Goldman et al.: Feeding habits of Pagrus pagrus and Batistes capriscus 
319 
Figure 1 
Map of catch locations off the southeastern United States, where speci- 
mens of red porgy {Pagrus pagrus) and gray triggerfish {Batistes ca- 
priscus) were collected for analysis of gut content in 2009-2011. Gray 
lines represent bathymetry (in meters). 
Ontogenetic, temporal, and spatial changes in diet Prey 
were pooled on the basis of taxonomy (e.g., decapods 
and gastropods). Percent composition by weight was 
calculated for guts grouped by intervals of TL, season, 
depth (in meters), and latitude, and this metric was 
used for all analyses. For analytic purposes, prey types 
that contributed less than 1% by weight to the diet 
were excluded. 
Canonical correspondence analysis (CCA; ter Braak, 
1986), a multivariate direct gradient analysis tech- 
nique, was used to determine the degree of variability 
in the diets of red porgy and gray triggerfish, explained 
by the canonical axes. The canonical axes are linear 
combinations of the 4 explanatory variables correlated 
to weighted averages of the prey within the cells of 
the response matrix (ter Braak, 1986; Garrison and 
Link, 2000). The CCA was performed with the com- 
munity ecology package vegan, vers. 2.0-10 
(Oksanen et al., 2013), an extension to the 
statistical software R, vers. 3.1.2 (R Core 
Team, 2014). 
Each element in the response matrix 
was the mean percent weight of each prey 
taxon in a given length category, season, 
depth, and latitude combination. Prey data 
(%W) were log-transformed (ln[x-(-l]) to 
normalize the data. The explanatory vari- 
ables were coded as ordinal variables with 
the exception of season, which was coded 
as a categorical variable. The variance 
inflation factor was used to detect nearly 
collinear constraints (environmental vari- 
ables), although it must be noted that these 
constraints are not a problem with the al- 
gorithm that is used in the CCA function 
of the vegan package to fit a constrained 
ordination (Oksanen et al., 2013). Any use- 
less constraints would have been removed 
from the estimation, and no biplot scores or 
centroids would have been calculated (Ok- 
sanen et al., 2013). Permutation tests were 
used to determine the significant explana- 
tory variables (ter Braak, 1986). A biplot of 
prey species and explanatory factors was 
constructed to examine the correlations be- 
tween the explanatory variables (factors) 
and the canonical axes and to observe any 
dietary patterns associated with these fac- 
tors. A descriptive analysis was generated 
for each of the significant factors identified 
by the CCA. 
Hydrographic conditions were used to 
derive seasonal categories: spring: April 
through June; summer: July through Sep- 
tember, and autumn: October through 
December. Latitudes were grouped into 
3 categories: southern (27-29°N), middle 
(31-32°N), and northern (33-34°N). To 
examine the effect of fish length on the 
diets of red porgy and gray triggerfish, 
specimens were grouped into 50-mm-TL categories so 
that all members of a category displayed a reasonably 
consistent diet composition, and %W was calculated 
for each group. Groups with low sample sizes (fz<3) 
were trimmed to minimize outliers. Cluster analyses 
(Euclidean distance, average linkage method) were 
used to group these length classes into broader cat- 
egories that represented relationships among the diet 
compositions. 
Feeding strategies The feeding strategies of each spe- 
cies were analyzed according to the graphical method 
of Costello (1990), modified by Amundsen et al. (1996). 
Through the use of this method, prey-specific abun- 
dance was plotted against %F, making it possible to 
explore feeding strategies as well as shifts in niche use. 
Prey-specific abundance was defined as 
