NOTE Wood: Using bone measurements to estimate the size of Pomatomus saltatnx 463 



( Observed - Predicted) 

 (Predicted) 



xlOO. 



To determine if any one bone or set of bones provided the 

 best predictor equation, comprehensive models involving 

 sets of bones were fitted in a stepwise linear algorithm 

 by using the Akaike information criterion (AIC) as the 

 criterion for model selection. Models were generated in 

 both a forwards and backwards manner in order to con- 

 firm that the same model was returned in all cases. 



Results 



Fork length (FL) and total length (TL) measurements 

 were taken from 58 bluefish ranging from 110 mm to 900 

 mm FL. The resulting regression equations correlating 

 skull bone measurements to FL (Fig. 2) were highly 

 significant (P=0.005 for the dentary correlation and 



P<0.001 for the rest of the models). The r 2 values for the 

 FL predictive equations ranged from 0.988 to 0.997, and 

 the mean percent predictive errors ranged from -0.03 

 to 1.19 (Table 1). Similarly, all of the resulting models 

 correlating the bone measurements to total length (Fig. 

 3) were highly significant (P<0.001, r' 2 values ranging 

 from 0.987 to 0.996, and mean percent predictive errors 

 ranging from -0.11 to 1.07 [Table 1]). 



Bones were ranked from best predictor to worst pre- 

 dictor for both the FL and TL models by using the 

 Akaike information criterion (AIC). In both cases the 

 premaxilla was ranked the best predictor bone, followed 

 by the maxilla, the opercle, the dentary, the cleithrum, 

 and finally dentary body length. The bone measure- 

 ments included in the stepwise multiple regression mod- 

 el for predicting fork length were PMXL, OPL, and DN 

 (Table 2). In the best predictor model for total length, 

 PMXL, OPL, DN and CL were included (Table 2). 



