512 
Fishery Bulletin 107(4) 
of stomach eversion, regurgitation, or feeding in the net. 
Fish were discarded if they showed signs of any of these 
actions. Each valid stomach was wrapped in a cloth bag 
and tied with a tag indicating the species type, sex, fork 
length (cm), date of sampling, and sampling location, 
and then preserved in a 10% buffered formalin solution. 
After arrival at the laboratory, the stomach samples 
were rinsed with water, and transferred and stored 
in 70% ethanol for later examination. For prey iden- 
tification and associated measurements, each stomach 
was cut open and the contents were blotted dry with 
126°W 124°W 122°W 
stomachs: Sebastes flavidus, S. entomelas, and S. pin- 
niger. Each symbol represents different sampling collec- 
tions: solid circles for the seasonal quarterly collections 
(April 1998 to September 1999); open circles for the 
1998 NMFS survey collections; stars for the 1980 NMFS 
survey collections. 
absorbent paper. The prey items were examined under 
a dissecting microscope and identified to the lowest pos- 
sible taxonomic level. The wet weight of each individual 
category of prey item was measured to the nearest 0.01 
gram and recorded. 
Data analysis 
To quantify and summarize the diets within each of 
the two data sets, the quarterly fishery data and the 
NMFS summer survey data, we calculated the percent 
frequency of occurrence and the percent by weight of 
each prey item for each fish species over each season in 
a given year. The frequency of occurrence was calculated 
by dividing the number of stomachs containing a par- 
ticular prey item by the number of nonempty stomachs in 
a given season. The percent by weight was calculated by 
dividing the total weight of a particular prey item by the 
total weight of the stomach contents in a given season. 
Samples in the quarterly fishery data set were grouped 
into four different seasons: winter (December-Febru- 
ary); spring (March-May), summer (June-August), and 
fall (September-November). 
Many factors potentially influence the diet patterns 
of fish, such as biological factors (species, sex, and 
length), spatial factors (latitude and depth), and tem- 
poral factors (daily, quarterly, and annually). Many 
methods of quantitative analysis, ranging from uni- 
variate to multivariate statistical techniques, have 
been proposed for examining the variation of diets 
in relation to extrinsic factors (Hyslop, 1980; Cortes, 
1997). However, no single method has been accepted 
as being the best to represent variability in fish food 
habits. For food-habit studies individual diet informa- 
tion is often aggregated at a population level or across 
certain extrinsic factors for a statistical comparison of 
diets. In most studies there is no attempt to formally 
evaluate diet variability, except at the level of aggre- 
gate mean values. Such data aggregation techniques 
attempt to overcome the multivariate nature of diet 
data and the problems associated with the high vari- 
ability and unequal weighting of contents of individual 
stomach samples. However, data aggregation across a 
particular factor (e.g., combining samples by latitude 
class) results in the loss of important diet information 
at the individual level. It also may result in erroneous 
conclusions as to the significance of a factor because 
an analysis based on aggregated data cannot simul- 
taneously account for other factors or assess possible 
interactions with other factors. 
For this study, we employed a multivariate method, 
principle component analysis (PCA), to examine the 
patterns of individual diet compositions in relation to 
sets of extrinsic factors. With the PCA, the measures 
of diet composition are extracted in a low-dimensional 
ordination space, based on the individual stomach diet 
information. We can then relate extrinsic factors to the 
ordination scores of stomach samples extracted from 
the PCA to test whether or not the factors are related 
to dietary variation. We can also gauge the relative 
