Scharf et a L Diet analysis of piscivorous fishes 
577 
est 0.01 gram (g) before external measurements were 
completed to the nearest 0.01 mm with digital cali- 
pers. To remove diagnostic bones, fish were placed 
in boiling water for 30-90 seconds, depending on fish 
size. Bones were then extracted from the soft tissue 
and measured to the nearest 0.025 mm with an ocu- 
lar micrometer (2.75x). 
Least squares regression equations (StataCorp., 
1995) were generated to predict original total length, 
fork length, and weight from measurements of body 
depth, eye diameter, caudal peduncle depth, pecto- 
ral-fin length, opercle length, cleithrum length, and 
dentary length. The left eye, left pectoral fin, and 
diagnostic bones from the left side of each fish were 
used consistently unless damaged. Body depth was 
measured as the maximum linear dorsoventral dis- 
tance with fins depressed. Eye diameter was mea- 
sured horizontally along the anteroposterior axis. 
Caudal peduncle depth was measured dorsoventrally. 
Pectoral-fin length was measured from the anterior 
most point of fin insertion to the tip of the longest 
fin ray. Opercle length was measured as the maxi- 
mum linear dorsoventral distance, usually from the 
dorsal most point to the tip of the primary ray (Fig. 1 ). 
Cleithrum length was measured from the tip of the 
dorsal process to the tip of the ventral process (Fig. 2). 
Dentary length was defined as the maximum linear 
anteroposterior distance from the symphyseal mar- 
gin located anteriorly between the left and right 
dentaries to the posterior tip of the dorsal or ventral 
process, whichever was longer (Fig. 3). A series of 
least squares regression equations to predict fish 
weight from total or fork length for each species was 
also generated. To further assess the strength of in- 
dividual bivariate relationships, mean percent pre- 
diction errors (Smith, 1980) were determined for each 
regression by averaging the percent prediction error 
calculated for each observation as 
[(Observed - Predicted) / Predicted ] x 100. 
Forward and backward stepwise linear regressions 
(StataCorp., 1995) were performed in an attempt to 
identify the best set of predictor variables. To ensure 
that all variables in the stepwise model were indi- 
vidually significant, the value of the ^-statistic used 
to determine variable inclusion was set at four 
(Draper and Smith, 1981). 
Opercles, cleithra, and dentaries were also care- 
fully examined to identify distinguishing character- 
istics that may be potentially useful for identifica- 
tion of each fish species. Several features of each bone 
were examined for differences among species, includ- 
ing the general shape of each bone; the numbers and 
orientation of ridges on the opercle and the curva- 
ture of opercle margins; the numbers, sizes, and 
shapes of cleithrum processes; and the presence or 
absence of teeth on the dentary, with attention given 
to overall tooth size, shape, and orientation, as well as 
the shapes and relative lengths of dentary processes. 
Resuits 
Predictive equations 
Regressions relating external morphological mea- 
surements to total and fork length were all highly 
significant (P<0.0001), with r 2 values ranging from 
0.63 to 0.99 and mean percent prediction errors rang- 
ing from 2.68 to 10.83 (Table 2). Regressions from 
measurements of eye diameter to predict original fish 
length were typically more variable than those from 
other external measurements. This was likely due 
to measurement error associated with damage of the 
soft adipose tissue surrounding the eye incurred dur- 
ing the freezing and thawing processes. Regressions 
relating diagnostic bone measurements to total and 
fork length were also highly significant (P<0.0001), 
with r 2 values ranging from 0.81 to 0.99 and mean 
percent prediction errors ranging from 1.26 to 8.48 
(Table 3). Compared with regressions from external 
morphological measurements, variation in diagnos- 
tic bone measurements typically explained more of 
the variation in original fish length (94% ofr 2 values 
>0.90) and bones were generally more precise in pre- 
dicting fish length (87% of mean %PEs < 5.00). 
Stepwise linear regressions indicated that cleithrum 
length was the most consistent predictor of original 
fish length and it was included in the best set of pre- 
dictor variables for 8 of 10 species (Table 4). Pecto- 
ral-fin length was included in the best set of predic- 
tors for 6 species, whereas dentary length was in- 
cluded for 5 species. The remaining independent vari- 
ables were included in the best set of predictors for 
either 3 or 4 species, respectively. 
Regressions relating external morphological mea- 
surements and diagnostic bone measurements to fish 
weight were also each highly significant (P<0.0001 ), 
with r 2 values ranging from 0.71 to 0.99 (Table 5). 
Mean percent prediction errors for regressions pre- 
dicting fish weight ranged from 5.97 to 39.17. These 
were typically higher compared with prediction er- 
rors for regressions predicting fish length, indicat- 
ing that estimates of original fish length were more 
precise than estimates of original prey weight. Simi- 
lar to length regressions, diagnostic bone measure- 
ments yielded higher r 2 values and lower mean per- 
cent prediction errors when regressed against fish 
weight compared with external morphological mea- 
