Gillanders et al,: Aging methods for Seriola lalandi 



815 



Figure 2 



Transverse sections of sagittal otoliths of kingfish over a range of sizes viewed 

 with reflected light against a black background: (Al 337 mm FL: (B) 638 mm FL; 

 (C) 850 mm FL; and (D) 1002 mm FL fish. Scale bar is 1.5 mm. 



classification matrix for the best model (see above) 

 was used to estimate the most probable age of each 

 fish by determining the probability that a fish would 

 be from each true age class given its two assigned 

 ages. Each fish was assigned to the age class with 

 the highest probability (Richards et al., 1992). This 

 estimate of age was used for growth models. Only 

 the first five age classes (0-4) were used in the clas- 

 sification matrix because, although older fish oc- 

 curred, sample sizes were small and some extrapo- 

 lation for missing age classes would be necessary. 

 Because data from older fish are important in esti- 

 mating growth models, fish not used in the classifi- 

 cation matrix were assigned an age by randomly se- 

 lecting one of their two age readings. 



The data matrix comprising the two age readings 

 for each of the three aging structures was then ex- 



amined to determine possible effects of the different 

 aging structures and to determine a final age for each 

 fish. Classification matrices for each aging structure 

 were determined following a modification of the 

 methods of Richards et al. (1992; outlined above). 

 Results from running the models for each structure 

 separately showed that the model of best fit was ob- 

 tained by using a normal model with a = 0; there- 

 fore, in its simplest form, the model contained the 

 parameters o^, o^, and terms for the relative bias of 

 each method (rj and rg; r^ - O-r^-r.2). Parameters de- 

 termining the change in bias with age ( y) and non- 

 linearity of bias with age (rj) were also added alone 

 and indexed by aging structure. Different combina- 

 tions of the model parameters were therefore tested 

 to determine which model provided the best fit to 

 the data and to determine relative biases by method. 



