Caldarone et al : Nonlethai techniques for estimating responses of postsmolt Salmo solar to food availability 
263 
18% coefficient of variation (CV) compared to 3.7% for 
CP% and 1.9% for TWa%. In continually fasted fish only, 
changes in two BIA measures from day-3 fed values were 
observed: capacitance significantly decreased beginning 
on day 11 (Table 2) and impedance increased on day 23 
only (not shown). 
Between- treatment effects 
Results of the MANCOVA indicated that of the vari- 
ables tested (body composition [%-WW], BIA measures, 
growth rate, K), only growth rate (PcO.OOOl, df=13) and 
K (P=0.0009, df= 16) revealed significant differences due 
to feeding treatment. Results of Tukey’s HSD multiple 
range tests indicated that beginning on day 11 and con- 
tinuing until the end of the experiment, growth rates 
(negative) of the continually fasted fish were statisti- 
cally significantly slower than the fed treatment (Fig. 
1A). Eight days after refeeding (day 19), growth rates 
of the refed group were significantly faster than those 
of the continually fasted treatment. On day 19 the fed 
treatment also had faster growth rates than those of 
the refed group, whereas the relation was reversed on 
day 23. On day 23, K values were significantly smaller 
in the continually fasted fish than those of the fed fish 
(Fig. IB). There was a trend of higher mean phase angle 
values in the fed treatment than those in the fasted 
treatment (Fig. 1C), but the values between the feeding 
treatments were not statistically different owing to high 
variability within a day’s sample. 
Prediction models 
Body composition (g), /? and Xc par measured in the 
postsmolts encompassed a range of values (Table 3). 
The best predictor models for body composition (g) all 
contained fff and FL as independent variables, with 
some models also including BIA measures or K (Table 
4). In these models, K can be viewed as an interaction 
term between fW and FL (i.e., a size-related variable). 
The models for TWa and CP had high predictive capabili- 
ties (coefficient of determination (r 2 )>0. 98), whereas the 
model for TF was less so (r 2 range: 0.74-0.76). Adding 
any of the BIA measures to models containing only size 
or size-related independent variables (size-based-only 
models) increased the explanatory capabilities by <1.5%. 
The best predictor models for body composition 
(%WW) also contained size-related variables (WW or 
FL, and often K) (Table 4). The models for CP% ad- 
ditionally contained two or three BIA measures. Add- 
ing the BIA measures to a size-based-only CP% model 
increased the explanatory capabilities by <3.4%. 
The best predictor models for growth rate included 
size-related variables plus two BIA measures (Table 4). 
Adding the two BIA measures to size-based-only models 
increased the predictive capability of the growth equa- 
tions by 18-22%. 
Models with AAICc values <2 are considered to be 
equally probable to the “best fit” model (Burnham and 
Anderson, 2002). Based on the AAICc and r 2 values, 
(A) Instantaneous wet-weight-based growth rate (per d); 
(B) Fulton’s condition factor (K= 100»final wet weight/ 
fork length 3 ); and (C) phase angle (arctangent reactance/ 
resistance converted to degrees) of laboratory-reared 
Atlantic salmon (Salmo salar) postsmolts measured 
over 27 days. Values are mean (±standard deviation 
[ SD I ) for each sampling day. Fish were either fed ad 
libitum , fasted, or fasted until day 11 and then fed for the 
remainder of the experiment (refed). Within a sampling 
day, food treatments sharing a common superscript or 
without superscripts do not differ significantly (Tukey’s 
HSD multiple range tests). For fed and fasted fish n = 4 
for each sampling day. For refed fish n = 22 for day 11, 
n = 5 for days 15, 19, and 23, and n = 7 for day 27. 
