to 



S.0 



< 



g 



u -60.5 

 K 



-121.0 



1 1 ! 1 1 ! 1 1 1 I 1 I 



FEMALE BLUE MARLIN (N-86J k * 



1 1 I 





- 



1 1 1 1 1 1 1 1 1 1 I 1 



1 I.I 





"f.,<« J- . 



_i i i i i i_ 



'40 80 120 160 200 240 280 320 360 200 240 280 320 360 400 440 480 520 





38.0 



- 



FEMALE BLUE MARLIN (N = 85K 



v 



- 







- 



■ 



* 



' 





19.0 





X « * 





■ 



to 



-J 

 < 



a 



id 



Ul 







..'"■• 





" 



0.0 





., r : . % 





- 





-19.0 





* ■* " ■ f 





- 





-38.0 









- 

















'40 80 120 160 200 240 280 320 360 200 240 280 320 360 400 440 480 520 



STRIPED MARLIN (N«53) 



* 



■ 1! 



! " ,' j ' x 



9 25 41 57 73 89 105 121 137 142 162 182 202 222 242 262 282 302 



WEIGHT (KG) FORK LENGTH (CM) 



Figure 3. — Plot of residuals from the nonlinear model for female blue marlin with 86 and 85 samples and for 



striped marlin with 53 samples. 



the cases (Table 2), the plotting of the residuals 

 indicated that there was no reason to reject the as- 

 sumption of constant variance. Hence, the log-linear 

 model seemed to be appropriate. 



Nonlinear Model 



The nonlinear model (Equation 3) was fitted to the 

 data for the large blue marlin (five relationships) and 

 the striped marlin (Table 3) in order to compare the 

 fit of this model to that for the log-linear model. Since 

 the estimate ofc 2 is biased in nonlinear regression 



and therefore tests of significance cannot be made, 

 the distribution of the error terms was not tested. 

 The estimates of "R 2 " (a biased estimator in this 

 nonlinear case) indicated that the nonlinear model 

 does not in general account for as much of the varia- 

 tion in the data and is, therefore, not as good a 

 predictor as the log-linear model .When the residuals 

 from the nonlinear regression lines were plotted 

 against the dependent and independent variables, it 

 was found in every case that the amount of error was 

 small for small values of the variables and large for 

 large values of the variables. Hence, the assumption 



133 



