Willis et al.: Feeding behavior of 3 sciaenids along the southeastern United States 
293 
were fourth-root transformed, and cluster analysis and 
nonmetric multidimensional scaling were performed 
on the resulting Bray-Curtis dissimilarity matrix (Mc- 
Cune and Grace, 2002; Bozzetti and Schulz, 2004) with 
the packages labdsv and vegan in R software, vers. 
2.15.2 (R Core Team, 2012). Predator species, region, 
year, and season were examined as categorical vari- 
ables and water depth (in meters), bottom temperature 
(in degrees Celsius), and salinity were examined as 
continuous variables by analysis of similarity. 
Three metrics of stomach content by prey category 
were calculated, as percentages, for each predator spe- 
cies, according to Hyslop (1980), where the total sample 
was defined as all prey within a given prey category: 
frequency of occurrence, composition by weight, and 
composition by number. An index of relative impor- 
tance (IRI) was determined by combining these met- 
rics to eliminate any biases created when each method 
is analyzed individually (Goldman and Sedberry, 2010; 
modified from Pinkas et ah, 1971): 
IRI = (N + W)xF, 
( 1 ) 
where F = frequency of occurrence; 
N = composition by number; and 
W - composition by weight. 
IRI has been included only in tabular form as a meth- 
od of data comparison across diet studies because it 
is calculated by combining indices that may produce 
varying results and, therefore, is not necessarily the 
most robust representation of diet (Cortes, 1997; Hans- 
son, 1998). To provide a more complete picture of prey 
consumption by each predator, IRI is presented by year. 
In additional, mean percent weight and mean per- 
cent number were calculated for each predator accord- 
ing to Chipps and Garvey (2007). Prey were analyzed 
at the lowest possible taxon, and bootstrapping was 
used to generate bias-corrected 95% confidence inter- 
vals for both means with the package boot in R, (vers. 
2.15.2). Prey were then grouped into higher prey cat- 
egories for the purpose of graphical presentation. 
Feeding strategy was determined with the method 
of Amundsen et al. (1996), modified from the method 
of Costello (1990), where prey-specific abundance (Pj) 
is plotted against frequency of occurrence. Expressed 
as a percentage, prey-specific abundance is a given 
prey taxon’s proportion in relation to all prey items ob- 
served in only those predator stomachs that contained 
the given prey taxon: 
Pi = 
£S t 
x 100, 
( 2 ) 
the feeding strategy of the predator population. Data 
points that cluster near the top of the y-axis indicate 
specialized feeding by individuals within the popula- 
tion. A cluster close to the origin describes infrequent 
consumption of a prey type (i.e., prey types that are 
not an important part of the predator’s diet). Data 
points that are scattered across the graph indicate 
that a population cannot be characterized as one that 
employs a single feeding strategy; a population may be 
specialized sometimes and generalized at other times. 
Data points clustered in the upper right quadrant of 
the graph indicate a population with a specialized feed- 
ing strategy, where a high percentage of the population 
consumes one or more specific prey types. 
Potential for trophic overlap among each predator 
pair was tested on mean percent weight and mean per- 
cent number with the simplified Morisita-Horn (M-H) 
index (O): 
0 = (2E"P?jP ik )/(E ; ''p^ +p^), (3) 
where n = the total number of prey item groups; 
Pij = the proportion of the prey item i used by 
predatory; and 
Pik = the proportion of the prey item i used by 
predator k (Wolda, 1981; King and Beamish, 
2000 ). 
Both mean percent weight and mean percent number 
were examined to provide the most robust evidence of 
potential overlap because both variables are biased 
when viewed independently owing to the broad range of 
prey sizes (Graham et al., 2007). The M-H index ranges 
from 0 to 1, with diet overlap increasing as the index 
approaches 1 (Zaret and Rand, 1971; Labropoulou and 
Eleftheriou, 1997; King and Beamish, 2000; Graham et 
al., 2007; Rodrigues and Vieira, 2010). 
Results 
Analysis of similarity revealed that differences among 
predator species were significant (goodness of fit: coef- 
ficient of determination [r 2 ]=0.112, P=0.001); however, 
for a given predator, the differences among years, re- 
gions, and seasons were not statistically significant and 
depth, temperature, and salinity were not correlated 
with diet composition. Therefore, for the remainder of 
the analyses, we combined seasons and regions to in- 
vestigate differences in diet composition among preda- 
tors and combined years for all analyses except IRI._ 
where Si = sum of prey i; and 
S t = sum of all prey items found in only those 
predator stomachs that contained prey I. 
Although the summed variable can be composed of 
the number or the volume or weight of prey items, we 
used weight and lowest possible prey taxon. The spread 
and location of points on an Amundsen plot indicate 
Weakfish 
For this study, 349 individuals with an average size 
of 21 cm TL for young-of-the-year fish through adult 
fish, were examined. Of the 349 stomachs extracted, 
276 contained food (79%; Table 1) and prey represented 
71 taxa (Table 2). 
