76 



Fishery Bulletin 97(1), 1999 



turned linear trends. When we replaced TOBEC with 

 sqTOBEC in the regression model for yellow perch to 

 account for nonlinearity, the r'^ value increased slightly 

 (+0.0002) and the mean percentage error decreased 

 (Table 1). The new yellow perch equation became 



WC = 0.767 + 0.638 x WWT + 0.996 x sqTOBEC 



(7-2=0.998). 



Backwards stepwise multiple linear regression 

 analysis of the yellow perch data eliminated LnTL 

 from the regression model and indicated that four 

 other independent variables iWWT, sqTOBEC, TL, 

 and an offset constant) be included. The inclusion of 

 TL in the new regression model increased the /~ value 

 by only 0.0002, but the mean PE also increased to 

 2.56% in the new model from 2.23'7( in the model that 

 included only WWT and sqTOBEC. Backwards step- 

 wise multiple linear regression analysis of the ale- 

 wife data indicated that only WWT was a sufficient 

 predictor of water content in that data set. 



Percentage errors in predicted values for water 

 content were consistently larger than the mean PEs 

 for TOBEC values < 15 (<30 g yellow perch; <35 g 

 alewife; Tables 1 and 2). Previous work indicated that 

 the scanner is accurate for predicting lean body mass 

 in birds down to a body weight of 20 g (Castro et. al. 

 1990 ). We found that fish sizes that returned TOBEC 

 values below 10 yielded such poor results that we 

 did not include them in our analysis (<24 g yellow 

 perch; <16 g alewife). 



Discussion 



In fish, water content of the individual is strongly 

 correlated with the whole-body wet weight and, 

 hence, many relationships developed for the EM- 

 SCAN use wet weight as an independent variable 

 (Brown et al., 1993; Bai et al., 1994; Jaramillo et al., 

 1994). Although wet weight alone could be used to 

 predict many of these values, assessment of the 

 subtle differences in tissue constituents relevant to 

 energy-balance calculations could not be accom- 

 plished. Many studies on terrestrial vertebrates have 

 used TOBEC alone to predict tissue composition 

 (Presta et al., 1983; Keim et al., 1988; Walsberg, 1988; 

 Castro et al., 1990); however, all fish TOBEC stud- 

 ies to date have used wet weight and TOBEC as the 

 predictor variables in regression functions. In our 

 work, TOBEC values did not increase the predictive 

 ability of regression functions with wet weight as the 

 other independent variable. 



Three previous studies have indicated that TOBEC 

 can accurately predict water content in fish (Brown 



et al., 1993; Bai et al., 1994; Jaramillo et al., 1994). 

 Each study indicated that TOBEC and wet weight 

 values were correlated to body tissue constituents 

 and water content. The inclusion of TOBEC values 

 in regressions with wet weight as the other predic- 

 tor of whole-body body water content produced slight 

 increases in r'^ values and decreased the mean square 

 error We ran regression analyses with backwards 

 stepwise elimination and found that TOBEC was 

 eliminated from the sunshine bass data set (Brown 

 et al., 1993) when wet weight was untransformed 

 but was included when wet weight was log^, trans- 

 formed. The regressions for sunshine bass with 

 untransformed wet weight alone and combined with 

 TOBEC both produced lower PEs (2.6 and 2.5 respec- 

 tively) than did the Brown et al. (1993) equation with 

 log^-transformed wet weight (PE=3.9). TOBEC was sig- 

 nificant in regressions with wet weight for red drum 

 (Bai et al., 1994) and channel catfish (Jaramillo et al., 

 1994). In these data sets, regressions, including the 

 TOBEC variable, increased the PE fi-om 2.2 to 3.0 for 

 red drum (Bai et al., 1994) and decreased the PE fi-om 

 4.0 to 2.7 for channel catfish (Jaramillo et al., 1994) 

 versus the regressions with wet weight as the only in- 

 dependent variable. Inclusion of TOBEC values in re- 

 gressions with wet weight did not substantially or con- 

 sistently improve the predictions of water content 

 over simple linear regressions with wet weight alone. 

 When fat is burned and replaced by water in a fish, 

 changes in the total quantity of electrolytic salts 

 should be reflected in TOBEC values. Hence, TOBEC 

 values alone should be able to successfully predict 

 water content. TOBEC values alone can predict wa- 

 ter content values in terrestial vertebrates (/•~>0.80) 

 and fish (/•2=0.67 to 0.988). In fish however, TOBEC 

 predicted total wet weight equally well. If more than 

 just test animal size affects conductivity readings, 

 then evidence of changes in the total body content of 

 electrolytic salts should also be apparent when 

 TOBEC and water content values are divided by wet 

 weight. Wet-weight-standardized TOBEC and wa- 

 ter content values were not related for alewife, sun- 

 shine bass, or red drum (Fig. 2). The apparent trend 

 in the yellow perch data is counterintuitive (Fig. 2) 

 to the expected trend of increased conductance with 

 increased water content. The strength of predictive 

 equations for fish in the above four data sets may be 

 solely due to effects from fish size (e.g. serum and 

 cellular fluid volumes). The expected relationship for 

 these parameters is apparent in our plot of the test 

 data set from Jaramillo et al. ( 1994; Fig. 2). The nu- 

 trient content of diets fed to fish in those experiments 

 was carefully controlled within groups and varied 

 only between groups. Evidence from our study indi- 

 cates some promise for using TOBEC for fish in situ- 



