building the models for the food habits of the 
three rockfish species from the quarterly fishery 
collections, we found that the other variables 
considered (depth, time of day, sex, and fish size) 
were not significant (P>0.05). The two sets of 
PCA scores predicted by the GLMs summarize 
the diet variation of species over the six quarters 
(Fig. 6) and show that the diet of S. pinniger was 
stable over the quarters, whereas the diets of S. 
flavidus and S. entomelas were not stable and 
did not vary in parallel. 
Separate groupings of the 1980 and 1998 
NMFS summer survey samples were fairly evi- 
dent in the PCA ordination space (Fig. 7), which 
had a triangular appearance similar to that of 
the PCA plot from the quarterly fishery collec- 
tions. Again, the proportions of euphausiids, fish- 
es, and jellyfishes were the primary prey items 
determining the shape of the PCA plot; the fish 
stomachs at the points of the triangle had only 
one prey-item species of euphausiids, fishes, or 
jellyfishes. However, there were no 1980 summer 
survey samples in the upper portion of the scat- 
ter plot, which corresponded to samples where 
jellyfishes and heteropods were major prey spe- 
cies in the 1998 samples. This finding, based on 
individual stomach sample data, corresponds 
with the result based on the weight proportions 
calculated from aggregated samples (Fig. 4), in which 
a dominance of these prey groups in the diets of S. 
flavidus was found in 1998 but not in 1980. 
According to the general linear model for the PCA 
scores, the diet composition of S. flavidus from the sum- 
mer survey collections was related with extrinsic factors 
in a more complicated fashion than was apparent in 
the quarterly fishery samples (F u 425 =18.28, P<0.0001, 
coefficient of determination (r 2 ) = 0.376 for axis 1; P 21 
418 =30.43, PcO.OOOl, r 2 = 0.395 for axis 2). Year, lati- 
tude, and depth were significant main effects for the 
first PCA component scores. The significant interaction 
terms were year by latitude and depth by latitude. The 
most influential variable was the interaction between 
year and latitude (MS = 3.714, df=2). The model for the 
axis-2 scores included terms for time of day and the 
interaction between time of day and latitude, along with 
those factors that were significant in the model for axis 
1. The interaction between time of day and latitude 
was the most influential term in the model (MS =1.646, 
df=5). In general, PCA axis 1 was mainly associated 
with differences in the food habits between the years 
and PCA axis 2 with latitudinal effects. The predict- 
ed values of the PCA-1 scores for each year and over 
latitude illustrated that the diets of S. flavidus were 
different in 1998 and 1980, and had different trends 
with latitude (Fig. 8). Similarly, the predicted PCA-2 
scores for time of day and over the latitude showed 
that the diet pattern differed with the time of feeding 
and the geographic location of feeding. The complicated 
models, with significant interactions between temporal 
and geographical factors, indicate that the S. flavidus 
food habits are not determined by single factors alone. 
Interestingly, as in the case of the quarterly fishery col- 
lections, the fish characteristics (sex and fish size) were 
not significant explanatory variables for the food habits 
of S. flavidus from the summer survey collection. 
Discussion 
The three rockfish species examined in this study, 
which covered a period of unusual oceanographic events 
(El Nino and La Nina), mostly preyed upon pelagic 
